THEORETICAL AND EMPIRICAL PERSPECTIVES ON THE GOVERNANCE OF RELATIONS IN MARKETS AND ORGANIZATIONS Vincent Buskens, Werner Raub and Chris Snijders ABSTRACT This introductory chapter places the contributions in this volume in the larger picture of research on governance in markets and organizations and highlights the structure of the volume. We argue that including embeddedness arguments in a model for purposive behavior is a fruitful way to extend theoretical work on governance that allows for consistent derivation of hypotheses. We hope that this theoretical focus combined with “empirical pluralism” induces a cumulative body of evidence in the new economic sociology.
INTRODUCTION Sociologists distinguish between “social exchange” and “economic exchange” (cf. Blau, 1964). A standard ingredient of this distinction is that social exchange differs on important dimensions from spot exchanges on perfect neo-classical markets. This volume proceeds from the observation that economic exchange itself also often deviates from the perfect market model. In particular, a typical feature of much economic exchange is that it is associated with various forms of risks for The Governance of Relations in Markets and Organizations Research in the Sociology of Organizations, Volume 20, 1–18 Copyright © 2003 by Elsevier Science Ltd. All rights of reproduction in any form reserved ISSN: 0733-558X/PII: S0733558X02200014
1
2
VINCENT BUSKENS, WERNER RAUB AND CHRIS SNIJDERS
the partners involved, risks that are “assumed away” for the case of the perfect market. The contributions to this volume explore the implications of such risks by focusing on how governance is used in markets and organizations for mitigating the risks of exchange. Consider an R&D alliance as one example of transactions between two firms. A major feature of such an alliance is the interdependence between actors: the results of the alliance for one of the partners depend on the behavior of the other and vice versa. Two features of the structure of this interdependence are characteristic for all economic exchange. First, the firms do not have strictly opposite interests and, hence, can jointly profit from cooperation. For example, they both profit from the successful development of a technological innovation. Second, the interests of the two firms do not coincide entirely. Innovations in an R&D alliance are profitable for both firms, but the engineers of one of the partner firms may have carried a much larger share of the development efforts, a problem that is similar to the one experienced sometimes by academics co-authoring a paper. More technically, the firms are involved in a mixed-motive game and conflicts of interest are not excluded. Exchange relations are prone to a number of risks. Parametric risks might cause the exchange to fail independently of the behavior of one of the two partners. Strategic risks such as coordination, distribution, or cooperation problems can be the reason that partners in an exchange relations end up in suboptimal outcomes if they follow their individual short-term incentives (for more extensive treatment of the distinction of several kinds of risks see Harsanyi, 1977, pp. 124–135; Milgrom & Roberts, 1992; Williamson, 1985, pp. 57–58). Partners have an incentive to cope with the risks they face, because they benefit unilaterally and often also jointly from measures that mitigate parametric and strategic risks. In this volume, “governance” refers to the measures the actors involved in exchange use or implement in order to mitigate the risks associated with economic exchange. The governance of transactions and relations in markets and organizations is a topic of a broad range of approaches in economics (for an overview, see the chapter by Voss in this volume) such as transaction cost theory (Coase, 1937; Williamson, 1985, 1996), principal agent theory (Pratt & Zeckhauser, 1985), contract theory (Salani´e, 1997), and the theory of mechanism design (Myerson, 1987). These approaches offer not only theoretical insights but have meanwhile also produced a solid body of empirical research (for examples, see Masten, 1996). The governance of economic exchange is likewise a topic of sociological analysis. After all, the insight that exchange can be problematic due to unforeseen or unforeseeable contingencies and to strategic behavior of the partners can be traced back to Durkheim 1893, book I, ch. 7). Durkheim argued – albeit using other terminology – that solving these problems using complete contracts that cover all possible contingencies
Theoretical and Empirical Perspectives
3
is (too) costly so that other forms of governance are needed (see Batenburg et al., in this volume). While a point could therefore be made that sociology preceded economics with respect to identifying the governance of exchange as a core problem of the social sciences, it is also true that initiating a “critical mass” of systematic theory and empirical research on this problem needed quite some time in sociology. Various chapters in this volume (Batenburg et al.; Lindenberg; Voss) show in more detail how research on governance developed within sociology, culminating in what is today often addressed as the “new economic sociology” (Smelser & Swedberg, 1994). This volume presents theoretical and empirical sociological studies on the governance of exchange from different, sometimes competing and often complementary approaches that contribute to the new economic sociology. A common assumption in the theoretical analyses offered in this volume is that governance is considered a result of purposive behavior of interdependent actors. Actors involved in transactions and relations in markets as well as in organizations are confronted with risks and they can benefit from mitigating these risks through appropriate governance. Purposive behavior then implies that governance will depend on the feasible alternatives for governance, the costs (including transaction costs) and benefits associated with these alternatives, and on the actors’ information with respect to these alternatives and outcomes associated with implementing them. Moreover, the costs and benefits of governance will depend on the kind and size of risks involved in the transaction or relation as well as on the efficiency of alternative mechanisms of governance in mitigating the risks at hand. This volume is divided in three parts. Part I of the volume offers surveys of the use of purposive behavior models in research on governance as well as theoretical work that aims at extending these models. Voss remains relatively close to “traditional” rational choice models and surveys present limitations but also their potential for future extensions. Lindenberg as well as Podolny and Hsu advocate theoretical approaches that are explicitly contrasted against the traditional rational choice approach. The contributions in Part II and III all have an empirical component, and deal with governance in markets (Part II) and organizations (Part III). In the remainder of this introductory chapter, we sketch the major building blocks underlying the contributions in this volume, highlighting new perspectives and insights for research on governance that hopefully foster the development of a body of cumulative knowledge in the new economic sociology. We start by treating the distinction between markets and organizations. Thereafter, we discuss the life-cycle of economic exchange, which provides the logic for the ordering of chapters within Part II and Part III. We suppose that an exchange relation consists of four main phases: search and selection of the partner; negotiating and contracting; contract execution and performance; and (possibly) conflict regulation. Clearly, these phases often overlap. Part II as well as Part III cover all four phases of
4
VINCENT BUSKENS, WERNER RAUB AND CHRIS SNIJDERS
Fig. 1. Structure of This Volume.
exchange relations. Figure 1 gives an overview of the chapters and indicates their position within this structure. While we include chapters 2 to 4 from Part I in the figure, it should be clear that the more general approach of these chapters somewhat complicates their positioning. After having discussed the features that structure this volume, we briefly elaborate on the basic theoretical assumptions used rather consistently throughout all contributions: governance is a result of purposive behavior and embeddedness is an important determinant of this behavior.
GOVERNANCE IN MARKETS AND ORGANIZATIONS In terms of research fields, this volume addresses exchange in two different contexts, markets and organizations. Part II of the volume is on markets and focuses on three types of market relations. First, there are two studies on buyer-supplier relations (Batenburg et al.; Buskens et al.). These studies both use data from two closely related large-scale surveys on the purchase of IT-products (hardware and software) by small- and medium-sized enterprises. Here, the focus is on the buyer’s governance of the transaction in order to mitigate risks for the buyer, such as delayed delivery, delivery of inferior quality and insufficient service and maintenance. A second type of market relations addressed in this volume are business alliances. While Stuart’s study examines the governance of alliances in biotechnology, Gulati and Wang use data on joint ventures by large firms. The risks involved in relations of this type (see, e.g. Blumberg, 2001a, p. 826) include the abuse of confidential information and technological skills and the opportunistic exit from an alliance under changing contingencies. A third type of market relations on which this
Theoretical and Empirical Perspectives
5
volume focuses are relations between professionals and their clients. More precisely, using survey data from a sample of Chicago lawyers, Kim and Laumann address the market for legal services, where a major risk for clients follows from information asymmetry, namely, the imperfect information of clients about the quality of the legal service provided. While Part II of the volume comprises studies of the governance of interorganizational relations, Part III is on intraorganizational relations. Here, the contributions address two types of relations. The employer-employee relation is studied in the contributions by Neckerman and Fernandez with data from a large retail bank and by Petersen et al. with comprehensive Swedish wage data. Relations between employees are the topic of Wittek et al. who use data from an ethnographic study in a management team. Risks in employment relations are often related to investments in human capital of the employee (e.g. Becker, 1964). Investments of the employer in general human capital of the employee are threatened by exit of the employee. Conversely, investments of the employee in firm-specific human capital run the risk of opportunistic behavior of the employer who might turn out to be unwilling to provide appropriate rewards after these investments have been pledged. Other risks involve problems that result from asymmetric information such as imperfect information of the employer about skills of the employee, or imperfect monitoring of employees allowing for various forms of “shirking.” Conversely, the employer may face opportunities and incentives for opportunistic behavior such as misrepresenting working conditions or violating implicit agreements on training, career perspectives, or fringe benefits. In relations between employees, risks include freeriding on the efforts of colleagues (with negative effects for the colleagues as well as the employer) but also the violation of peer norms with respect to (reduced) effort (with negative effects for the peers but positive effects for the employer). The focus of this volume on the governance of relations in markets as well as in organizations and also the focus on different types of relations in markets and different types of relations in organizations is not accidental but results from a research strategy that we would like to label “empirical pluralism.” Arguably, the mainstream in sociology as a problem-driven discipline focuses on what might be called “theoretical pluralism.” Theoretical pluralism implies seeking progress by developing and testing alternative theories, often in the form of relatively small sets of testable hypotheses not systematically derived from an underlying model, for one and the same empirical phenomenon or for a small set of similar empirical phenomena. We submit that “empirical pluralism” is an alternative way of seeking progress, particularly in a more theory-driven discipline that aims at deriving testable hypotheses systematically from more general and abstract underlying models. Here, we seek progress by applying basically the same set of theoretical principles to a broad class of empirical phenomena in diverse fields of sociology.
6
VINCENT BUSKENS, WERNER RAUB AND CHRIS SNIJDERS
This volume fits into the strategy of empirical pluralism. The volume starts out with a general problem, namely, risks in different sorts of economic exchange. Subsequently, the theoretical analyses in Part I as well as the more applied contributions in Part II and Part III develop general concepts and ideas on the basis of which theory and testable hypotheses on the governance of risks can be developed. Applications are then selected in organization studies, labor market research, and the sociology of professions, trying to exploit structural similarities between, at first sight, rather different types of relations in markets and organizations such as R&D-alliances, buyer-supplier relations, the professional-client relation, the employer-employee relation, and relations between employees.1
GOVERNANCE IN THE LIFE-CYCLE OF ECONOMIC EXCHANGE The governance of economic exchange can be based on different mechanisms for mitigating risks and the literature on governance offers various approaches towards more systematic distinctions between different types of mechanisms. More specifically, one can distinguish different phases in the life course of an exchange relation: search and selection, negotiating and contracting, contract execution and performance, and (eventually) conflict regulation. In Part II on markets as well as in Part III on organizations, the sequence of contributions roughly follows the “life course perspective” on economic exchange. This means that the volume offers, first, analyses of governance through careful search and selection of the partner. While search and selection processes on the labor market are much researched (for an overview, see Devine & Kiefer, 1991), mitigating risks of exchange by searching for and selecting a competent, reliable, and trustworthy partner has been, surprisingly enough, a rather neglected topic in the literature on the governance of transactions and relations between firms (Blumberg, 2001b). It has only recently attracted more systematic work in the theoretical (e.g. Bala & Goyal, 2000) and empirical (e.g. Gulati & Gargiulo, 1999) literature on interorganizational network formation and dynamics. In this volume, Buskens et al. focus on search and selection in a market setting. They ask how buyer firms select their suppliers and conceptualize supplier selection as a means of mitigating the buyer’s risks associated with the subsequent transaction. The related contribution in Part III of the volume on relations in organizations is the study by Neckerman and Fernandez who also focus on selection processes: the hiring of employees by the employer in the employment relation. A second feature of ex ante governance is contracting between the partners. Various contributions focus on contractual features of governance and also on the
Theoretical and Empirical Perspectives
7
choice between mitigating risks through costly formal contractual means versus less costly but perhaps less effective informal governance based on trust in incomplete or partly implicit contracts (Azariadis, 1987; Hart, 1987; and see the contributions by Batenburg et al.; Lindenberg; Voss). In turn, this trust can be based on information about the partner’s trustworthiness, such as information on the partner’s behavior in previous transactions. Another basis for trust are opportunities for informal reciprocity in future transactions, that is, opportunities to use future transactions for rewarding trustworthy and sanctioning opportunistic behavior of the partner. Or, trust may be based on commitments of the partner not to behave opportunistically, where commitments may result not only from formal, contractual arrangements but may also be of a non-contractual, informal nature. In Part I, Lindenberg considers how trust and cooperation depend on cognitive processes and on “managing” the link between cognition and motivation. Voss discusses how we can understand relational contracting through a rational choice approach using repeated games. In Part II on markets, Batenburg et al. empirically analyze the choice between formal contracting and informal reciprocity in the governance of transactions in buyer-supplier relations. Stuart’s contribution likewise addresses features of contractual planning of market relations, by looking at financial investment decisions in strategic alliances in biotechnology. In Part III on the governance of relations in organizations, Petersen et al. focus on a core feature of contractual governance of the employment relation, the setting of wages. Ex post features of governance include contract execution and performance, the adaptation of contracts in case unforeseen contingencies emerge, and conflict regulation. We distinguish contributions that focus on performance from those that focus on conflict regulation. In Part II on markets, Gulati and Wang analyze performance features of joint ventures. They conceptualize the performance of a joint venture as reflected in stock market gain of the firms entering the joint venture. Gulati and Wang analyze both joint success in terms of the aggregated gains of the participating firms as well as the relative gains of the firms. Kim and Laumann analyze performance aspects of professional-client relations. The performance measure of Kim and Laumann are earnings of the professional, thereby focusing on individual instead of joint success. In Part III on organizations, performance features are the core topic of the contribution by Neckerman and Fernandez. They study how ex ante governance, namely, the selection process leading to hiring of employees, is related to performance. Better performance is then assumed to be reflected in lower turnover among new employees. Finally, two contributions in this volume focus on conflict regulation as a feature of ex post governance. Conflict regulation of market relations is the topic of the chapter by Lazega and Mounier. They offer an empirical case study of the Commercial Court of Paris, the Tribunal de Commerce de Paris. They analyze a specific mode of conflict regulation in
8
VINCENT BUSKENS, WERNER RAUB AND CHRIS SNIJDERS
inter-firm relations where judges of the Tribunal de Commerce who are involved in conflict regulation between firms are “consular” judges representing other firms and industries. Conflict regulation in organizations is the topic of Wittek et al. They study conflict regulation in a management team and focus on social escalation in conflict regulation. Finally, Podolny and Hsu do not focus on one specific phase in the exchange relation. They study effects of uncertainties that apply in different phases of exchange and in different types of markets and alternative “exchange domains.” Note that the different mechanisms of governance studied in this volume typically imply “private ordering” (Macaulay, 1986; Williamson, 1985) of risks. The focus is on endogenous, self-enforcing governance rather than on external, centralized policing and enforcement. While central authorities may be needed to create a context that makes certain mechanisms feasible – contract law being an obvious example – the mechanisms of governance are typically not imposed but are used voluntarily by the partners. Also, while analyses of governance in economics and, more specifically, in the transaction cost approach tend to focus on contractual features of ex ante governance, more sociological approaches accentuate non-contractual governance ex ante as well as ex post. This difference between economic and sociological approaches is reflected in this volume. Most contributions in this volume highlight non-contractual features of ex ante governance such as search and selection and informal reciprocity. The contributions also address a broad range of ex post governance, again focusing on non-contractual features.
GOVERNANCE AS A RESULT OF PURPOSIVE BEHAVIOR A common feature of the theoretical analyses offered in this volume is that governance is seen as a result of purposive behavior of interdependent actors. The contributions differ with respect to the way they interpret what “purpose behavior” means. Some contributions are based on rational choice assumptions as developed in (expected) utility theory and game theory. Voss surveys applications of rational choice assumptions in analyses of intra- and interorganizational governance. The chapters by Buskens et al. as well as Batenburg et al. explicitly use such assumptions as the theoretical core of empirical analyses of governance. Other contributions (Lindenberg; Podolny & Hsu) focus on explications of purposive behavior that are alternatives to or complements for standard rational choice assumptions as we know them from utility and game theory, highlighting applications for the analysis of governance. These approaches replace restrictive and empirically problematic assumptions of the rational choice model, such as
Theoretical and Empirical Perspectives
9
assumptions on available information as well as assumptions on information processing. Cognitive selection processes defining or framing the action situation are also addressed. An example from Lindenberg’s contribution are assumptions on cognitive selection processes that define the goal an actor pursues in a given situation so that the standard rational choice assumption of exogenously given preferences is abandoned. Still other ways of enriching the rational model embrace the possibility of myopia, the tendency to neglect or distort long-term consequences of behavior. An important feature of these approaches is, however, that adaptations of the purposive behavior model are not an aim as such. The focus remains on applying a possibly more complex model of purposive behavior for the explanation of governance features. Lindenberg mainly addresses the implications of purposive behavior models with more complex assumptions on cognitive processes for the motivation of employees, focusing on the governance of risks from opportunistic behavior in intraorganizational relations. Wittek et al. offer an empirical study using such assumptions to account for conflict regulation in relations between members of a management team. Podolny and Hsu take issue with information assumptions used in standard rational choice models. Standard rational choice models assume that actors can foresee possible outcomes of their actions and can assign probabilities to possible future events. Podolny and Hsu explore the implications of dropping the assumption that probabilities can be assigned to outcomes or that such probabilities are well-defined. They further assume that actors try to avoid such situations of “Knightian uncertainty” with undefined probabilities for outcomes. Kim and Laumann’s contribution on the lawyer-client relation can be seen as an empirical study in the spirit of Podolny and Hsu’s theoretical approach.
Governance in Embedded Settings Depending on the circumstances, different kinds of governance are more appropriate. But what are these circumstances? Economic approaches (e.g. transaction cost theory) tend to focus on transaction characteristics such as the nature and amount of specific investments associated with a transaction, the degree of asymmetric information and monitoring problems, and the volume of a transaction. The feasibility of governance also relies on actor characteristics, for example, on the size of a firm or on actor’s resources. Many studies in this volume take transaction and actor characteristics into account. Some do so by explicitly deriving hypotheses on the effects of such characteristics on governance (Batenburg et al.; Buskens et al.), others at least control for such effects in the empirical analyses (Gulati & Wang; Kim & Laumann; Lazega & Mounier; Neckerman & Fernandez; Petersen et al.; Stuart).
10
VINCENT BUSKENS, WERNER RAUB AND CHRIS SNIJDERS
Nevertheless, transaction and actor characteristics are not at the center of the theoretical focus of this volume. The focus is rather on conditions that seem to be crucial from a sociological perspective on governance. Such a perspective tends to explain governance as a result of the fact that (risks associated with) transactions are not isolated events but are embedded in a social context (see Granovetter, 1985). This embeddedness affects the size of risks as well as the feasibility and efficiency of different mechanisms of governance. We briefly discuss three major types of embeddedness that are prominent in the contributions in this volume, namely, embeddedness on the dyadic and on the network level (see Buskens & Raub, 2002, for a more detailed discussion) as well as institutional embeddedness (see Weesie & Raub, 1996).
Dyadic Embeddedness Several empirical studies in this volume address dyadic embeddedness and its effects on governance. Dyadic embeddedness refers to characteristics of the pair of actors as opposed to either characteristics of the goods or services they exchange or individual characteristics of the partners (Gulati & Wang refer to dyadic embeddedness as “relational embeddedness”). Assume that transactions between two actors are not one-shot but ongoing. Then, the actors share a common history of past transactions as well as a common future of upcoming transactions. A common history of transactions in the past can have a number of effects. On the one hand, an actor can learn from previous transactions about unknown and not directly observable characteristics of the partner. For instance, a history of mutually beneficial transactions may positively affect an actor’s belief that the partner is trustworthy and does not succumb easily to incentives for opportunistic behavior. Hence, positive past experiences may reduce the expected size of present risks from interdependence. Obviously, subtle processes of signaling and reputation building are involved. For example, even if a partner is not trustworthy, he or she may choose not to reveal this immediately but to imitate trustworthy behavior, waiting for a “golden opportunity” with rewards for opportunistic behavior that compensate for later sanctions. Learning effects of previous transactions between partners are analyzed in various studies in this volume. A core problem addressed by Buskens et al. is how previous experiences with a supplier affect the governance of buyer-supplier relations via partner selection of the buyer. Batenburg et al. analyze how previous experiences affect investments in contractual governance. Effects of prior transactions on funding of biotechnology alliances are a research problem in Stuart’s contribution, and Gulati and Wang investigate effects of prior transactions between partners on value creation and value appropriation in joint ventures.
Theoretical and Empirical Perspectives
11
While learning effects are an important offshoot of prior transactions, they are certainly not the only one. Another result of past transactions can be relationship specific investments (see, e.g. Williamson, 1985). These are investments that make the relation more profitable for both partners but have to be depreciated should the relation terminate prematurely. As an example consider an actor’s experiences of dealing with the partner’s idiosyncrasies. The value of these experiences is reduced drastically if the relation is terminated. Likewise, the value of smooth decision procedures fine-tuned to specific characteristics of their alliance depends on alliance stability. Unilateral relationship specific investments of an actor increase the actor’s dependency on the partner and, hence, the partner’s incentives for opportunistic behavior through a hold-up. On the other hand, mutual relationship specific investments will tend to reduce the incentives for opportunistic behavior and will stabilize the relation (see Williamson, 1985, ch. 8). Such effects of previous transactions on governance are analyzed for the case of buyer-supplier relations in the piece by Batenburg et al. Positive experiences with a partner or (mutual) investments in a relation through previous transactions can be assumed to increase trust, thus reducing, for instance, the need for costly contractual safeguards. Not all previous transactions have such beneficial effects for the partners and for the costs of current governance. Various studies in this volume discuss more problematic effects of previous transactions, even if these previous transactions have not produced negative experiences with the partner. For the case of joint ventures, Gulati and Wang address adverse performance effects from previous transactions through becoming overly dependent on each other, not responding optimally to changes in the environment, underemphasizing the exploration of new opportunities with other partners, etc. More generally, in his discussion of cognitive features of governance and focusing on relations in organizations, Lindenberg offers a theoretical discussion on how previous transactions are related to such cognitive features of ex ante as well as ex post governance. Lindenberg focuses on framing effects and processes of the definition of the situation, including problems such as “crowding out” intrinsic motivations through material incentives. Wittek et al. apply a related approach empirically in their study of conflict regulation in a management team. One of their problems is the erosion, over time (i.e. successive transactions), of cognitive barriers to conflict escalation. Dyadic embeddedness of a focal transaction refers not only to a common history of previous transactions between the partners but also to expected future transactions. A major effect of expected future transactions is that risks from interdependency can be mitigated through conditional cooperation (Axelrod, 1984; Taylor, 1976/1987). This is the control effect (“Tit-for-Tat”) of dyadic embeddedness. Opportunistic behavior can be mitigated through threats of future
12
VINCENT BUSKENS, WERNER RAUB AND CHRIS SNIJDERS
sanctions as well as promises of future rewards for cooperation. In general, if the long run costs of opportunism are sufficiently high to compensate short run gains, opportunism can indeed be mitigated, notably without any interventions of outside third parties. Expecting more future transactions with the partner not only allows for control effects but likewise tends to make investments in the governance of today’s risks more attractive if these investments are not completely transaction specific, that is, if they can be redeployed at least to some degree for future transactions (cf. Williamson, 1985, pp. 60–61 discussion of what he calls the “frequency dimension” of transactions). Effects of expected future transactions are discussed in various contributions to this volume. Voss offers a theoretical discussion on such effects as a core feature of rational choice models of the governance of relations within as well as between organizations. Batenburg et al. present specific hypotheses derived from a rational choice approach on how expected future transactions will affect contractual governance of buyer-supplier relations and offer empirical tests. Rational choice models of conditional cooperation presuppose that the actors value future outcomes sufficiently. An important argument in Lindenberg’s contribution in this volume is that actors are typically prone to myopia and systematic underestimation of future costs and benefits. He outlines implications of such myopia for the governance of intraorganizational relations.
Network Embeddedness The second dimension of social embeddedness is network embeddedness (Gulati & Wang use the label “structural embeddedness”). With the exception of the contributions by Batenburg et al. and Petersen et al., effects of network embeddedness on governance are extensively addressed in all of the empirical studies in this volume. Network embeddedness refers to the partners exchanging not only with one another but also with third parties. Consider the relation between two firms such that one is a supplier for the other. The buyer may have access to and communicate with other buyers of his supplier. Likewise, the buyer may know and have access to other potential suppliers. This implies that individual behavior in a relation with a given partner may affect relations with other partners. Hence, network embeddedness provides opportunities for learning and control effects similar and in addition to learning and control through dyadic embeddedness (see Buskens, 2002 for a more detailed discussion and analysis).2 First, consider learning through network embeddedness. Through his network with other buyers, an actor can collect information on the behavior of the supplier in transactions with other buyers, which can be used to update beliefs about
Theoretical and Empirical Perspectives
13
the characteristics of the supplier. Moreover, information from other buyers or observation of other buyers may provide information not only about the supplier but also about the feasibility and the expected costs and benefits of certain governance mechanisms: the buyer may try to improve on the governance of his relation with the supplier by learning from, imitating, and adopting successful, prevalent or legitimate management of other actors. Network embeddedness not only allows for learning but also for control because it offers new and additional options for sanctioning opportunism. First, network embeddedness allows for reputation effects through diffusion of information in the network (Raub & Weesie, 1990): in a network, bad behavior will spread for others to know about it and reputations will suffer.3 Another sanction that is feasible through network embeddedness is “exit.” If the supplier turns out to be incompetent or opportunistic and the buyer has access to other suppliers, ego may choose to terminate the relation and enter a new relation with an alternative supplier. Again, the long run costs for the supplier associated with exit of the buyer may induce the supplier to withstand the temptations of the short run incentives for opportunism in today’s transaction. Different types of effects of network embeddedness on governance are examined in the following contributions. Buskens et al. analyze learning effects of network embeddedness on governance through selection in buyer-supplier relations. Neckerman and Fernandez assume learning as well as control through network embeddedness and focus on performance effects of search and selection using networks in the case of employment relations. Stuart studies effects of network embeddedness on contractual governance. He focuses on learning through network embeddedness and shows how this affects the funding of biotech alliances. Performance effects of network embeddedness of market relations are the topic of the contributions by Gulati and Wang and by Kim and Laumann. Gulati and Wang investigate joint as well as individual performance of partners forming a joint venture. They assume learning as well as control through network embeddedness. Moreover, they investigate if network embeddedness can also have negative effects on performance because information one receives from partners might become redundant or otherwise less valuable and because high network embeddedness may hamper innovativeness. Kim and Laumann’s case is the lawyer-client relation. They focus on network embeddedness in the sense of a lawyer’s ties to prestigious other lawyers and ask whether such ties have positive signaling effects for clients whose uncertainty about the ability and trustworthiness of a lawyer can be reduced through such ties. Finally, two contributions are on network embeddedness and conflict regulation as an element of ex post governance. Lazega and Mounier look at conflict regulation in market relations and analyze network embeddedness of the consular judges in the Tribunal de Commerce of Paris. Wittek et al. consider
14
VINCENT BUSKENS, WERNER RAUB AND CHRIS SNIJDERS
relations within an organization and address network effects on conflict escalation in a management team.
Institutional Embeddedness An important additional kind of embeddedness is what we call “institutional embeddedness,” as studied by Lazega and Mounier and by Petersen et al. Institutions can be broadly conceived as constraints on human action that result from human action itself and structure the incentives for transactions. They constitute the formal or informal “rules of the game” in which actors are involved (North, 1990, ch. 1). Through the appropriate design of institutions (Myerson, 1987), risks associated with exchange can be mitigated. We distinguish two types of institutional arrangements that mitigate risks. First, there are constraints that directly affect individual incentives for opportunism by modifying the associated costs and benefits. Threatening “deviant behavior” with additional costs through external sanctions such as penalties or other punishment is a rather obvious example. Other institutional arrangements modify incentives only indirectly, by providing opportunities for actors to actively modify their own incentives. This includes ex ante features of governance such as mitigating risks by incurring voluntary commitments or “hostages” (Schelling, 1960; Williamson, 1985, chs. 7, 8). Voluntary commitments such as guarantees or warranties moderate risks associated with transactions because opportunistic behavior becomes less attractive, because they imply compensation for the partner who would suffer from opportunism, and because they may signal reliability and trustworthiness (see Raub & Weesie, 2000b). Also, institutional embeddedness may affect ex post governance. An example is conflict regulation through third parties. Given appropriate institutional embeddedness, conflict regulation through third parties may no longer be restricted to using litigation and the courts but might include the use of private mediation and arbitration or complex systems involving courts operated by members of a business community rather than professional judges (the study by Lazega & Mounier in this volume provides an intriguing case). One could easily take from this that institutional embeddedness is provided exclusively by the law, but it is not. For example, as Lazega and Mounier demonstrate in their chapter, branch organizations are an important ingredient of institutional embeddedness. Such organizations may provide standard contracts and make them easily available for trading partners. Or, such organizations may provide opportunities for private mediation and arbitration in the case of conflict. Institutional arrangements of this type are obviously close to the focus of this volume on governance in the sense of private ordering.
Theoretical and Empirical Perspectives
15
CONCLUSION Conceptualizing governance as a result of purposive behavior and accentuating the effects of embeddedness on governance allows to indicate how two broad intellectual streams within the social sciences and sociology may be fruitfully merged, namely, purposive action theory and what is sometimes labeled as the “new structuralism.” Rational choice theory provides parsimonious and powerful micro-level theories of purposive individual action. Limitations in the empirical adequacy of rational choice theory and social science applications of the theory (see, e.g. Green & Shapiro, 1994) have often been highlighted. Simplifying to some degree, such inadequacy can be attributed to two causes: either the behavioral decision theory is incorrect, or the assumed decision situations like the available alternatives and the potential consequences of behavior are incorrectly specified. Much recent research aims at developing a less simplified and more realistic model of individual decision making. This includes work in psychology and decision theory but – as this volume shows (see the contributions by Lindenberg; Podolny & Hsu) – it meanwhile also includes applied work in fields like the new economic sociology and organization studies. Sociologists tend to emphasize the way in which social conditions such as the embeddedness of transactions affect behavior. Thus, they emphasize the second possible cause for inadequacy of rational choice theory. With such an emphasis, the approach of Coleman (1990) and, notably, Granovetter (e.g. 1985, pp. 505–506), might become attractive. They maintain a simple behavioral theory of purposive behavior, possibly even rational choice or a tractable variant of rational choice, and focus attention on the impact of social conditions. The new structuralism in the study of social networks (see Burt, 1992; Nohria & Eccles, 1992; Wellman & Berkowitz, 1988 for representative volumes) emphasizes the importance of socialstructural conditions for the explanation of individual behavior and collective effects. It provides many useful theoretical notions and tools for describing, analyzing, and modeling the social context of behavior and collective organization and processes resulting from behavior. Many contributions in this volume contribute to the construction of an interface for both approaches that profits from their strengths while avoiding their weaknesses. Simple principles of purposive behavior like those developed in rational choice theory are often lacking in the new structuralism. On the other hand, the contributions systematically introduce aspects of social contexts like dyadic and network embeddedness into the analysis that differ from the context typically studied in micro-economics, the major field of application of the rational choice approach. Thus, an implicit reduction of purposive behavior models in general and, more specifically, of rational choice models to perfect market theory can be
16
VINCENT BUSKENS, WERNER RAUB AND CHRIS SNIJDERS
avoided. This volume hopefully shows that considerable progress can be made by replacing the assumption of perfect market coordination and by analyzing other forms of social embeddedness of behavior.
NOTES 1. It has been sometimes argued (Becker et al., 1977; Ben-Porath, 1980; Raub & Weesie, 2000a) that such an approach might even try to include analyses of still other social relations such as households or friendships as well as, e.g. relations between political parties such as in coalition government since these are likewise relations with partners facing structurally similar types of risks and incentives to mitigate such risks through governance. In a meanwhile classic paper on divorce, Becker et al. (1977, p. 1185) have nicely summarized the underlying rationale: “The approach to marital dissolution developed here should also prove useful in analyzing the dissolution of (implicit as well as explicit) contracts of indefinite duration between employees and employers, business partners, friends, etc. The case for a common theoretical approach to all social behavior would be greatly strengthened if the same theory is applicable to employee turnover and the termination of friendships, as well as marital dissolution.” 2. Note that Burt (e.g. 1992, 2000) discusses a related distinction between “information benefits” and “control benefits” associated with social capital and “structural holes.” Burt focuses on information and control benefits for individual performance in competitive situations. Our discussion shows that the governance of transactions and mitigating risks through governance can be beneficial for both partners in a transaction. Therefore, information and control may be beneficial for both partners rather than favoring one of them while putting the other at a disadvantage. 3. It has been observed (e.g. Lorenz, 1988; Raub & Weesie, 1990, pp. 648–649; Williamson, 1996, pp. 153–155) that control through network embeddedness and in particular control through reputation effects is often problematic. Such control effects presuppose reliable information diffusion as well as the willingness of third parties to implement sanctions, both of which are often problematic.
ACKNOWLEDGMENTS The order of authorship is alphabetical. Work on this chapter and on the volume as a whole was supported by a grant from the Netherlands Organization for Scientific Research (NWO; PGS 50–370). Buskens and Snijders acknowledge support of the Royal Netherlands Academy of Arts and Sciences (KNAW), while Raub acknowledges support of the Netherlands Institute for Advanced Study in the Humanities and Social Sciences (NIAS), Wassenaar, Netherlands. We would like to acknowledge that NIAS provided the environment for a number of authors (Lindenberg, Raub, Voss and Wittek) to work on chapters of this volume as part of a NIAS research group in the academic year 2001/2002. The context of the NIAS research
Theoretical and Empirical Perspectives
17
group was essential for making this volume possible. We thank Arnout van de Rijt for his excellent assistance in reviewing contributions to this volume and in other editorial work. We are indebted to former editor Sam Bacharach for offering the Research in the Sociology of Organizations series as a forum for this volume and thank Ed Lawler for acting as an intermediary and bringing us in touch with Sam. Mike Lounsbury, the new series editor, likewise supported this volume.
REFERENCES Axelrod, R. (1984). The evolution of cooperation. New York: Basic Books. Azariadis, C. (1987). Implicit contracts. In: J. Eatwell, M. Milgate, & P. Newman (Eds), The New Palgrave: Allocation, Information and Markets (pp. 132–140). London: Macmillan. Bala, V., & Goyal, S. (2000). A noncooperative model of network formation. Econometrica, 68, 1181– 1229. Becker, G. S. (1964). Human capital (2nd ed.). Chicago, IL: University of Chicago Press. Becker, G. S., Landes, E. M., & Michael, R. T. (1977). An economic analysis of marital instability. Journal of Political Economy, 85, 1141–1187. Ben-Porath, Y. (1980). The F-connection: Families, friends, and firms and the organization of exchange. Population and Development Review, 6, 1–30. Blau, P. M. (1964). Exchange and power in social life. New York: Wiley. Blumberg, B. F. (2001a). Cooperation contracts between embedded firms. Organization Studies, 22, 825–852. Blumberg, B. F. (2001b). Efficient partner search: Embedded firms seeking co-operative partners. Journal of Mathematical Sociology, 25, 329–354. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Burt, R. S. (2000). The network structure of social capital. Research in Organizational Behavior, 22, 345–423. Buskens, V. (2002). Social networks and trust. Boston, MA: Kluwer. Buskens, V., & Raub, W. (2002). Embedded trust: Control and learning. Advances in Group Processes, 19, 167–202. Coase, R. (1937). The nature of the firm. Economica, 4, 386–405. Coleman, J. S. (1990). Foundations of social theory. Cambridge, MA: Belknap Press. Devine, T. J., & Kiefer, N. M. (1991). Empirical labor economics. New York: Oxford University Press. Durkheim, E. (1893). De la division du travail social. 9. Aufl., Paris: PUF 1973. Granovetter, M. (1985). Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91, 481–510. Green, D. P., & Shapiro, I. (1994). Pathologies of rational choice. A critique of applications in political science. New Haven: Yale University Press. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104, 1439–1493. Harsanyi, J. C. (1977). Rational behavior and bargaining equilibrium in games and social situations. Cambridge: Cambridge University Press. Hart, O. (1987). Incomplete contracts. In: J. Eatwell, M. Milgate, & P. Newman (Eds), The New Palgrave: Allocation, Information and Markets (pp. 163–179). London: Macmillan.
18
VINCENT BUSKENS, WERNER RAUB AND CHRIS SNIJDERS
Lorenz, E. H. (1988). Neither friends nor strangers: Informal networks of subcontracting in french industry. In: D. Gambetta (Ed.), Trust: Making and Breaking Cooperative Relations (pp. 94–107). Oxford: Blackwell. Macaulay, S. (1986). Private government. In: L. Lipson & S. Wheeler (Eds), Law and the Social Sciences (pp. 445–518). New York: Russell Sage. Milgrom, P., & Roberts, J. (1992). Economics, organization and management. Englewood Cliffs, NJ: Prentice-Hall. Myerson, R. B. (1987). Mechanism design. In: J. Eatwell, M. Milgate, & P. Newman (Eds), The New Palgrave: Allocation, Information and Markets (pp. 191–206). London: Macmillan. Nohria, N., & Eccles, R. G. (Eds) (1992). Networks and organizations: Structure, form, and action. Boston, MA: Harvard Business School Press. North, D. C. (1990). Institutions, institutional change and economic performance. Cambridge: Cambridge University Press. Pratt, J. W., & Zeckhauser, R. J. (Eds) (1985). Principals and agents: The structure of business. Boston, MA: Harvard Business School Press. Raub, W., & Weesie, J. (1990). Reputation and efficiency in social interactions: An example of network effects. American Journal of Sociology, 96, 626–654. Raub, W., & Weesie, J. (2000a). The management of durable relations: Introduction and outline of a research program. In: J. Weesie & W. Raub (Eds), The Management of Durable Relations. Theoretical and Empirical Models for Households and Organizations (pp. 1–32). Amsterdam: Thela Thesis. Raub, W., & Weesie, J. (2000b). Cooperation via hostages. Analyse und Kritik, 22, 19–43. Salani´e, B. (1997). The economics of contracting. Cambridge, MA: MIT Press. Schelling, T. C. (1960). The strategy of conflict. London: Oxford University Press. Smelser, N. J., & Swedberg, R. (Eds) (1994). The handbook of economic sociology. New York: Russell Sage. Taylor, M. (1976/1987). The possibility of cooperation. Cambridge: Cambridge University Press (revised edn. of Anarchy and cooperation. London: Wiley). Weesie, J., & Raub, W. (1996). Private ordering: A comparative institutional analysis of hostage games. Journal of Mathematical Sociology, 21, 201–240. Wellman, B., & Berkowitz, S. D. (Eds) (1988). Social structures: A network approach. Cambridge: Cambridge University Press. Williamson, O. E. (1985). The economic institutions of capitalism. New York: Free Press. Williamson, O. E. (1996). The mechanisms of governance. New York: Oxford University Press.
THE RATIONAL CHOICE APPROACH TO AN ANALYSIS OF INTRA- AND INTERORGANIZATIONAL GOVERNANCE Thomas Voss ABSTRACT Rational choice theory has numerous implications for the analysis of organizational governance structures. This chapter reviews some of these applications. The main emphasis is on relational contracting. It will be argued that repeated games theory, that is, a variant of rational choice that deals with rational agents who repeatedly interact, can explain the outcomes of relational contracting. There is some controversy about the merits of rational choice explanations. Can they deal with inefficient structures and their (alleged) stability, with path dependence and mimetic processes? Many of these issues have been addressed by new sociological institutionalists. It is argued that rational choice analysis is in fact consistent with many of these observations. There is, in other words, some convergence between rational choice and institutionalist approaches.
INTRODUCTION Governance structures can be thought of as “institutional frameworks” which affect decisions within a transaction or a related set of transactions (see Williamson, The Governance of Relations in Markets and Organizations Research in the Sociology of Organizations, Volume 20, 21–46 © 2003 Published by Elsevier Science Ltd. ISSN: 0733-558X/PII: S0733558X02200026
21
22
THOMAS VOSS
1996, 11ff.). Governance structures can be related to interorganizational transactions, for instance a relation to a supplier. They can also refer to intraorganizational relationships of various kinds. Given the generality of the concept of governance or governance structure, the concept encompasses formal and informal institutional frameworks such as conventions, norms, and networks. The concept can therefore refer to institutions that guarantee the enforcement of hierarchical authority within a firm, the rules that govern individual careers within an organization, and the procedures that are used in inter-profit center transactions. The basic idea of rational choice theory in this field is as follows: Governance structures are chosen or evolve such that there are efficiency gains for the participants. Appropriate governance structures may in particular generate efficiency gains, if the focal transactions are subject to considerable opportunism problems. In this case, opportunism may be mitigated by choosing adequate governance modes. There are various, quite different, branches of rational choice thinking that relate to organizational governance structures. Not all of these can be reviewed here in detail. This chapter focuses on transaction cost thinking as the most prominent approach that has had an enormous impact on theoretical thinking and empirical research in many fields, including organizational sociology. One of the most important explananda of this approach is vertical integration. This is a subject that is most relevant to organization studies because it touches the most fundamental question that can be posed in the field: Why do organizations exist at all? It is also interesting to focus on the integration problem, because much criticism of the transaction cost approach refers to vertical integration. In particular, sociological new institutionalists and representatives of the new economic sociology suggested frequently that transaction cost analysis neglected the role of informal structures, of cultural rules, social norms, and social networks in analyses of the integration problem. It has been argued that economic transactions are “embedded” into social relations which provide important mechanisms to cope with the opportunism or (Hobbesian) order problem that is present in an economic transaction. This chapter argues that the rational choice approach is in fact capable to deal with informal governance structures. More recent work uses elementary tools of game theory to deal with interacting agents. If the modern theory of repeated games is employed to analyze transactions among economic agents, it will become fairly obvious that informal structures and outcomes of so-called relational contracting can be explained within a rational action framework. From this perspective, some convergent ideas that characterize the new institutionalism and rational choice analysis emerge. However, the strength of rational choice theory is the availability of a model building tool-box that has deductive and heuristic power to systematically generate empirical predictions.
The Rational Choice Approach
23
RATIONAL CHOICE AND ORGANIZATIONS: SOME PERSPECTIVES There are various contributions of rational choice to the analysis of organizational governance. This section aims to give a sketch of some core components of rational choice reasoning in general and its application to an analysis of governance structures. Most approaches focus on intraorganizational governance. Transaction cost theory, on the other hand, simultaneously deals with both, namely intra- and interorganizational governance. This will be subject of the next section. Rational choice ideas in the social sciences aim to explain social phenomena by assuming goal-directed behavior of the agents (see for a more general survey of this approach from a sociological perspective Voss & Abraham, 2000). The rational choice approach is based on the principle of methodological individualism. This requires that micro- and macro-level variables are to be explained by lawful regularities of individual action. In the context of organizational analysis, this may be consistent with the concept of corporate actors (in Coleman’s sense, see Coleman, 1990). That is, the actors may not be natural persons but collectivities such as states, organizations, groups, or agents who act on behalf of principals. One must, however, keep in mind that corporate actors, their objectives, and their internal functioning finally should – in principle – be analyzed as results of the actions of the participating natural persons, their interactions, and the collective decisions generated by their preferences. Rationality means that an agent behaves in accordance with certain rationality criteria. The set of possible rationality postulates that are used in rational action theory is comprehensive. Which rationality postulate is at work depends on the context. The simplest version of rational action theory assumes that an agent acts as if she were endowed with a consistent set of preferences over sure outcomes. She realizes an alternative (outcome) that ranks highest in the preference ordering. It is important that only a small subset of alternatives may be feasible. In other words, rationality means to choose under restrictions, constraints, or resources. In organizational analysis, constraints are influenced by institutional rules. There is a large body of work that uses this basic idea of “choice within constraints” in the field of organizational studies (see Ingram & Clay, 2000 for a recent review that also covers other institutions). For instance, in the property rights tradition within institutional economics (see Barzel, 1989; Eggertsson, 1990; Furubotn & Pejovich, 1974 for surveys of the literature) standard neoclassical economic theory (based on the principle of rational choice among sure alternatives within constraints) is applied to firms and other types of organizations. This work heavily focuses on the effects of various institutional restrictions on incentives of participants in organizations to choose options that may be consistent with desirable outcomes
24
THOMAS VOSS
(from the point of view of organization goals or the interests of a principal, etc.). To illustrate, an ideal typical bureaucrat (as described in Max Weber’s concept of legal-rational authority) will be faced with much different restrictions compared to a typical capitalist entrepreneur. The bureaucrat’s rewards comprise his salary, promotions to higher office, and the social approval that is attributed to his rank, among others. Therefore, the structural interests of the bureaucrat correspond to actions enhancing the chances of receiving higher salaries, that is, he is interested in the rise of the organization’s budget, because this may mean that more positions can be created, which in turn results in increasing promotion chances for the average bureaucrat (if personnel in higher office is recruited from inside), etc. An entrepreneur’s rewards, on the other hand, depend on the firm’s profit. This profit will be the most important determinant of her well-being, because it is the main source of income for an entrepreneur under the institutions of capitalism. Given this income, other (private) goals such as physical well-being and social approval can be realized. The capitalist will typically not be interested in the growth of the firm (or its budget, the number of employees, etc.) per se, but only when this relates positively to increases in profitability. In many contexts of analysis, the assumption of sure outcomes may not be appropriate. If there are risks or uncertainty with respect to the outcome that results if a particular action is chosen, more advanced rationality postulates have to be used. In this case, expected utility theory (as developed in modern decision theory) will be applied. The basic principle of action is often expressed as the idea that actors maximize expected utility (see Coleman, 1990, ch. 5 for a simple example of this reasoning in a sociological analysis of trust). It should be recognized that this is not a basic behavioral axiom of rational choice theory, but a logical consequence of more fundamental assumptions about preferences over risky or uncertain prospects. These consistency assumptions about preferences are analogous to the transitivity and completeness axioms employed in classic utility but must albeit be supplemented by additional axioms (in particular, an axiom of independence) to yield cardinal utility scales that have the expected utility property. Contemporary rational choice theory is dealing in most cases with social interactions among rational agents. This means, however, that a hypothesis of expected utility maximization will not be sufficient but only necessary to predict outcomes of interacting rational actors. Social or strategic interactions are given, in terms of rational action theory, if the outcome of actor A is determined not only by A’s decisions but also by the choices of other actors B, C, . . . Situations of this kind clearly correspond to what Max Weber has called rational social action or interaction. The branch of rational choice theory that deals with interacting agents is game theory. The most fundamental concept of (non-cooperative) game theory is the Nash equilibrium. A Nash equilibrium is defined as a profile of strategies such
The Rational Choice Approach
25
that no agent has an incentive to unilaterally deviate from that profile. One of the most important ideas that emerged from game theory is the insight that individual rationality need not be consistent with collective rationality (or efficiency) in the Pareto sense, that is, there are social dilemmas. The Prisoner’s Dilemma is a case in point. In a social dilemma, there may not exist a Nash equilibrium that is Pareto optimal. From such elementary reasoning it may at least be concluded that rational choice theory implies that we do not necessarily live in the best of all possible worlds. In other words, it cannot be expected under all circumstances that efficient institutions and governance structures will emerge and persist. Given a rational action approach, it is not necessary to assume that agents are only self-interested. However, the egoism assumption has dominated at least in economics for quite a long time. There are many substantive as well as methodological arguments that favor the assumption of actors who are primarily self-interested. The discussion of these arguments is beyond the scope of this essay. The self-interestedness assumption is powerful to generate predictions in many different contexts whereas other assumptions (of altruistic or other pro-social motivations) seem more context specific. Many classics of social theory, for example David Hume, have argued that altruism may only be in force within elementary interactions in primary groups, in particular the family. Another aspect is the danger of immunizing explanations by introducing motivational assumptions into the utility function that in an ad hoc manner just explain a particular behavioral outcome ex post, but are not useful to generate novel predictions in other situations. Despite the predominance of selfish motives in rational choice approaches, in recent times, non-standard preferences are increasingly introduced. Besides altruism these comprise fairness considerations and reciprocity (see Gintis, 2000, ch. 11 for an excellent introduction and overview). Note, however, that a nonstandard approach not necessarily is at odds with the rationality postulates. There are, for instance, some promising approaches to model non-standard preferences such that these are represented by von Neumann-Morgenstern utility functions, as used in standard decision and game theory (see, for example, Bolton & Ockenfels, 1999; Fehr & Schmidt, 1999). Another aspect of this body of work is that standard preferences are a limiting case of non-standard preferences. That is, depending on parameters of the model, the degree of fairness or other social orientations in the utility function may vary. This seems a clearly more fruitful idea than to assume simply that most actors are non-selfish (e.g. driven by certain moral norms) most of the time, as is apparently implied in some older work in sociology. Frank (1985) extensively makes use of the assumption that workers are motivated by their social standing, that is, by their local status with regard to material rewards such as wages in comparison to the status of their co-workers. This gives rise to numerous new predictions and may help to explain that wages are only loosely connected
26
THOMAS VOSS
with individual marginal productivity. This type of argument is closely related to the sociological concept of relative deprivation. On the other hand, it is derived from rational choice reasoning that takes into account non-standard preferences. Using this reasoning, some counter-intuitive hypotheses can be proposed that may be tested. Another aspect of motivation is that it is sometimes argued, at least tacitly, that agents will predominantly be governed by material interests. In much work in the field of principal agent theory (see Milgrom & Roberts, 1992 for a survey), it is assumed that agents are motivated by income (such as wages) and a taste for leisure or, conversely, a distaste for effort. Given these assumptions, many insights can be generated. For instance, one can show that market oriented incentives within work organizations may enhance productivity (see Petersen, 1992 for an empirical study). Lazear (2000) impressively shows that there is indeed a huge increase in productivity given the implementation of an incentive based wage system within an organization that produces automotive components (auto glass). This confirms classic principal agent theory. However, one can argue that this result may no longer hold for jobs with more complex tasks. Lazear’s example comprises relatively simple, routine tasks with a labor output that is measured (with regard to quality) easily by objective criteria (in this case, low quality work becomes immediately visible because glass breaks). Many jobs involve, on the other hand, complex tasks with an output that may not at all be metered by simple objective standards. It can be argued, in addition, that one must consider, in the case of more complex and responsible tasks requiring a substantial amount of professional education and training (such as jobs in academia) that people may not primarily be motivated by material (extrinsic) rewards but also by intrinsic rewards. These intrinsic motivations imply that an agent may be interested to do a good job because she values this per se. In this case, the employee will not necessarily reduce work effort to a zero level if monitoring is absent. Research in psychology suggests that there may be a crowding out effect among workers with a high degree of intrinsic motivation if external monitoring is increased or an incentive based reward system is implemented (see Frey, 1994 for reasoning along these lines): highly motivated workers may reduce their intrinsic motivation, given an increase in external incentives provided by the principal. Under certain conditions, the reduction of effort due to the crowding out of intrinsic motivation may be quantitatively more severe than the incentive effect on work effort due to the provision of external rewards. In these cases, incentive wages or other schemes of giving rewards on the basis of external motives, lead to a decreasing productivity. Note that incorporating intrinsic motivation components into the utility function is broadly consistent with rational choice theory. There is no assumption in the basic axioms of rational action theory that payoffs or outcomes or utility arguments are material; it is only sometimes convenient to use such an assumption.
The Rational Choice Approach
27
There are different research lines within the rational choice analysis of organizational governance structures. Principal agent models and other work in the area of human resource management and personnel economics (see Baron & Kreps, 1999; Lazear, 1995) deal with many topics and problems that are traditionally approached in organizational sociology, in particular in Max Weber’s theory of bureaucracy and related work. Given that modern organizations require that employees who receive an extrinsic compensation in terms of salary are working on behalf of a principal (for example, a capitalist entrepreneur), various questions arise: Under what conditions do these salaried employees have an interest to exert effort with respect to activities that favor the principal’s interests? Under which conditions are long-term employment relations and prospects of promotion into higher office providing incentives to cooperate? Can compensation schemes that make wages dependent on the employee’s positional rank (like in Weber’s bureaucracy) be compatible with incentives to work hard? Is it possible that even self-interested and rational employees are loyal and choose high levels of effort if there is only a mild level of direct monitoring? Answers to these questions are of course quite complex. However, they are in many cases roughly consistent with Weber’s ideas that bureaucratic organizations with life-long employment contracts, internal career ladders and salaries that rise over the working life may in fact have positive effects on employee effort and productivity. Economic approaches usually favor market incentives or performance pay in organizations. More recent research has demonstrated that such incentives in many instances may exhibit dysfunctional consequences. In addition to crowding out effects for intrinsic motivation, there may be serious problems if performance is measured by objective criteria (which can in principle be verified by third parties). This is so (see Gibbons, 1998; Kerr, 1975; Prendergast, 1999), because with regard to more complex tasks (e.g. teaching) it is difficult to define standards that unambiguously encourage the behavior that the principal desires. A teacher who is rewarded only in proportion to the academic gradings his students achieve will have incentives to discourage the entrance of less able students into his class. He will also emphasize academic subjects and neglect the teaching of social skills, humanitarian values etc. (which may be reasonable education goals). This phenomenon is, of course, an example of what has been called “goal displacement” in organizational sociology. Subjective performance standards may be more adequate for evaluating output in complex jobs. However, there may be dysfunctions too (see Gibbons, 1998, 1999): the application of such standards may be highly costly for the principal in terms of time and effort. In addition, there may be a considerable amount of lobbying, rent seeking, and other influence cost (Milgrom & Roberts, 1988). For example, research universities may evaluate the performance of individual researchers and teams based on a complex system of objective (such as quantity of refereed publications) and subjective criteria (chosen and implemented by the dean). Then it may be expected
28
THOMAS VOSS
that researchers will use considerable time to influence the dean’s decisions by providing him with information that may somehow be biased. In some European university systems, external evaluations by deans or by other referees use extensive self-evaluations as the input of information on which the final judgment will rest. It seems safe to say that the effort applied to such self-evaluation reports has the opportunity cost (in terms of the whole system) that a large number of research papers will not be published (although this is the target the incentive system was designed to encourage). This seems to be an outcome that is second best: Every agent would be better off if influence costs of this kind were saved because more scientific output would be produced and the ranking of researchers’ performance level would be unaffected (given the assumption that there is homogeneity within the population with respect to the ability to use influence activities). That is, the evaluation procedure yields a kind of Prisoner’s Dilemma among researcher and research teams. However, the productivity of the entire system may be even lower if there were other, less intensive types of evaluations or no evaluations at all. As illustrated, rational choice is capable to analyze various aspects of organizational governance structures by pointing out efficiency properties of intraorganizational incentive systems. In this way, some major problems of the field of organization studies are addressed that have been important at least since Max Weber’s seminal contributions on bureaucratic motivation structures. Transaction cost analysis is another rational choice perspective that has been most influential within recent decades. It deals with interorganizational, and, albeit less extensively, also with intraorganizational relations. As is well known, Coase (1937) submitted in his seminal paper on the “The Nature of the Firm” the argument that prohibitively large transaction costs, that is costs of searching for partners as well as costs of bargaining, specifying contracts, and enforcing contracts will shift transactions from the price system to the firm. This internalization of transactions is a major cause for the existence of hierarchical organization. Williamson (1975, 1985) introduced more specific indicators which are related to the amount of transaction costs. Subsequently, transaction cost analysis has become a major theoretical and empirical research program within various branches of the social sciences.
TRANSACTION COST ANALYSIS OF GOVERNANCE STRUCTURES Transaction cost analysis deals with many different explananda. As already emphasized in Coase’s classic work (1937), first the boundaries of organizations are
The Rational Choice Approach
29
explained. The focus is primarily on economic organizations (firms) in modern capitalism, but in principle other kinds of organizations can be dealt with. One should, however, notice that other economic institutions than capitalism or nonprofit organizations in capitalism are associated with different incentives for the participants with respect to economizing activities. Managers of non-profit firms or of public bureaucracies may not be subject to the same competitive pressure that motivate actors in others contexts to economize on transaction costs. Williamson (1985) provided typologies of transactions implying empirical dimensions or indicators for the amount of transactional problems or transaction costs that can be expected: transaction specific investments (asset specificity), uncertainty, frequency, and measurement problems. It is argued that vertical integration can be expected to occur, if there are large specific investments, uncertainty is high, the transaction is frequent, and measurement of quality difficult (see Williamson, 1985, ch. 3). Other seminal contributions in this field provide additional theoretical insights into the conditions that foster vertical integration or disintegration (Grossman & Hart, 1986; Hart, 1995; Klein et al., 1978). Numerous empirical studies have tested hypotheses related to transaction cost determinants of vertical integration. A survey of some older work is given in Williamson (1985, ch. 5). Joskow’s paper (1993, pp. 124–130) includes some further results. To illustrate this body of research, consider two well-known examples. They comprise “make” or “buy” decisions in the automobile industry. Monteverde and Teece (1982) have tested and corroborated the hypothesis – suggested by transaction cost arguments – that components which require a higher amount of application “engineering effort” and are thus more “complex” are more likely produced within the firm. Less complex, standardized products are more frequently bought from an independent supplier. Complexity in this interpretation means, among other aspects, a higher amount of specific investments, and thus a higher potential to “hold-up” the partner. To “make” such a component within the firm reduces supplier opportunism because the supply relation is brought under the regulation of the firm’s authority. Another famous, albeit historical case study comprises Klein’s (Klein, 1988; Klein et al., 1978, pp. 308–310) description of the integration of Fisher Body into General Motors (GM) in the 1920s. Fisher Body was a supplier of bodies for GM’s automobile production. Technological changes in the 1920s shifted – as Klein argues – the production largely from relatively flexible wooden parts to closed metal bodies. The latter require relatively large specific investments in the stamping machines and dies. This equipment cannot easily be used to produce bodies for other automobile corporations. The partners agreed on a long-term contract that aimed to protect them against hold-up and opportunism of each other. The contract contained detailed price specifications. After an increase in demand for automobiles, GM argued that Fisher should agree
30
THOMAS VOSS
on price reductions because larger scale production would decrease average production costs. The conflicts over these and related issues, according to Klein’s narrative, were finally solved by merger in 1924, namely, GM integrated Fisher into its hierarchy. The Fisher story is a paradigmatic example for transaction cost analysis and has found its way into many textbooks. However, as often with famous examples, it appears that it does not fit historical data (Casadesus-Masanell & Spulber, 2000; Coase, 2000; Freeland, 2000). It may be instructive to have a somewhat closer look at the reasons for this. First, technological change and therefore asset specificity seems less dramatic than Klein’s story suggests. Wooden parts have still been in use for an extended period of time. There are also some hints from historical records that the relationship between the partners has been relatively harmonious before the merger. The integration decision seems to be a consequence of a change in GM’s general management policy that favored the integration of suppliers, independent of transactional problems in specific cases. Finally, after the integration, opportunism and hold-up were not absent, but became visible as employee opportunism (see Freeland, 2000). Coase (2000, p. 30) refers to the Fisher Body case when he says: The view that I formed in 1932 (. . .) was that the asset specificity problem was normally best handled by a long-term contract rather than by vertical integration (. . .) This reexamination of the Fisher Body-General Motors case has not changed my mind. It is true that I have said nothing about the role of reputation in the events discussed. However, I have no doubt that concern for their reputation would also have deterred the Fisher brothers from engaging in the kind of practices described by Klein.
Thus, even Coase himself stresses the role of alternatives to vertical integration that may contribute to limit opportunism incentives in economic relationships. Long-term contracts and also informal mechanisms as “reputation” are indeed very important even in relationships with large asset specificity (see Joskow, 1993; Williamson, 1985, 1996). However, the transaction cost framework does not provide a comprehensive analysis of these informal structures. This is so because this approach lacks the micro-foundations that are appropriate to explain the emergence and stability of informal cooperative relationships among rational and interdependent agents. Contemporary game theory as applied in the theory of repeated games and other related areas is suitable to address this topic.
RELATIONAL CONTRACTS AND REPEATED GAMES: INFORMAL GOVERNANCE STRUCTURES Economic transactions are to a large extent governed by long-term contracts that may be formal but also contain informal agreements. These contracts and mutual
The Rational Choice Approach
31
agreements are referred to as relational contracts. A relational contract is given whenever a transaction is not discrete (Macneil, 1980). It is a characteristic of discrete transactions that buyer and seller come together in order to agree on a single transaction that involves no opportunism problems. There is no common history nor a common future. In other words, the shadows of the past and of the future are completely absent. Reputations with respect to third market participants are not important. Discrete contracts are enforced by physical means (that is, cheating is made impossible physically) or by third parties such as the law. It appears that spot transactions are typical examples of discrete transactions whereas social exchange relations are in almost all cases relational (see Voss, 1985). Relational contracts are common in business transactions in modern societies. Macaulay’s (1963) seminal “preliminary empirical study” has demonstrated that a large fraction of business deals are governed only to a minor degree by formal contracts that are, in principle, enforceable by the courts. If conflicts between the parties occur, legal sanctions are applied infrequently because law enforcement seems to be prohibitively costly. The mechanisms that help to create cooperation in business are, according to Macaulay, long-term relationships (that is, the shadow of the future), the value attributed to a good reputation, and a general moral code of business ethics and norms. Numerous empirical studies confirm the view that long-term relationships facilitate the spontaneous emergence of a private social order. Even if a Leviathan that enforces cooperation is virtually non-existent, as in many developing countries, there may be informal institutions that help to make some business transactions stable and reliable (see McMillan & Woodruff, 2000, pp. 2443–2445 for a survey of the literature). The basic theoretical ideas that may explain these findings can be elaborated by elementary game theoretic tools. In the following, the simplest, though illuminating, interaction structure that represents exchange will be referred to, the Trust Game (Dasgupta, 1988; Kreps, 1990, pp. 66–67; see also, for example, Buskens, 2002; Snijders, 1996 for more extensive theoretical and empirical analyses of the Trust Game). The Trust Game is a two person non-cooperative game that represents a one-sided Prisoner’s Dilemma. There is a trustor (actor 1) and a trustee (actor 2) who sequentially decide to place trust (actor 1) and to honor trust (actor 2) if actor 1 has given trust. The payoffs of the trustor are 0 (no trust given), G (gain from honored trust) and −L (loss from abused trust). It is assumed that −L < 0 < G. The trustee’s payoffs are 0 (no trust placed), R (honored trust) and T (trust abused). It is assumed that T > R > 0. Thus, if no trust is placed, both actors receive a payoff of 0 which is less than the payoff that results from honored trust (G for the trustor and R for the trustee). In other words, if there is trust, an efficient outcome will be realized. However, in this game, the second mover (the
32
THOMAS VOSS
trustee) has an advantage: Given that actor 1 has placed trust, actor 2 alone will determine the outcome of the game. Since T > R is assumed for the trustee’s payoffs, there is a temptation to abuse trust. Technically speaking, actor 2’s best response to actor 1’s placement of trust is to abuse trust. Rational actors will therefore not cooperate (in the sense of giving and honoring trust) in the single-shot Trust Game. This profile of strategies represents the unique Nash equilibrium. The result that no trust occurs is relevant for the analysis of discrete transactions which do not have a common past and a future. Relational contracts, however, per definitionem will not be single-shot interactions. Consider as a concrete example of a relational transaction the buyer-seller interaction with a buyer who frequently orders components that he needs for his production processes from the same seller. The seller will promise his partner certain services. Analogously, the buyer can promise to deal with the seller honestly. Then, the seller may have, in every exchange, an incentive to cheat on quality or he may not deliver in time, ask for an increase in price, etc. Analogously, the buyer may be considered a trustee from the point of view of his partner. The buyer may not pay for delivered services in time, cancel an order without announcement, etc. The critical feature of relational contracts is that they comprise repeated interactions between the same partners. Consider the repeated game with the Trust Game as its stage game. The usual infinitely (indefinitely) repeated games assumptions (see, for example, Raub & Voss, 1986 for an explication from a sociological perspective, and Gintis, 2000, ch. 6, for a more recent and comprehensive exposition) mean, among other aspects: The actors discount their future payoffs with a constant discount factor a (1 > a > 0). This discount factor represents the “shadow of the future” (in Axelrod’s (1984) technical sense of the term). It can be interpreted as a probability that the next repetition of the game will occur, under the condition that some stage of the game has been reached. The actors can act in the repeated game as if they reflected about trigger strategies. A trigger strategy uses information that has become public from the past behavior of the players to decide about the choices in the next period. A simple trigger strategy for the trustor is to give trust in the very first round. Thereafter, the trustor places trust only, if it has been placed in every round and if it has never been abused in the past. Otherwise, the trustor will never place trust. The trustee might reflect about the trigger strategy to honor trust if the trustor has placed trust in all previous periods. Otherwise, he will abuse. It can be shown that a Nash equilibrium of trigger strategies exists such that both partners will cooperate in every period of the game, provided the shadow of the future is large enough. The size of the shadow of the future is important because the incentive to break a relational contract becomes larger, if the sanctions (which are expected due to a refusal of any trust in the future) are low relative to the size of the rewards that a defection yields. More precisely, in a trigger strategy Nash equilibrium,
The Rational Choice Approach
33
the condition for a requires that a · (T − R)/(T − 0) for the payoff values of the trustee. The critical value of a that an equilibrium of trigger strategies requires, therefore depends on T − R, the temptation to abuse trust. The larger this payoff difference, the larger the temptation to renege on a contract, ceteris paribus. In other words, T − R gives the size of the potential gain that results if the trustee abuses trust in a single period. Thus, the possibility of mutual cooperation, in this simple model, depends on two interdependent parameters, namely, the (ratio of) payoff differences and the size of the shadow of the future. This line of reasoning can be enlarged and elaborated in various ways. First, consider a trust relation that is embedded into a social network. Then multilateral reputation effects are possible (see, among others, Buskens, 2002; Greif, 1994; Raub & Weesie, 1990): if partners A and B cooperate, then this may induce third partners C, D, . . . to cooperate with A or B likewise. Conversely, if the information about the abuse of trust by actor A spreads through the network, third partners will refuse to place trust in A. In many cases, the embeddedness of actors into networks of various types can make cooperation easier to achieve because the threshold that the shadow of the future must fulfill may be lower (see Raub & Weesie, 1990). Second, in many real world interactions, the repertoire of reactions on the partner’s abuse of trust (and thus strategies) may be quite comprehensive and complex. The basic interaction structure may, for example, contain certain retaliatory, direct sanctions such that the trustor may use more severe punishment threats than implied by the trigger strategy in the Trust Game. Given an enlarged Trust Game with the option to actively punish the partner (which is formally and strategically an entirely different game), social norms with sanctions may emerge (see, for example, Voss, 2001 for an overview with respect to enlarged Prisoner’s Dilemmas). Current work on the emergence of norms from a rational choice perspective (such as Bendor & Swistak, 2001; Binmore, 1994, 1998; Ellickson, 1989; Hechter & Opp, 2001; Young, 1998) may be highly relevant to cast light at this issue. Third, there is some evidence that supports the assumption that actors are not entirely self-interested but are endowed with non-standard motivation functions. For example, with respect to a public goods problem with an option to punish partners who free ride, this punishment option in fact led to higher cooperation levels, even in single-shot situations (Fehr & G¨achter, 2002). It is puzzling that the punishment option meant that a subject could bear a positive cost to apply a sanction to another subject that will never be met again as an interaction partner. Rational actors who care only about material outcomes would never use such punishments nor would the threats to use these sanctions be credible. In other words, there seems to be “altruistic punishment” (Fehr & G¨achter, 2002), because this punishment is individually costly (to the punisher) but profitable to the group because it leads to higher levels of cooperation. Such evidence suggests to introduce more complex
34
THOMAS VOSS
motivational assumptions into the utility function, because this may yield new empirical predictions and explain some important empirical evidence. It seems obvious, for example, that actors who are disposed (via certain emotions) to react on unfairness or free riding behavior by using sanctions that are materially costly not only to the defector but also to themselves, may achieve better outcomes than actors who are purely self-interested. This is because such emotions or non-standard preferences serve as commitment devices to make threats to punish defectors credible (Frank, 1988). One kind of emotional disposition which may have numerous powerful consequences in many settings is reciprocity (see Fehr & G¨achter, 2000). On the other hand, such emotions and non-standard preferences will probably be more important, the more an interaction is similar to face-to-face interactions among natural persons. Conversely, it may be the case that large, formalized corporate actors will more strictly comply to behavioral standards of self-interested preferences than natural persons. In large organizations, institutional rules and procedures may contain commitment devices. That is, an organization may use the rule that every instance of cheating will be sanctioned even if this will be more costly to the organization than the damage due to cheating (some large department stores and insurances companies announce publicly to use such rules). Another aspect that would explain this is reputation: It can be perfectly rational that an organization reacts in a provocable way on cheating, if this will be publicly known and helps to stabilize a reputation with regard to provocability. Fourth, there are in addition many other mechanisms which help to sustain cooperation. To mention just one that might be particularly important for interorganizational relations, actors can invest in scrutinizing their trading partners. Consider the case of incomplete information about the preferences of potential partners. Then, there may exist partners with preferences such that it would be rational to place trust, whereas vis-`a-vis other partners it would be better not to do so. Careful partner search and selection may in this case help to cope with the problem that the a priori probability estimate that a randomly chosen partner will be trustworthy may be too low. There is some empirical evidence that such search and selection indeed is very important in terms of management effort (Blumberg, 2001; Buskens et al., 2003; McMillan & Woodruff, 1999). How can game theoretic ideas in general and repeated games in particular be applied to relational contracts and vertical integration? Consider the main features of the repeated Trust Game. This game covers some of the important elements of relational contracts. The actors are involved in an on-going relationship with each other. Agreements about terms and conditions of the contract must be selfenforced to a certain degree. This is so for at least two reasons: First, the amount of “presentiation” (Macneil, 1980) is low in relational contracts. The contracts are thus incomplete, the set of possible future contingencies cannot be specified
The Rational Choice Approach
35
ex ante completely due to excessively large information costs. Second, the ex post monitoring and enforcement of the contract by a third party (the law, in particular) is very limited because under these circumstances of contract incompleteness only the parties themselves will be able to recognize and construct a common understanding about defections. In sum, the participants have to adapt, monitor, and enforce relational contracts themselves. What does this sketch of analysis imply for the choice of formal governance structures, such as the make or buy decision or the decision to incur management effort with regard to a transaction? Applying some recent, technically more sophisticated reasoning (Baker et al., 2001, 2002; see also Gibbons, 2001), the answer is: formal governance structures, in particular vertical integration, will affect the conditions of the relational contract between partners. To be more concrete, a relational contract depends on parameters that measure the incentives to defect (shadow of the future, payoff quotients). Given appropriate parameter values, certain Nash equilibria will exist that are compatible with an efficient outcome. Consider a supplier-buyer relationship with asset specificity. Supplier A offers a product or service to buyer B that may create an opportunity for a hold-up. The reason for this is that asset specificity leads to incentives to collect a rent (Klein et al., 1978). Supplier A could offer the product to a third party C who does not value the non-standardized product as much as B and therefore would only pay a lower price. The difference between the price that C would offer and the value of the product to B is the rent that could be appropriated. It may be possible that, by offering to invest specifically into the product along lines that might fit C’s interests better, A could strategically increase the price that C would be willing to pay. In doing this, A could improve its bargaining position in the relation with B. This would, however, reduce the total surplus in the A–B relationship, which is inefficient. Both parties would be better off if they refrain from doing so (see Gibbons, 2001). Efficient relational contracts between A and B contain an agreement such that A promises to accept a certain distribution of the surplus. On the other hand, consider an integrated relation between A and B. In this case, A will be a department that is subject to B’s hierarchical authority. This new department would, in principle, still have an interest to bring B into a hold-up position. The chances to do so are, however, very low. At least, they are low, given the bargaining techniques that can be used in the non-integrated situation. The basic reason is that the outside option C is no longer viable as a means to increase A’s bargaining position vis-`a-vis B. This is implied by the subordination under the hierarchical authority of B. Therefore, under conditions of vertical integration, the parameters of the relational contract may be different: The incentive to renege on the agreement may be lower for A under integration than under non-integration. This does, however, not mean that under integration no opportunism problems will
36
THOMAS VOSS
occur. The historical Fisher Body-General Motors merger apparently is a case in point (Freeland, 2000). Other examples are well-known from the literature on profit-center pricing (see, for example, Eccles & White, 1988). In other words, the decision to integrate or, more generally, to change the formal authority structure of the organization affects the conditions of the relational contract between the parties. Under certain circumstances, changing the formal authority structure will enable the realization of equilibrium outcomes of the relational contract that are more efficient. But this need not be the case (see Baker et al., 2002 for a more detailed elaboration of this). To reiterate, the integration decision does not imply that a relational contract ceases to exist. But the parameters and therefore the set of feasible Nash equilibria will change. One might pose the question whether this in fact explains the choice of formal governance structures. Basically, this approach of Baker et al. is suitable to draw out the consequences of certain institutional alternatives in terms of their efficiency properties. In order to explain organizational change and, more specifically, the choice of alternative formal governance structures, the implicit functionalism in the argument must be eliminated by specifying mechanisms that lead to more or less efficient governance structures. It appears that an explanation of governance structures (derivation of proposition 6. below) relies on the following assumptions and premises: (1) At time t, a relational contract with parameters P and a corresponding set of Nash equilibria E is given. (2) A particular equilibrium E1 (an element of set E) is realized. (3) A change in formal governance structures changes the parameters and corresponding Nash equilibria of the relational contract such that P and E came into existence. (4) A particular equilibrium E2 (an element of set E ) would be selected (realized) after the change of governance structures. E2 is Pareto superior to E1. (5) There are mechanisms such that the efficiency gains that are linked to a shift from E1 to E2 set in motion processes of institutional change that yield the intended outcome at t . (6) At time t a change of formal governance structures occurs. As always, sketches of explanations rely on more or less problematic assumptions. Two classes of assumptions are most problematic. One is related to premise 1, namely the equilibrium selection problem. In game theory in general, and in repeated games in particular, this problem has been discussed for many years. It connects to the fact that in most cases of non-cooperative games and most dramatically in repeated games, there are multiple Nash equilibria. Sociologically,
The Rational Choice Approach
37
equilibrium selection seems to be a problem to a lesser extent than analytically. This means that real societies, communities and organizations always have selection devices which are quite obvious. Greif (1994) coined the term of “cultural beliefs” to emphasize that equilibrium selection will be deeply dependent on history and therefore path-dependent and one of the most important aspects of what can be called “culture.” Miller (1992) argues that within organizations, leadership will mean to select equilibria out of an enormously extensive set of possible solutions. It is the task of leadership and management in organizations to consciously select particular equilibria by choosing, for example, pre-commitments to certain strategies. Then, the other actors who are involved in the game, will react by selecting their best responses. The importance of relational contracts is, of course, a basic insight of much research in the sociology of organizations. For instance, Gouldner’s classic on Patterns of Industrial Bureaucracy (1954) contains ethnographic and qualitative descriptions of the effects that a leadership turnover had on the employees-management relations within a business organization. Since the newly imposed leadership did not recognize the informal agreements between workers and management, which had been established in the previous periods, the new relational contract generated inefficient outcomes, that is, lower levels of cooperation and productivity than before. To cite just another well-known example, Blau’s (1964) notion of legitimate authority that can be understood as a rational reconstruction of Weber’s ideas on legitimacy, implies that formal authority relations are embedded into informal agreements. Authority means, on one side, that an appropriate informal contract is tacitly created and enforced such that both parties, the principal and its management representatives and the employees, gain. The enforcement of this form of vertical cooperation depends on the horizontal cooperation among the peer workers. There is a kind of community enforcement system (similar to those described in other contexts by more analytic means, see Greif, 1994; Kandori, 1992) such that workers who deviate from the contract are targets of informal sanctions from their peer workers, and not from the leadership. Another, more recent case in point are implicit contracts within industrial relations in the Japanese economy, as analyzed by Aoki (1988) who explicitly employs a repeated games approach.
SOME CONTROVERSIAL AND UNRESOLVED ISSUES Rational choice analyses of organizations have always been controversial. In particular, advocates of so-called sociological institutionalism in economic and organizational sociology argue that rational choice approaches are inadequate for methodological, theoretical, and empirical reasons. It is beyond the scope of this
38
THOMAS VOSS
essay to give a balanced and thorough discussion. However, some aspects of the criticism will be alluded to in the following. A major aspect of criticism of rational choice approaches is the neglect of relational contracts, informal relationships and social networks (Dore, 1983; Eccles, 1981; Granovetter, 1985 develop this point with regard to Williamson’s transaction cost analysis of vertical organization, to mention just a few representative contributions): Vertical integration is empirically less frequent than some versions of transaction cost reasoning suggest. There is a large variety of forms that are in-between pure hierarchy (vertical integration) and a spot market. This point is hardly disputed. However, it is obvious from our discussion that relational contracts can be matched by game theoretic ideas. Using game theory, it is clear that relational contracts allow for a considerable amount of cooperation. It is also the case that the interplay between formal and informal structures can be described endogenously by arguments that are in line with rational choice reasoning. Among the more recent contributions to economic sociology, Granovetter’s (1985, 1992) papers have become famous due to their rigorous appeal to social network thinking. According to Granovetter, the “embeddedness” of economic transactions into social networks can account for the solution of Hobbes’ problem of order within economic relations. Granovetter points out two types of embeddedness, namely relational and structural embeddedness. Structural embeddedness is given whenever “economic action and outcomes, like all social action and outcomes, are affected by actors’ dyadic (pair wise) relations and by the structure of the overall network of relations” (1992, p. 33). It is not difficult to see that the repeated games approach can be used to analyze effects of relational and of structural embeddedness on cooperation in relational contracts. Effects of the “shadow of the future” or of the “shadow of the past” (Buskens & Raub, 2002) on cooperation represent effects such that trust or cooperation may be more easily achieved due to dyadic or relational embeddedness. Similarly, multilateral reputation is an instance of “structural” embeddedness, since the conduct of the partners A and B within their relationship affects the possibilities of cooperation with third partners C, D, . . . No doubt, there is much need for additional theoretical analyses of structural embeddedness. But it seems safe to say, on the other hand, that in principle rational choice ideas can be merged with network thinking and that a unified theoretical program of rational action principles is useful to analyze effects of different forms of embeddedness on cooperation. Another issue is related to efficiency. It is often argued that organizational governance structures are selected with respect to efficiency criteria, for example, such that transaction costs are saved. Fligstein (1985, 1990; see also Freeland, 1996 for different arguments against the efficiency of the M-form) aims to demonstrate that a particular organizational innovation, namely the spread of the multidivisional
The Rational Choice Approach
39
form (M-form), is governed by a diffusion pattern that is analogous to other contagion processes. There are apparently imitation or social learning effects such that the adoption of the M-form by one major firm within an organizational field triggers the adoption by other firms. DiMaggio and Powell (1983, p. 152) argue, more generally, that organizations “tend to model themselves after similar organizations in their field that they perceive to be more legitimate or successful.” They also posit that such “mimetic” processes can account for institutional change and the ubiquity of certain structures to a higher degree than efficiency advantages. Similar ideas are relevant for organizational demography and population ecology (Carroll & Hannan, 2000; Hannan & Freeman, 1984) where much of organizational change with respect to core structures is attributed to the cultural legitimacy of organizational forms and the imitation of these forms in organizational birth processes. Another mechanism of change is the response of organizations to changes in the societal environment, such as the introduction of new legal rules. Dobbin and Sutton (1998) found that the creation of human resource management divisions within American public, for-profit and non-profit organizations in the 1970s can be attributed to organizational responses to the federal employment rights revolution. Only in later periods, these organizational structures tended to become rationalized by efficiency arguments by the participating actors within the field. Rational choice arguments would indeed point out that these institutions affect efficiency because they increase workers’ loyalty and cooperation in the employment relation. However, matters are not as simple as they sometimes appear. Rational choice theory is not a kind functionalist theory which pretends that every efficiency gain will in fact be realized. There are many obstacles to the realization of particular efficient outcomes. First, efficient outcomes may not be feasible at all for rational actors. To give an example, this may be so, because they tend to be consequences of a public good. It may depend on appropriate conditions whether the free riding problem can be resolved. In the literature on public goods production and collective action, many mechanisms that foster cooperation, and thus yield efficiency, are pointed out (see Hardin, 1982; Ostrom, 1989). One mechanism is the provision of selective incentives by third parties which link contributions to the collective good with private rewards. Then, the collective good may be a by-product of the external organization’s activities. I do not want to argue that structural changes in organizations necessarily must have collective good properties for the set of participating actors (investors, managers, employees, suppliers, customers). But insofar as they tend to be associated with those goods, the line of reasoning would apply. Safety regulations with respect to labor may in fact solve public goods production problems, because they improve the average laborer’s position. But there may be disincentives to adopt the respective institutions from the point of view of the organization and the workers (!) since they are clearly costly (Frank, 1985). Therefore,
40
THOMAS VOSS
selective incentives may be necessary to bring about the change. Second, even if appropriate incentives exist for supporting an efficient outcome, it may be difficult to implement. This is because of the equilibrium selection problem. Game theory, and repeated games theory in particular, is capable to describe the set of (Nash) equilibria that exist under specific conditions. In most cases, there are numerous equilibria in a game, and there may also exist multiple efficient equilibria. Game theory does not provide unequivocal hints on predicting the equilibrium that will be selected. But this may, on the other hand, be more puzzling a problem from the point of view of the foundations of game theory. It is less so from the standpoint of applied research. Equilibrium selection is a product of historical processes. Greif (1994) argues that “cultural beliefs” are involved in this process of selection. Notice that an equilibrium selection problem is a coordination game (and possibly also a bargaining game) for the population of actors. Therefore, there is a demand for coordination devices. A society’s culture is impregnated with such coordination rules. More precisely, cultural beliefs are rational beliefs “that capture individuals’ expectations with respect to actions that others will take in various contingencies” (Greif, 1994, p. 915). These beliefs refer to rational enforceable expectations that a particular equilibrium will be realized. Given convergent expectations of this kind, the actors will select a particular equilibrium. The emergence of convergent cultural beliefs is obviously a product of socialization or other cultural inheritance processes. These processes are presently not easily described by rational choice models. It is also fairly obvious that outcomes of processes with multiple equilibria are path dependent if they are affected by cultural beliefs. Consequently, given identical conditions with respect to opportunities, resources, and preferences, but different sets of cultural beliefs, huge divergences in institutional evolution may result (see Greif, 1994 for an instructive historical study). Thus, path dependencies and the impact of cultural elements on the evolution of governance structures are by no means surprising from a rational choice perspective. However, notice that only those outcomes will be realized that in fact correspond to an equilibrium (which is not necessarily efficient). In contrast, some institutionalists seem to argue that outcomes of processes of institutional change are independent of efficiency criteria and also independent of equilibria (in the Nash sense). Third, the line of reasoning discussed so far implies that there will be a considerable amount of institutional inertia. That is, even inefficient outcomes, being consequences of equilibria, can be stable, provided structural conditions remain unchanged. Fourth, rational choice arguments seemingly are not very strong when the task is to describe processes that yield specific outcomes. They are, for instance, not extremely good in predicting the exact pattern of timing of the adoption of certain innovations. However, as Strang and Macy (2001) point out, efficiency and optimality criteria are somewhat more plausible with respect to explaining the abandonment of “worthless strategies.” With regard to adoption decisions, there is large uncertainty
The Rational Choice Approach
41
about the potential merits of an item. Therefore, there is much need to draw on experiences of other (successful) actors and thus social learning and herding behavior abound. Once adopted, an organization makes its own experiences with a new structural element and can therefore judge whether it will probably pay off to retain it. Finally, with respect to mimetic processes, it may even be the case that inefficient governance structures emerge and persist due to the following mechanism. Consider a repeated game played by corporate actors with a stage game that may have similarities with the Trust Game or other social dilemma. Then, relational contracts that yield sufficiently high degrees of cooperation can be problematic, under particular conditions, because the shadow of the future is too low. Another factor that limits the chances of cooperation is incomplete information about the partner’s preferences. Relational contracts may under such circumstances be linked to signals (in the game theoretic sense) that credibly communicate an actor’s (in this case, a corporate actor) trustworthiness or, more generally, cooperativeness. One could also say that these signals are related to “legitimacy.” It could be a signal for an organization’s trustworthiness to adopt an apparently fashionable but economically worthless or costly innovation (introduce a new organization structure) if it were the case that only those actors would be capable to afford this innovation that are also inherently trustworthy. A similar line of reasoning is presented, albeit informally, in Posner’s (2000) theory of social norms. This theory could be adapted to the explanation of informal social norms within and between organizations. Presently, rational choice analysis seems to be somewhat stronger with respect to an analysis of equilibria and their efficiency consequences in various social situations. It is less strong, when it deals with the mechanisms of equilibrium selection. Some more recent work blends evolutionary games with boundedly rational agents (see Young, 1998 for an excellent example). This approach is promising because it allows for an endogenous analysis of equilibrium selection and path dependence. The objective of this evolutionary analysis of social dynamics is to illuminate the probability distribution of (stochastically stable) equilibria in the “long run” (that is, in the presence of stochastic shocks). Using this approach, the evolution of conventions can be explained endogenously (Young, 1996). There are not many empirical analyses for the time being. However, Young and Burke (2001) have used a variant of the evolutionary approach to deal with an empirical analysis of contracting.
CONCLUSION This chapter has given a review and discussion of rational choice explanations of organizational governance. There are many contributions to an analysis of
42
THOMAS VOSS
intraorganizational governance from a rational choice perspective that point out efficiency properties of certain institutional arrangements within organizations. The main focus of this chapter was on transaction cost arguments about vertical integration and relational contracting. These refer to both intra- and interorganizational kinds of relations. It was shown that game theory and the theory of repeated games in particular can cast light on the outcomes of relational contracting. This approach in principle also is capable to model various forms of network embeddedness. However, much more research is still needed that attempts to blend choice theoretic and network analytic concepts. Another aspect that can be dealt with is the link between formal and informal governance. Recent research aims to demonstrate that both intraas well as interorganizational relationships are governed by relational contracts. Formal governance structures may shift the parameters of a relational contract such that different equilibria will be feasible. This may yield efficiency gains. The strength of rational choice analysis is to demonstrate that certain formal or informal governance structures have particular consequences with regard to efficiency. This is possible if the underlying social situation is represented by an appropriate game theoretic model such that (Nash) equilibria of this game can be analyzed. However, in most instances, there are multiple equilibria. Which of these equilibria will be realized or, in other words, which governance mode will be favored, is explained less unequivocally. With regard to this equilibrium selection problem, coordination devices are important: cultural traditions, norms and conventions, managerial leadership, and so forth. These coordination devices can be described within the rational choice framework. Nevertheless, this description remains mainly exogenous. There is clearly some convergence and, maybe also some complementarity, between rational choice analyses and sociological institutionalism. Both approaches emphasize that the outcomes of equilibrium selection processes are dependent on historical, cultural, and other accidental factors. The description of these social mechanisms is one of the predominant aims of the new sociological institutionalism. However, sociological institutionalists would go too far if they claimed that organizational analysis could completely dispense with an analysis of equilibria and efficient equilibria. An analysis of equilibrium selection processes would be beside the point without specifying equilibria of certain social situations in the first place. Rational choice analyses of governance structures will have to address the social mechanisms of equilibrium selection more explicitly. This means that, first, additional empirical work will be needed about the social processes of equilibrium selection. It is largely a matter of empirical research to determine, for example, which cultural beliefs are at work in certain social situations. Greif’s work (1994) represents an excellent example of an historical analysis along these lines, but other
The Rational Choice Approach
43
types of empirical research are also desirable. Second, there is some promising theoretical work about evolutionary game theory that is based on micro-assumptions of boundedly rational or adaptive behavior. This area of research is highly relevant because an endogenous analysis of coordination mechanisms will be possible within such an approach. Moreover, the long run properties of social institutions (and, consequently, organizational governance modes) can be analyzed (see Young, 1998). The evolutionary analysis provides the tools for pointing out which equilibria will be favored by evolutionary forces.
ACKNOWLEDGMENTS This chapter was completed during a stay as a fellow at the Netherlands Institute for Advanced Study in the Humanities and Social Sciences (NIAS) in Wassenaar, The Netherlands. Detailed and helpful comments by the editors of this volume are gratefully acknowledged.
REFERENCES Aoki, M. (1988). Information, incentives and bargaining in the Japanese economy. Cambridge: Cambridge University Press. Axelrod, R. (1984). The evolution of cooperation. New York: Basic Books. Baker, G., Gibbons, R., & Murphy, K. J. (2001). Bringing the market inside the firm? American Economic Review Papers and Proceedings, 91, 212–218. Baker, G., Gibbons, R., & Murphy, K. J. (2002). Relational contracts and the theory of the firm. Quarterly Journal of Economics, 117, 39–83. Baron, J. N., & Kreps, D. M. (1999). Strategic human resources. New York: Wiley. Barzel, Y. (1989). Economic analysis of property rights. Cambridge: Cambridge University Press. Bendor, J., & Swistak, P. (2001). The evolution of norms. American Journal of Sociology, 106, 1493– 1545. Binmore, K. (1994). Game theory and the social contract. Volume 1: Playing fair. Cambridge, MA: MIT Press. Binmore, K. (1998). Game theory and the social contract. Volume 2: Just playing. Cambridge, MA: MIT Press. Blau, P. M. (1964). Exchange and power in social life. New York: Wiley. Blumberg, B. (2001). Efficient partner search: Embedded firms seeking co-operative partners. Journal of Mathematical Sociology, 25, 329–354. Bolton, G., & Ockenfels, A. (1999). ERC: A theory of equity, reciprocity, and competition. American Economic Review, 100, 166–193. Buskens, V. (2002). Social networks and trust. Boston: Kluwer. Buskens, V., Batenburg, R. S., & Weesie, J. (2003). Embedded partner selection in relations between firms. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 107–133). Oxford: Elsevier Science.
44
THOMAS VOSS
Buskens, V., & Raub, W. (2002). Embedded trust: Control and learning. Advances in Group Processes, 19, 167–202. Carroll, G. R., & Hannan, M. T. (2000). The demography of organizations and industry. Princeton, NJ: Princeton University Press. Casadesus-Masanell, R., & Spulber, D. F. (2000). The fable of Fisher Body. Journal of Law and Economics, 43, 67–104. Coase, R. H. (1937). The nature of the firm. Economica, 4, 386–405. Coase, R. H. (2000). The acquisition of Fisher Body by General Motors. Journal of Law and Economics, 43, 15–31. Coleman, J. S. (1990). Foundations of social theory. Cambridge, MA: Belknap Press of Harvard University Press. Dasgupta, P. (1988). Trust as a commodity. In: D. Gambetta (Ed.), Trust. Making and Breaking Cooperative Relations (pp. 49–72). Oxford: Blackwell. DiMaggio, P. J., & Powell, W. W. (1983). The iron cage revisited: Institutional isomorphism and collective rationality in organizational fields. American Sociological Review, 48, 147–160. Dobbin, F., & Sutton, J. R. (1998). The strength of a weak state: The rights revolution and the rise of human resource management divisions. American Journal of Sociology, 104, 441–476. Dore, R. (1983). Goodwill and the spirit of market capitalism. British Journal of Sociology, 34, 459– 482. Eccles, R. G. (1981). The quasifirm in the construction industry. Journal of Economic Behavior and Organization, 2, 335–357. Eccles, R. G., & White, H. C. (1988). Price and authority in inter-profit center transactions. American Journal of Sociology, 94(Supplement), S17–S51. Eggertsson, T. (1990). Economic behavior and institutions. Cambridge: Cambridge University Press. Ellickson, R. C. (1989). Order without law. Cambridge, MA: Harvard University Press. Fehr, E., & G¨achter, S. (2000). Fairness and retaliation: The economics of reciprocity. Journal of Economic Perspectives, 14, 159–181. Fehr, E., & G¨achter, S. (2002). Altruistic punishment in humans. Nature, 415, 137–140. Fehr, E., & Schmidt, K. (1999). A theory of fairness, competition, and cooperation. Quarterly Journal of Economics, 114, 817–868. Fligstein, N. (1985). The spread of the multidivisional form among large firms. American Sociological Review, 50, 377–391. Fligstein, N. (1990). The transformation of corporate control. Cambridge, MA: Harvard University Press. Frank, R. H. (1985). Choosing the right pond. New York: Oxford University Press. Frank, R. H. (1988). Passions within reason. New York: Norton. Freeland, R. F. (1996). The myth of the M-Form? Governance, consent, and organizational change. American Journal of Sociology, 102, 483–526. Freeland, R. F. (2000). Creating holdup through vertical integration: Fisher Body revisited. Journal of Law and Economics, 43, 33–66. Frey, B. S. (1994). How intrinsic motivation is crowded in and out. Rationality and Society, 6, 334–352. Furubotn, E., & Pejovich, S. (Eds) (1974). The economics of property rights. Cambridge, MA: Ballinger. Gibbons, R. (1998). Incentives in organizations. Journal of Economic Perspectives, 12, 115–132. Gibbons, R. (1999). Taking Coase seriously. Administrative Science Quarterly, 44, 145–157. Gibbons, R. (2001). Trust in social structures: Hobbes meets repeated games. In: K. S. Cook (Ed.), Trust in Society (pp. 332–353). New York: Russell Sage. Gintis, H. (2000). Game theory evolving. Princeton, NJ: Princeton University Press.
The Rational Choice Approach
45
Gouldner, A. (1954). Patterns of industrial bureaucracy. New York: Free Press. Granovetter, M. (1985). Economic action and social structure. American Journal of Sociology, 91, 481–510. Granovetter, M. (1992). The old and the new economic sociology. In: R. Friedland & A. F. Robertson (Eds), Beyond the Marketplace (pp. 89–112). New York: Free Press. Greif, A. (1994). Cultural beliefs and the organization of society. Journal of Political Economy, 102, 912–950. Grossman, S., & Hart, O. (1986). The costs and benefits of ownership: A theory of vertical and lateral integration. Journal of Political Economy, 94, 691–719. Hannan, M., & Freeman, J. (1984). Structural inertia and organizational change. American Sociological Review, 49, 149–164. Hardin, R. (1982). Collective action. Baltimore, MD: Johns Hopkins University Press. Hart, O. (1995). Firms, contracts, and financial structure. Cambridge: Cambridge University Press. Hechter, M., & Opp, K.-D. (Eds) (2001). Social norms. New York: Russell Sage. Ingram, P., & Clay, K. (2000). The choice-within-constraints new institutionalism and implications for sociology. Annual Review of Sociology, 26, 525–546. Joskow, P. (1993). Asset specificity and the structure of vertical relationships: Empirical evidence. In: O. E. Williamson & S. G. Winter (Eds), The Nature of the Firm (pp. 117–137). New York: Oxford University Press. Kandori, M. (1992). Social norms and community enforcement. Review of Economic Studies, 59, 63–80. Kerr, S. (1975). On the folly of rewarding A, while hoping for B. Academy of Management Journal, 18, 769–783. Klein, B. (1988). Vertical integration as organizational ownership: The Fisher-General Motors relationship revisited. Journal of Law, Economics, and Organization, 4, 199–213. Klein, B., Crawford, R. G., & Alchian, A. A. (1978). Vertical integration, appropriable rents, and the competitive contracting process. Journal of Law and Economics, 21, 297–326. Kreps, D. M. (1990). Game theory and economic modelling. New York: Oxford University Press. Lazear, E. P. (1995). Personnel economics. Cambridge, MA: MIT Press. Lazear, E. P. (2000). Performance pay and productivity. American Economic Review, 90, 1346–1361. Macaulay, S. (1963). Non-contractual relationships in business: A preliminary study. American Sociological Review, 28, 55–67. Macneil, I. R. (1980). The new social contract. New Haven, CT: Yale University Press. McMillan, J., & Woodruff, C. (1999). Dispute prevention without courts in Vietnam. Journal of Law, Economics, and Organization, 15, 329–354. McMillan, J., & Woodruff, C. (2000). Private order under dysfunctional public order. Michigan Law Review, 98, 2421–2458. Milgrom, P., & Roberts, J. (1988). An economic approach to influence activities in organizations. American Journal of Sociology, 94, Supplement, S154–S179. Milgrom, P., & Roberts, J. (1992). Economics, organization, and management. Englewood Cliff, NJ: Prentice-Hall. Miller, G. J. (1992). Managerial dilemmas. Cambridge: Cambridge University Press. Monteverde, K., & Teece, D. J. (1982). Supplier switching costs and vertical integration in the automobile industry. Bell Journal of Economics, 13, 206–213. Ostrom, E. (1989). Governing the commons. Cambridge: Cambridge University Press. Petersen, T. (1992). Individual, collective, and systems rationality in work groups: Dilemmas and market type solutions. American Journal of Sociology, 98, 469–510.
46
THOMAS VOSS
Posner, E. (2000). Law and social norms. Cambridge, MA: Harvard University Press. Prendergast, C. (1999). The provision of incentives in firms. Journal of Economic Literature, 37, 7–63. Raub, W., & Voss, T. (1986). Conditions for cooperation in problematic social situations. In: A. Diekmann & P. Mitter (Eds), Paradoxical Effects of Social Behavior: Essays in Honor of Anatol Rapoport (pp. 85–104). Heidelberg: Physica. Raub, W., & Weesie, J. (1990). Reputation and efficiency in social interactions: An example of network effects. American Journal of Sociology, 96, 626–654. Snijders, C. (1996). Trust and commitments. Amsterdam: Thesis Publishers. Strang, D., & Macy, M. W. (2001). In search of excellence: Fads, success stories, and adaptive emulation. American Journal of Sociology, 107, 147–182. Voss, T. (1985). Rationale Akteure und soziale Institutionen. Beitrag zu einer endogenen Theorie des sozialen Tauschs. M¨unchen: Oldenbourg. Voss, T. (2001). Game theoretical perspectives on the emergence of social norms. In: M. Hechter & K.-D. Opp (Eds), Social Norms (pp. 105–136). New York: Russell Sage. Voss, T., & Abraham, M. (2000). Rational choice theory in sociology: A survey. In: S. R. Quah & A. Sales (Eds), The International Handbook of Sociology (pp. 50–83). London: Sage. Williamson, O. E. (1975). Markets and hierarchies. New York: Free Press. Williamson, O. E. (1985). The economic institutions of capitalism. New York: Free Press. Williamson, O. E. (1996). The mechanisms of governance. New York: Oxford University Press. Young, H. P. (1996). The economics of convention. Journal of Economic Perspectives, 10, 105–122. Young, H. P. (1998). Individual strategy and social structure. Princeton, NJ: Princeton University Press. Young, H. P., & Burke, M. A. (2001). Competition and custom in economic contracts: A case study of Illinois agriculture. American Economic Review, 91, 559–573.
THE COGNITIVE SIDE OF GOVERNANCE Siegwart Lindenberg ABSTRACT In this chapter, governance in organizations is seen primarily as the governance of motivation of employees. It is argued that motivation is steered by cognitive frames (goal driven definitions of the situation), so that governance in organizations should focus mainly on the establishment and maintenance of frames. The chapter discusses how this may be done and how this cognitive approach to governance can be seen as an integration of transaction cost economics and the organizational behavior approach.
INTRODUCTION The term “governance” in organization studies is usually applied to contractual relations in and between organizations. Transaction cost economics has strongly influenced our understanding of governance in this sense. “Good cooperation” is seen as a transaction with low transaction costs. The relevant independent variables are factors that potentially affect transaction costs (such as asset specificity and uncertainty) and/or boost concern over contractual performance (such as volume of the transaction, lack of alternative partners, and lack of past experience with the particular partner), and/or influence sanctions against breach (such as recourse to courts, reputation, networks). The dependent variables are amount and kind
The Governance of Relations in Markets and Organizations Research in the Sociology of Organizations, Volume 20, 47–76 © 2003 Published by Elsevier Science Ltd. ISSN: 0733-558X/PII: S0733558X02200038
47
48
SIEGWART LINDENBERG
of measures taken by the parties to reduce transaction costs (such as extra care in finding a contractual partner, kind of contract, extra clauses in the contract, specific hostages, and, inside organizations, such “private orderings” as internal labor markets, etc.). This approach has been applied both to transactions between and within organizations. The inside of organizations is also studied by a different approach, one concerned with performance rather than with contracts: the organizational behavior approach. Good “cooperation” in this approach is understood in terms of satisfaction of the various parties in organizations. The major dependent variable is performance of employees. The independent variables are factors that influence performance (such as perceptual distortions, wrong attitudes, leadership style, role pressure, miscommunication, group conflict, lack of fairness). Of particular importance is the use of research on performance for devising means to increase performance (such as adapting reward and evaluation schemes; improving decision making, ways of communication, and leadership style; redesigning work and the division of labor). The advantage of the transaction costs approach is its theoretical unity and high tractability.1 A few assumptions go a long way. Its disadvantage is that many aspects presumably important inside organizations are not considered. The converse applies to the organizational behavior approach. It is very rich in the aspects that are being considered but it is theoretically eclectic and not well tractable. Given the complementarity of advantages and disadvantages, it is not surprising that there have been efforts to combine or link the approaches in some way. For example, Baron and Kreps wrote a book on “Strategic Human Resources” (1999). They observed that we initially expected that ‘the Economic way of thinking’ and the ‘Organizational Behavior way of thinking’ would be substitutes for one another . . . We’ve found instead that the disciplines are complementary; each helps to fill in holes left open by the other, thereby sharpening and clarifying what the other has to say (Baron & Kreps, 1999, p. vii).
Indeed, the book is remarkable in the mix of economic, sociological, and socialpsychological aspects. Economics offers the general framework, and sociology and social psychology fill in the questions on how expectations are formed and how goodwill is created. One problem with this approach is that the “model of man” of the general framework and of the theories for expectations and goodwill are not only different, they are often contrary in their assumptions. For example, the basic framework is transaction cost economics with the assumption that people are opportunistic and pursue their self-interest with guile whereas goodwill is supposedly a matter of intrinsic motivation. Thus, there is an important combination of aspects but they do not fit together in the microfoundations.2
The Cognitive Side of Governance
49
Rousseau attempted such an integration with her concept of “psychological contract” (Rousseau, 1995). This concept draws attention to the subjective side of contracts, to unwritten agreements and expectations and this is a strong point of her approach. However, it still offers no theory for the integration of the economic and the organizational behavior approaches. Clearly, there is room for a healthy difference in vision between economists and sociologists, leading to different caveats. Still an integration of the microfoundations is desirable, not just for the theoretical purist. Without it, ad hoc reasoning or simply stonewalling must help out anytime the two approaches would come to different conclusions. For example, from an economic point of view, one would recommend monetary incentives to motivate people. From a social psychological point of view, economic incentives are seen as decreasing intrinsic motivation. Instead, one would recommend job enrichment schemes and increased participation in decision making. The way Baron and Kreps deal with this problem is illustrative. They admit that economic and social-psychological theories could come to different conclusions, but that they will deal with these issues in a later chapter (1999, p. 195). In the later chapter (1999, p. 268) they reiterate that pay-for-performance schemes can dull intrinsic motivation. But instead of saying how this should be resolved, they refer to the earlier chapter in which intrinsic motivation was introduced. In short, the issue is not dealt with. In addition, the empirical evidence itself is “rich” enough to support vastly different and incompatible theories. For example, long ago, the human relations school pointed to an interaction between the model of man held by management and the productivity of the employees. If employees are considered as calculating, untrustworthy and shirking they will behave that way. If they are seen as honest, responsible and motivated, they will behave much better (see McGregor, 1960). This point has recently been reiterated by researchers who point to the importance of intrinsic motivation (for example, Osterloh & Frey, 2000). The model of man used in management practices is said to create a self-fulfilling prophecy. If you expect people to shirk, you will thwart their intrinsic motivation. Yet, whereas it may be true that people respond to the way they are expected to be, there is no denying that they also respond to opportunities to shirk, cut corners, and breach agreements even if they were thought to be non-calculating, honest, and motivated. For example, in the Netherlands, faculty has a certain amount of vacation days per year. If vacation days are not used this year, they can be saved for later years. There is an honors system. Faculty report themselves each year how many vacation days they have used. A recent study of the use of vacation days in the University of Groningen revealed that faculty who work more often at home do take up far fewer of their vacation days each year than faculty who work more often in their office. This is hardly a coincidence. The most likely interpretation of this finding
50
SIEGWART LINDENBERG
is that many days in which faculty members “work at home” are in fact used as unreported vacation days. Consistent with this interpretation is the fact that the faculty members who report using fewer of their vacation days are, on average, not more productive than the rest. For principal-agent theorists, this evidence would come as no surprise. Still, even if close supervision of the use of vacation days could be realized at low cost, would it increase or further decrease productivity? Seemingly, the empirical evidence allows both the human relations and the principal agent (and transaction cost) approach to claim that they have been right all along. If people do respond to how they are expected to be but also respond to opportunities they get, then it is important to look much more closely at how motivation can be managed. This is what governance is all about. Simply collating variables from different disciplines will tell us nothing about how they interact to form the patterns we observe. How then can motivation be managed? The remainder of this chapter presents an integrated approach to answer this question, based on the central place of cognitive processes for motivation.
JOINT PRODUCTION AND THE LINK BETWEEN MOTIVATION AND COGNITION The most important premise of our cognitive approach to the study of governance is that cooperation in groups and organizations is first and foremost contributing to joint production, rather than “exchange” (as seen from the point of view of transaction cost economics) or “performance” (as seen by the organizational behavior approach). There are two very important hypotheses of this approach. The first hypothesis is that governance of joint production is to a large degree the management of motivation for adaptive behavior.3 Setting the stage for joint production entails decisions on division of labor and rules of coordination. Of course this coordination is necessary. However, given the genuine incompleteness of contracts in complex organizations, tasks can only be incompletely specified. The view that coordination sets out the interlinking tasks and governance sees to it that people do what is expected fails when tasks cannot be well specified. In such situations with a great deal of tacit knowledge (see Osterloh & Frey, 2000) and need for adaptation,4 the most problematic ingredient that makes joint production work, is that each individual is motivated to use his/her intelligent effort adaptively to advance the joint production. This entails two kinds of behavior that are often merged in real life situations. First, joint production is adaptively advanced if members to the joint production act in such a way that, given the circumstances, their action instrumentally advances the common goal. Second, joint production is adaptively advanced if members behave in such a way that they contribute to the motivation of
The Cognitive Side of Governance
51
other members of the group to adaptively advance the joint production. Together, these two kinds of actions can be called solidarity (see Lindenberg, 1998). The motivation to behave in a solidary way, if present, tends to decay unless special care is taken to keep it up (see Andreoni, 1988). Thus, the governance of joint production is predicted to be to a large degree the management of motivation. It follows that a person, motivated to behave solidarily, will become part of the lateral governance of the organization by behaving in such a way that other members’ motivation for solidary behavior is maintained or increased. The second hypothesis is that managing motivation is inextricably linked to managing cognitive processes. A theory of governance must thus prominently deal with the cognitions and how they can be managed. What are the links between motivation and cognition? One link is the “definition of the situation.” Durkheim already claimed that ambiguous or contradictory information leads to lack of motivation (Durkheim, 1964, p. 96ff.) This was also experimentally shown by Raven and Rietsema (1957). If a situation is ambiguous, an individual will not be highly motivated to act in any particular direction (except, maybe, to improve the clarity of the situation). But then, in most cases there are many possible ways in which a situation can be defined. What the individual will be motivated to do depends heavily on which of these various possible definitions of the situation (or “frames”) is actually realized (see Lindenberg, 2001a). Thus, if we simply assume a structured situation (as we ordinarily would when using microeconomic theory or the theory of subjectively expected utility), we brush away the question by what processes the situation became so well-structured and why it is this rather than another definition of the situation that obtains. A second link between motivation and cognition, related to the previous one, is the competition between goals in an action situation. Goals are intimately linked to the mobilization of energy (motivation), but often an individual has conflicting goals. Take an individual who is about to conduct business with a supplier who also happens to be a member of the local business community. What is the main goal for the transaction? To make as much profit as possible out of this particular transaction? To consider the long-term profit potential of transactions with this particular partner? To deal collegially with the business partner? To retaliate for a nasty remark the other had made during yesterday’s Rotary Club meeting? The particular goal the individual focuses on will mobilize motivation for different kinds of action. The standard assumption that the person wants to maximize utility presupposes ordered preferences and thus simply brushes away the possibility that goals can be in conflict and ignores the fact that there is no superordinate goal and appropriate numeraire to make all subgoals compatible. It also ignores that the weight of various aspects depends on the goals that are pursued at the moment (see Gollwitzer & Moskowitz, 1996; Kruglanski, 1996). Thus, the various goals
52
SIEGWART LINDENBERG
cannot be simply pressed into a utility or objective function.5 From a cognitive point of view, conflicting goals imply a conflict for cognitive resources, that is, for access to attention, knowledge, memory, and processing capabilities. Certain goals will win this competition (more or less) and others will lose it (more or less), with important consequences for cognitive processing and motivation. The combination of the first and second points implies that the less clear the victory of one goal, the more likely that the goal conflict will lead to some form of half-hearted motivation to act in a particular way. A third link of motivation to cognition is that an individual’s cognitions are influenced by the behavior of others (see Fiske & Taylor, 1991). Cognitions of interacting individuals are interdependent and, as a consequence (due to point one and two above), so is their motivation to act in a certain way. For example, the definition of the situation (the frame) by one person will be affected by signals given off by the behavior of other persons in that situation, behavior that is in turn affected by cognitions the others have. Mismatches in frames can lead to great problems in the motivation to cooperate. For example, if A gives advice to B (defining the situation as one in which he can help) and if B interprets A’s advice as the attempt to increase his status by showing that he knows things better than B, then B’s reaction will be defensive or aggressive towards A, not just lowering A’s willingness to help but also the probability that A even perceives B’s need for help. Thus, the management of motivation entails the need to coordinate cognitions.6 Even though there is no room to present the theory of framing in any detail, a few essential points should be mentioned (see Lindenberg, 2001a for more detail). The basic mechanism of the cognitive approach to governance consists of a number of interrelated processes. People’s perception of a situation is selective. When we say that a person has a certain frame we mean that, compared to another frame, this person thinks of certain things more readily, is more sensitive to certain kinds of information, perceives certain alternatives more readily than others, and assigns different weights to certain aspects.7 To focus on certain aspects also means that other aspects are cognitively pushed into the background. This process is governed by goals in the sense that in the competition between goals in a situation one wins out and dominates the foreground as well as the major cognitive processes while the other goals are pushed into the background. From there, they can only influence the frame indirectly via the strength (or “salience”) of the frame. Compatible goals increase and incompatible goals decrease the strength of the frame. For example, when the frame is linked to the goal “to help a person in need” then the goal “to guard one’s resources” is in the background and the value of money is less than if the position of these goals were reversed. Nonetheless, money spent by helping a person in need is incompatible with the background goal, so that the more is being spent, the more the background goal decreases the
The Cognitive Side of Governance
53
salience of the frame. Lowering salience means that second and third best alternatives have an increasing probability of being chosen. The lower the salience of a frame, the more likely that the frames switch. In our example, the background goal will become the frame and the formerly dominant goals will be pushed into the background.
COGNITIVE ASPECTS THAT ARE PARTICULARLY RELEVANT FOR THE MANAGEMENT OF MOTIVATION One important consequence of the framing approach is that it draws our attention to phenomena and to aspects of governance that are vital to the analysis of the problems and the instruments of governance and yet have not been adequately considered by either transaction cost theory or organizational behavior approaches. First, there is the phenomenon of motivational decay, as mentioned above. This phenomenon makes it imperial that motivation in organizations and groups is managed. Second, there is myopia. Even though myopia has been studied extensively by psychologists and social scientists, it has not played an important role in the study of governance so far. Third, once the relation of goals to framing and the relation of framing to behavior are recognized, the analysis of goals becomes important for the study of governance. If we manage goals, we manage framing and thereby motivation and behavior.8 I will briefly deal with the last two points in somewhat more detail. Myopia. There is considerable empirical evidence for the point that human beings are forward-looking, that is, they anticipate consequences and generate expectations about the future, but they are not farsighted, which means that they cannot look far into the future and in fact often do not even look as far as they might, given the information they have (see, for example, Loewenstein & Elster, 1992; Loewenstein & Thaler, 1989). In fact people are generally quite myopic, the more so the more “hedonic” the situational stimuli. It is quite misleading to see myopia as either “just” a kind of (rational or hyperbolic) form of discounting or as a form of limited information processing ability, even though both phenomena do exist. In both ways of looking at myopia, the most important aspects of myopia for contracting are ignored.9 These aspects have to do with cognitive processes that allow short-term temptation against long-term self-interest. They are more easily captured in terms of framing effects (see Gattig, 2002). In the competition among goals in an action situation, goals related to short-term (hedonic) aspects (such as “to feel good/better”) are likely to win out because they are directly tied to emotions. For example, a manager who decides to give in to a golden opportunity with a third party (even though it means that he cannot honor his
54
SIEGWART LINDENBERG
contractual agreements with his current business partner) does not necessarily weigh the advantages of the golden opportunity against the breach of contract with his business partner. Rather, the aspects connected to the golden opportunity make a short-term goal very salient and thereby mobilize a frame in which the attention is focused on these short-term aspects. Knowledge and memory chunks related to these aspects become cognitively more accessible. The long-term aspects in the situation are pushed into the background from where they may still lower the salience of the (hedonic) frame but without appearing in any rational weighing of short-term versus long-term aspects in the situation. As a consequence, the alternatives are selected and ordered according to the short-term goal. From this, a number of interesting conclusions can be drawn. First, myopia is not primarily a matter of information; it is rather a matter of the situational selection of aspects. Second, in order to prevent the intrusion of myopia into long-term plans, a person must have ways to stabilize the current frame against a frame switch. We will discuss below how this is likely to happen and what social processes are involved. But it can already be gleaned from the exposition so far that for any kind of behavior, but especially for contractual behavior, frame stabilization is an important aspect of governance. The analysis of goals. In order to know what frames govern human behavior, we have to know the goals that human beings pursue. Of course human beings have many goals and it is almost impossible to come up with a limiting list. However, in all likelihood, the human goals are hierarchically ordered so that if we look high in the hierarchy, we can come up with a small number of goals that govern many lower-level goals. What kind of hierarchy and what kind of goals? Space does not permit a full exposition of a theory of goals (for a fuller exposition of this theory, see Lindenberg, 2001a). However, a number of important points can briefly be summarized. First, people are actively involved in shaping their lives. This implies that it is more useful to see them as producers than as consumers. Their productive efforts are directed towards the realization of certain substantive goals. Here the assumption is that they care especially for physical well-being – in terms of feeling comfortable and getting stimulation – and for social well-being – in terms of achieving status, getting confirmation from others on one’s opinions and actions, and giving and getting affection (see Ormel et al., 1999). It is also useful to make explicit what it is that people want to achieve with any particular substantive goal, what is their operational goal. Is it maximization of some overall utility; is it satisficing, or still something else? The assumption most compatible with human beings as producers and with much social psychological evidence is that they are mostly concerned with the improvement of their position (either their position right now or their position some time in the future). Technically speaking, the assumption of an “improvement” goal is for many situations identical to the
The Cognitive Side of Governance
55
assumption of maximization of utility. However, it is also different in important respects. It draws attention to the fact that people judge the value of something relative to some point of reference and that for many goods this point keeps shifting toward the status quo. Human beings actively search for ways to improve their position and this includes efforts to change their constraints over time rather than just choose from within the set of alternatives. Search behavior and resourcefulness are both important aspects of improvement. The combination of substantive and operational goals yields a few very highlevel and cognitively very powerful goals. First of all, there are the goals to improve physical and social well-being directly (i.e. right now). For example, a person may want to increase his comfort right now by getting rid of a feeling of hunger. Improvement in physical and social well-being is registered by emotions, that is, by a change in the way people feel. It is oriented towards the short-term and its power derives from its direct link to emotions. For easy reference, I call the frame that is related to this goal a hedonic frame. People can also be oriented towards the improvement of their resources. For example, they may invest in education in order to get better earnings in the future. The effect on physical and/or social well-being is mostly indirect because for most resources it is true that not their growth itself but their subsequent use and/or public display creates improvement in physical and social well-being.10 The frame that is related to this goal can be called gain frame for ease of communication. The orientation of a gain frame is towards the long(er) term, and, in comparison to a hedonic frame, the tie to emotions is much weaker (or absent). There is a third major goal that takes some extra explanation. It is the goal “to act appropriately.” Seemingly it has nothing to do with improvement. However, the relation to improvement is only veiled. There is a general regulatory interest that individuals conform to norms even when nobody is watching. For this very reason, there is a universal tendency to mistrust people who seem to conform to norms in order to get social approval or to avoid disapproval. If people are not in some sense “intrinsically” interested in conforming to the norms, they cannot be trusted to withstand the temptation to deviate when nobody is watching. For this very reason, it is likely that parents teach their children early on that they should behave appropriately for its own sake and that parents reward their children richly with social approval for showing a genuine sense of obligation and richly sanction children with disapproval for showing a purely instrumental attitude towards norm conformity (see Kochanska et al., 2000). In this way, the goal “to act appropriately” becomes a major source of social well-being and, at the same time, its relation to the improvement of social well-being is cognitively and socially veiled (i.e. it is in the cognitive background). The frame that this goal is related to can be called a normative frame.
56
SIEGWART LINDENBERG
In the literature, one finds other researchers deal empirically with the distinction between “orientations” that are quite supportive of the theory of “master” frames. However, rarely do we find all three in the same research. This probably is due to the fact that gain orientation in one research tradition is identified with a “longterm” or investment orientation (as against a short-term, hedonic or consumption orientation) whereas in another research tradition, gain orientation is contrasted with a normative orientation. Thus, we find research on “hedonic versus utilitarian orientation” (Dhar & Wertenbroch, 2000; Gattig, 2002; Mano & Oliver, 1993). We also find research on normative versus gain orientations. For one, research shows that action is actually guided by overriding goals and that one simply cannot presume that the overriding goal will always be “gain.” For example, de Dreu and Boles (1998) found that individuals with a “social value orientation” (comparable to a normative frame) chose significantly more frequently “appropriateness” as their guiding goal for negotiation whereas individuals with a competitive value orientation (comparable to a gain frame) chose significantly more often “effectiveness” as their guiding goal for negotiation. Carnevale and Lawler (1986) also found that the goal to act cooperatively and the goal to act competitively activate very different patterns of attitudes, expectations and behavioral repertoires. Ligthart (1995) directly tested normative and gain frames against each other and found that they predictably lead to different actions in bargaining. Clark has set up a longrunning research program around the distinction “communal versus exchange orientation” (see Clark, 1981; Williamson & Clark, 1989 for an overview of this program). She finds clear differences between these orientations that parallel the normative and gain frames. Research done by the organizational consulting firm “Motivaction” has lead to a differentiation between four types of orientation toward work that are linked to goals: “traditional” (conform to a general expectation that one ought to work), “postmodern” (make use of the opportunity to contribute to a collective good), “modern” (earn as much money as you can), and “hedonic” (take it easy).11 The first two types of orientation can be identified as being normative frames (with different sets of norms), the “modern” type is closely related to a gain frame and the fourth type is clearly related to a hedonic frame. There are good reasons to assume that people have stable traits that would make one of these orientations more likely than another. However, from a governance point of view, stable traits would restrict us to selection processes, whereas framing would allow in addition to select the direct governance of frames and thereby of behavior. In the literature we clearly find that even though there are stable differences between people, framing effects do override these differences to a considerable degree (see Gattig, 2002; Ligthart, 1995). This issue also came up in Clark’s research program. For example, she was very surprised to find that the Communal Orientation Scale and the Exchange Orientation Scale seem to be uncorrelated (Williamson &
The Cognitive Side of Governance
57
Clark, 1989, p. 97). From a framing theory point of view one would not expect that people are either one or the other but one would investigate the conditions under which people change frames and how these conditions can be influenced.
WHAT KINDS OF FRAMES SHOULD BE MAINTAINED FOR ADAPTIVE MOTIVATION IN JOINT PRODUCTION? How can motivation for adaptive behavior in joint production be managed? We now have all the tools in place to answer this question. The question boils down to what kind of frames would have to be present for this kind of motivation. Once we know this, the question then is: how precarious are these frames and how can they be stabilized? Before we start answering these questions, it is important to have a brief look at the relative strengths of the master frames. A priori, a frame can be assumed to be the stronger (i.e. the more salient) the more directly it is tied to emotions and to the improvement of the conditions of self. Thus, unless there are additional stabilizers, a hedonic frame (being high on emotions and direct concerning improvement) is likely to displace gain and normative frames; a gain frame (being low on emotions but direct concerning improvement) is likely to displace a normative frame (which is low on emotions and indirect concerning improvement). For this reason, governance would have to create special stabilizers in order to guard a gain or a normative frame against the intrusion of a hedonic frame. Because a normative frame is a priori the weakest, its stabilization against displacement by the other two frames is likely to necessitate the most governance effort. How vulnerable a normative frame can be is illustrated by the empirical finding by Frey and G¨otte (1999) that financial rewards reduce the work effort of volunteers. In terms of framing, this finding can be interpreted as a situation with a normative frame (for example with the goal to help or to contribute to the common good) that is being displaced by a gain frame (with the goal to earn money) the moment financial compensation is offered. In a gain frame, the hours worked are influenced by the amount of pay and because the pay for volunteer work is generally not very high, the average work effort is likely to decrease when financial incentives are introduced to boost volunteer work.
The Hedonic Frame: Not a Likely Candidate for Useful Governance Let us take a hedonic frame as the point of departure and the point towards which motivational decay will lead when governance fails. When individuals in the organization are in a hedonic frame, this means that they are oriented towards the
58
SIEGWART LINDENBERG
short-term; they seek out opportunities to improve their physical and social wellbeing directly. For example, they may enjoy their task which to them is stimulating, not very tiring, and a source of self-pride and positive feedback from supervisors and colleagues. This looks like a good things and indeed sometimes it is seen as ideal for “intrinsic motivation” (see Deci & Ryan, 1985). However, because enjoyment-based intrinsic task motivation is linked to a hedonic frame, its sustainability is likely to be weak (see Lindenberg, 2001b). The reason is that people in a hedonic frame are also very sensitive to moods and tend to focus on the enjoyable aspects of the task, avoiding the unenjoyable ones. This has negative cumulative consequences for the alignment of tasks and enjoyment, which work against a constant flow of self-pride and positive feedback from others. When that happens, employees will attempt to improve the way they feel by any means at their disposal in the work context. They seek stimulation (say, by surfing the internet, gossiping about others, walking about in the halls), and cater to their comfort (say, by frequently going to get a cup of coffee, worry about draughts from open windows, keep shifting work-related but tiring or anxiety provoking activities to a time in the near future). For their social well-being, they may seek out status-enhancing encounters with inferiors; exchange mutual pats on the back with colleagues for behavioral confirmation; and, for a bit of affection, help a friend down the hall to deal with his marriage problems, or complain about the management of “them” to people from whom they hope to get sympathy. From the point of view of a hedonic frame, work-related requirements are seen as so many opportunities or barriers to direct improvement of physical and social well-being. Close monitoring combined with frequent feedback in the form of social (dis)approval will increase the willingness to do what is expected. Yet, it does not help to create a motivation for intelligent behavior that is adaptive with regard to the (organizationally) intended results of joint production, let alone encouraging for the motivation of others to behave solidarily. Instead, it follows from the logic of framing that in a hedonic frame, the most salient aspects of joint production will be its externalities on the physical and social well-being of the employee him- or herself. If people make noise, or are in a bad mood, they ruin comfort and social well-being for others and are seen as inconsidered with regard to this interdependence. Conversely, people in a good mood and sociable spirit will be seen as laudable contributors to the overall atmosphere. A hedonic frame is thus not very useful for achieving adaptive behavior in joint production and yet it is likely to be the dominant frame when governance fails. This does not mean that enjoyment is unimportant. To the contrary, enjoyment can be a very important factor as a background goal that increases the salience of a gain or a normative frame. I will come back to this point. If adaptive behavior in joint production is deemed desirable (as it is likely to be), then governance has to be able to stabilize other frames than hedonic ones. Which other frames?
The Cognitive Side of Governance
59
Governing by Gain Frames: Possibilities and Limitations The obvious candidate for a useful frame is a gain frame. Employees in such a frame will act adaptively to the degree that they believe they can improve their resources. This makes them particularly sensitive to incentive instruments. It has been an important contribution of transaction cost economics to drive home the point that people in a gain frame will be strategically opportunistic, that is, they will use guile and anything they can get away with.12 They try to get as much out of a transaction as they can, put as little as possible into it, and no alternative is left unconsidered simply because it might be against the rules or unethical. Following the principal-agent theory, transaction cost economics has also found the solution to the problem of strategic opportunism: interest alignment via credible commitments. For organizations, this means that the way to steer the motivation of employees is to align their interests with the interests of the organization, say, by credible commitments to positive and negative conditional incentives, such as pay increases, bonuses, stock options, growing scope of influence, and career advancement opportunities, as well as negative sanctions, all conditional upon performance. However, gain frames cannot be the whole answer. There are too many problems left that potentially disturb the interest alignment and thereby the governance of motivation. One problem is the measurement of performance. To the degree that performance is difficult to measure, conditional rewards and punishments become problematic. Behavior then is likely to shift towards the best measurable components of the task, which is the opposite of adaptive behavior in joint production (Meyer, 1994). Without good measurement, rewards and punishments also seem subject to arbitrariness, which is likely to undermine the credibleness of the company’s commitments (M¨uhlau, 2000). When contracts are very incomplete, essential elements of tasks cannot be put into the contract and can thus not be made subject to contingent rewards. As a consequence, noncontingent boosts of employee motivation are necessary (see Osterloh & Frey, 2000). There is a number of other problems that are widely known but they cannot be adequately described (let alone analyzed) without considering cognitive effects on motivation. For example, arbitrariness does not just undermine the credibleness of the company’s commitments, it also creates feelings of unfairness that make people, even at their own peril, reduce intelligent effort favorable to the employer (see Folger, 1998; Folger et al., 1978; Kim & Mauborgne, 1998; Moorman, 1991). In transaction cost economics, people are by default assumed always to be in a gain frame. The very concept of “fairness” does not fit into this frame in which self-interest is pursued with guile and in which ethical standards have no effect on behavior. Emotional reactions to rule infractions and unfairness do not fit into
60
SIEGWART LINDENBERG
such a frame. They belong to a hedonic frame and presuppose effects of ethical standards on behavior. Even more serious is the problem (already mentioned in the beginning of this chapter) that people’s behavior is seemingly not unaffected by how they are seen by others. If the governance structure is based on the assumption that employees pursue self-interest with guile, intrinsic motivation is likely to be “crowded out” (see Osterloh & Frey, 2000). Thus, making people very sensitive to incentive instruments makes them less obligated to their tasks and to organizational goals and derive less enjoyment from them because they focus only on contingent rewards and punishments. This “crowding out” effect is very likely an effect of framing (see Lindenberg, 2001b). A third problem is the overwhelming evidence that people are often myopic (see above), and thus also are myopically opportunistic. This severely limits the workings of interest alignment. Myopic opportunism is most likely in a hedonic frame because the time horizon in such a frame is very short. However, a gain frame is no guarantee against myopic opportunism because, unless it is strongly stabilized, it can easily be displaced by a hedonic frame. The mechanism of displacement involves the motivational power of loss (see Kahneman et al., 1991). If people in a gain frame perceive a “golden” opportunity for a quick and sizable gain they are likely pursue it. If, however, pursuit of this opportunity is barred because it goes against a standing agreement or promise, then the very idea of having to miss out on such an opportunity creates a strong feeling of loss. Unless the gain frame is very strong, it will give way to a hedonic frame in which the goal “to feel better” is dominant. In this case, “to feel better” means avoidance of the feeling of loss and the most likely action that can achieve this is to take advantage of the golden opportunity. Thus, even a gain frame is no shield against myopic opportunism, that is, against pursuing short-term advantages even if there is a good objective chance that the aligned long-term interests are in jeopardy as a consequence. Newspapers frequently report cases in which against their long-term interest people are tempted to steal from the company; to gamble with company money; to relax the strictness of their control task, say in approving a loan for another member of the country club; to be bribed for relatively small amounts; to fail to pass on bad news in order to avoid embarrassment; to cut corners on set procedures; to pursue a golden opportunity that goes against long-term agreements, etc. Solutions to these problems have often been offered on an ad hoc basis, without any consideration of cognitive effects on motivation. For example, it has been observed that wages do not just have an effect as contingent rewards but, if they are high enough, also as gifts of the employer. They thereby create obligations in the employee to do his/her best (Akerlof, 1982; Baron & Kreps, 1999, p. 109ff.) The creation of obligations (by gifts or any other way) is not covered by the assumption
The Cognitive Side of Governance
61
of self-interest pursuit with guile. Such an effect cannot simply be added to the workings of a gain frame.
Governing by Normative Frames: Possibilities and Limitations As we have just seen, a gain frame does offer some advantages through the possibility of interest alignment but it leaves many problems unsolved and it creates some serious problems of its own. Would governance by normative frames be a better answer? Let us have a closer look. Maybe the most elaborate suggestion in this direction comes from the “new institutionalism” mainly developed by March and Olsen (1989, 1995). For reasons of space, I will take their approach as representative of kindred approaches in this direction. March and Olsen contrast what they call “the exchange perspective” of rational choice (roughly what we have just called a gain frame) with an institutional perspective in which human action [is] driven less by anticipation of its uncertain consequences and preferences for them than by a logic of appropriateness reflected in a structure of rules and conceptions of identities (March & Olsen, 1995, p. 28).
Even though they work out the theory of governance mainly for democracies as political systems, many of their conclusions hold equally well for organizations for which adaptive behavior in joint production is important. The authors see the “logic of appropriateness” as a cognitive process of the definition of the situation and of selective matching “the obligations of an identity to a situation” (March & Olsen, 1995, p. 38). This comes quite close to what has been called a normative frame in this chapter. Institutionalized rules (covering duties, rights, routines, roles, informal obligations, and standard operating procedures) define acts as appropriate or inappropriate. “Individuals come to define themselves in terms of their identities and to accept the rules of appropriate behavior associated with those identities. They seek the competencies required to fulfill their identities.” Governance in March and Olsen’s view thus has two main leverages: the creation and maintenance of identities and of the capabilities that are needed to act appropriately (such as rights, endowments, knowledge, skills). There are clear advantages to this view, compared to the “gain frame” view, in terms of instruments of governance. For one, the problem of strategic and myopic opportunism seems less serious than in the other two master frames. In a normative frame, improvement as a goal is “veiled” (see above) and thus does not play an important direct role in goal pursuit. To do the right thing (appropriateness) is a matter of matching rules and behavior, not of improvement. Thus opportunities to
62
SIEGWART LINDENBERG
improve one’s feelings or resources are not as easily perceived as in the other two master frames and that may reduce the susceptibility to opportunism. Second, we get rather concrete directions on what to do in order to create a normative frame. According to March and Olsen, clear identities have to be forged and individuals who assume these identities have to be endowed with the necessary resources to act appropriately. One of the major instruments to motivate people to use the logic of appropriateness is accountability. To make people accountable is to make them more careful in the definition of the situation and more sensitive to social pressure and to standards of appropriate behavior associated with their roles. Another major instrument is adaptiveness through experiential learning. For this to happen, accountability must be linked to accounts from which one can learn (for example, what went wrong and why) and which will be remembered, necessitating investing in records and intelligent retrieval. In addition, the information must be enriched by the experience from other organizations, necessitating network formation (March & Olsen, 1995, p. 233ff.). All in all, March and Olsen offer a variety of very concrete governance instruments that, at the face of it, are theoretically tied to a normative frame. Yet, there are also serious problems with this approach. The most important one is that March and Olsen are too short on the elaboration of the cognitive side of governance. Although the logic of appropriateness is described as if it were a frame, no other frames are elaborated, and therefore there is no serious treatment of the precariousness of this normative frame. To what degree is it threatened by what they call “self-interested exchange” or by a hedonic orientation? What is needed to stabilize a normative frame against the onslaught of a hedonic or a gain frame? Because March and Olsen have no conception of precariousness of a frame, they cannot systematically analyze what stabilizes frames. When one looks at governance from a cognitive point of view, governance is first and foremost the stabilization of certain frames. By admitting only a normative frame, March and Olsen cannot but believe that it is all a matter of rules. As a consequence, they even believe that self-interested behavior can best be captured as part of the logic of appropriateness: Self-interested calculation can be seen as simply [sic!] one of the many systems of rules that may be socially legitimized under certain circumstances (March & Olsen, 1995, p. 29).
Here we get dangerously close to role theory of the traditional sociological kind for which human beings are a tabula rasa, shaped by socialization and the given set of institutions.13 Governance is then just a matter of defining the right roles. March and Olsen clearly don’t want to go this route. They claim that “Any treatment of governance must include substantial elements of an exchange perspective to be plausible” (1995, p. 25). However, they simply
The Cognitive Side of Governance
63
drift into this tabula rasa scenario by not taking cognitive processes seriously enough. Another important problem of considering only a normative frame is that there is no analysis of the negative sides of behavior that is mainly driven by the wish to act appropriately. Assume that the frame is stable, that is, not subject to being pushed aside by a hedonic or a gain frame. Then, unless special care is taken, behavior in day-to-day interactions is likely to be strongly oriented toward the avoidance of social disapproval (since disapproval is the living proof that one did not act appropriately). This will lead to optimal results only if the norms of all relevant players are perfectly aligned with reaching optimal results and if all relevant players are in a stable normative frame. This however is unlikely. For one, it is very difficult to keep stable normative frames for all players. Second, the alignment of norms with results is not easily maintained because in daily interaction, others react mostly to infractions of the norms rather than to causal links to results. For example, a person who obviously free rides and lets the others do the work is not judged by how much he lowers the likelihood that the joint product will be finished in time. Rather, he will be disapproved of by the fact that he seemingly shifts the entire burden to the others (a breach of solidarity). Third, in daily interaction, the terms of accountability and experiential learning, two important governance instruments for March and Olsen, are then also likely to focus on conformity to norms rather than to instrumental action as such. For example, after the fact, a team leader may be criticized for having covered up wrongdoings of members of his team. Yet, in terms of the logic of appropriateness in daily interactions, the cover up is appropriate. Experiential learning will focus mainly on ways to avoid disapproval. What people find socially appropriate in daily interaction cannot be easily pushed around by changing the formal rules (see also Nee, 1998).
GOVERNANCE BY DYNAMIC INTERDEPENDENCE OF FRAMES Clearly, governance should focus on avoiding widespread hedonic frames in the organization. However, as we have seen, this does not mean that governance should focus on gain frames or on normative frames. What then? The answer takes a few steps to build. The first step is that we need both gain and normative frames to keep each other in check. The second step rests on the fact that important stabilizers of frames are salience enhancing background goals. We thus need the right background goals to stabilize frames. The third step is based on the realization that frames are mostly stabilized in interaction with others and thus governance
64
SIEGWART LINDENBERG
must be centrally concerned with creating the right conditions for frame stabilizing social interaction. Lets look at each step in more detail. In the rough distinctions between major frames, we recognized only three frames, one of which we dismissed as being harmful for adaptive joint production. Neither of the other two would do by itself. But maybe there are possibilities to combine the advantages and reduce the disadvantages of both by combining them in a dynamic way. Among the most important advantages of a gain frame are (a) that it is clearly linked to an improvement goal and to individual initiative; (b) that behavior emanating from it is sensitive to (material and status) incentives; and (c) that, by proper interest alignment, its orientation can be linked to organizational results. Among the most important disadvantages of a gain frame are (a) that it is wide open to strategic opportunism when interests are not well aligned and that it is quite vulnerable to myopic opportunism even if interests are aligned; (b) that it is easily displaced by a hedonic frame if individuals experience loss (especially, but not exclusively, due to agreements against breach even in case of golden opportunities and due to frustrated expectations, such as unfairness); (c) that it tends to crowd out enjoyment and obligation as sources of motivation; and (d) that it needs good measurability of performance. Among the most important advantages of a normative frame are (a) that it effectively deals with myopic opportunism by making obligations an important source of motivation; (b) that it makes individuals sensitive to rules and rule following; and (c) that its requirements for the measurability of performance are relatively low. Among the most important disadvantages of a normative frame are (a) that performance is not directly linked to improvement and individual initiative; (b) that it tends not to be result-oriented; and (c) that it is difficult to stabilize against being displaced by a gain or hedonic frame. How can we maximize the advantages and minimize the disadvantages of both? Remember that a frame switch comes about when the salience of the frame is too low. One way this can happen is that background goals are increasingly violated by the behavior that is generated by the foreground goal (i.e. the frame). This allows a dynamic relationship between frames such that if behavior in a frame violated important goals in the background, the frame is likely to switch, only to be replaced again when, in turn, behavior in the new frame violates important background goals. Such a dynamic interdependence can be achieved between gain and normative concern. When a person is in a gain frame and there is normative concern for the relationship with others, then, as the pursuit of gain puts a burden on the ongoing relationships, the salience of the gain frame will decrease, making it likely that the gain frame will be displaced by a normative frame. For example, an employer is under pressure to lower costs of production and, bit by bit, increases the workload of certain employees. As he does so, normative concerns for the relationship with the employees in the background become stronger, lowering
The Cognitive Side of Governance
65
the strength of the gain frame, possibly to such a degree that it will be replaced by a normative frame, in which the concerns for acting appropriately within the relationships with his employees come to inhabit center stage. The concerns for gain are pushed into the background. While he is in this frame, his actions are focused on doing what is “right” for his employees. He will show concern for their plight, maybe make some special effort to spot problem cases, hire temporarily some extra help to mitigate the worst problems, and promise to take employee concerns well into consideration when taking on orders and making commitments for delivery deadlines. As he does so, he reduces his flexibility, he pays for extra help, he loses time for other tasks by being concerned with the employees, and the goal of gain pursuit, in the background, increasingly comes under pressure, lowering the strength of the normative frame, and likely effecting a switch back to a gain frame, restarting the cycle, but from a different status quo (one in which expenses have been incurred, reassurances given and promises made). Observe that the process is not one of maximizing a complex function with both gain and relational elements, even though, in hindsight, the frame switch episodes may be experienced as simultaneous concerns for gain and for the relationship with employees. The difference between maximizing one function for both concerns and dynamic frame interdependence is that in the first case, we assume that there are no cognitive processes, such as myopic opportunism, that interfere with such a maximization exercise. There are such processes, and they necessitate special governance instruments. Thus, we do not gain anything by assuming, for simplicity’s sake, that the employer maximizes a complex function. Rather, by doing so, we lose sight of the most important instruments needed for effective governance of adaptive behavior in joint production. For the following, we thus try to answer the question how such a dynamic frame interdependence can be achieved and maintained.14 In order to be potential background goals, gain and normative concerns must both be important goals in the organization. There are various ways in which gain and normative concern may become important in an organization. The central point here is that for the dynamic interdependence between the two, one should not be made important in such a way that it jeopardizes the importance of the other. For example, an internal labor market, with strong status distinctions between ranks coupled with large differences in power and compensation is an excellent instrument for making gain an important goal in the organization. However, if one does that, many hurdles for normative frames may arise. For one, in the absence of good performance measures, arbitrariness is likely to undercut any attempt to make normative concerns important. Rather, feelings of being treated unfairly will make hedonic frames prominent. Second, even if good performance measurement is possible, the normative concern for relationships will very likely be dwarfed
66
SIEGWART LINDENBERG
by the salience of gain, no matter how incompatible the gain is with normative concerns. Conversely, take a flat organization with a strong ideological foundation (which helps to define appropriateness for most situations); a leader who requires accountability; and high exit costs. In such a “sect-like” organization, normative frames are likely to be so strong that it will be virtually impossible simultaneously to encourage gain frames for interaction among the members and to encourage a normative frame towards individuals outside the organization. The dynamic frame interdependence should thus make possible what I have elsewhere called “weak solidarity” (Lindenberg, 1998). In weak solidarity, the norms that belong to the “normative concern” are tied to a much more limited sacrifice that can be legitimately expected from the individual than in strong solidarity. In that sense, weak solidarity is indeed a weakened version of strong solidarity. However, the norms that belong to weak solidarity are not just weaker versions of the norms of strong solidarity because the latter are not compatible with gain. Norms of weak solidarity stress the importance of the individuals or the transacting dyad (as opposed to the importance of the group in strong solidarity). As distributional norm, weak solidarity has equity (as opposed to equality in strong solidarity). As with strong solidarity, authority in weakly solidarity groups must be legitimized and maintained in terms of the contribution to the common goal. If this is the case, status differences are likely to develop on this basis (see Ridgeway & Walker, 1995). Because of the equity principle, status differences (which are based on difference in contribution) are likely to be accompanied by different remuneration.15
Alignment by Mutual Commitment: The Role of Task Enjoyment, Improvement, and Quality of Relations In order to get gain and normative frames to coexist dynamically, it is important to make gain and normative concern strong enough to be frames (i.e. to be focal goals) but weak enough to be displaced by the other when incompatibility is more than minor. Observe that it is not just a matter of making the frames weak enough to be displaced, but weak enough to yield to the other frame rather than to a hedonic frame. The guiding idea about how to do it is actually quite simple. The organization must convey convincingly that it is all about joint production (rather than workers versus capital, leaders versus followers, superiors versus subordinates). Since Durkheim, sociologists have known how this jointness can be achieved. It happens through rituals and symbols of jointness (including emblems, joint excursions, festivities, etc.) and to an important degree through ingroup/outgroup measures of fostering norms of solidarity and thus of relational concern within the ingroup (see Lindenberg, 1998). For example, an organization may encourage
The Cognitive Side of Governance
67
three kinds of comparisons to other organizations in the industry. One with regard to “inferior” organizations, one with regard to competitive organizations, and one with regard to exemplary organizations (past or present) that should be emulated. In a fractal design, similar but less dramatized comparisons may be encouraged among divisions and subdivisions inside the organization. This is a very powerful tool, but it generates high relational expectations. If an organization creates relational concerns through governance measures that emphasize jointness of the entire organization and within all its parts, it is most important that the relational concern clearly shows up in commitments of the strongest partner to each and all of the other partners, especially the individual employee and the (sub)division he or she belongs to. In order to keep the gain frames tied to instrumentally adaptive behavior towards the realization of organizational goals (i.e. in order to reduce strategic opportunism), the organization must align the gain interests of the employees with the organizational goals, as has been elaborated by Williamson (1985). However, it was one of the major points of this chapter that this alignment alone will not do. There are too many threats against solidary behavior left even when interests are aligned. The dangers lurk in low measurability of performance, in motivational decay, and in myopic opportunism. Thus, in addition to interest alignment, the organization has to show commitment to the individual employee, in terms of task enjoyment, in terms of individual improvement, and in terms of quality of its relation with the employee and of the relations between employees. In return, the organization can legitimately expect commitment of the employee to the goals of the organization and to the quality of relations within it. For this, the employee can be held accountable without negative relational consequences. Employees who are approached with such commitments by the organization will not see its intrinsic motivation crowded out. They are clearly not approached as strategic actors and are unlikely to respond as strategic actors. If successful, we have a fundamental alignment of frames between employees and the organization. Interest alignment must thus not be seen as something separate, as a strategic move of the organization but as part and parcel of the relational interest of the organization in the employee. This overall alignment by mutual commitment to adaptively advance a common goal can be made more likely by structural arrangements (see below) but it will have to be gained and regained through deeds that signal continued commitment.16 Most governance tools can be interpreted as instrumental towards the achievement of this fundamental alignment. For example, schemes to share in the positive results of the organization (for example through stock options) cannot always be applied (due to the kind of organization, the measurability of the performance etc.). But where it can be applied, it is a good instrument to combine commitment to individual improvement and to relational quality. Here, the lessons from transaction
68
SIEGWART LINDENBERG
cost economics (interest alignment through credible commitments), from the organizational behavior approaches (with emphasis on fostering commitment) are actually combined in a theoretically fairly coherent way. As discussed above, task enjoyment as a background goal is quite different from creating the atmosphere of “work as a fun place,” since the latter caters directly to a hedonic frame with all the problems that entails. Also, the commitment to individual improvement is quite different from offering an internal labor market with a structure of positions for which individuals may compete (internal tournament). Tournaments signal lack of commitment of the organization to employees who lose in the competition and they reduce relational concerns among employees. Conversely, the commitment to relational quality is quite different from making rules and rule following behavior the central focus of governance. Emphasis on rules may foster normative frames, but it can be quite incompatible with relational concerns. In general, there are a number of management tools available that are likely to achieve commitment to individual improvement and relations, and to task enjoyment in the background (see Lindenberg, 1993, 2001b). What are they? To begin with, the hierarchy in the organization may not be a status command structure. Strong emphasis on rank is incompatible with task enjoyment for the lower-ranked members and with relational concerns for all involved. One way to avoid the negative aspects of rank is to have flat organizations. But in larger and more complex organizations, this cannot always be done. Activities need to be coordinated. Then the solution is to legitimize hierarchy in terms of the various functions that have to be fulfilled and in terms of the necessary information, knowledge, and responsibilities that cumulate in certain functions and functionaries. A directive is then not the order of a superior (the famous “fiat” in transaction cost economics that reduces transaction costs because it is a command). Rather, a directive is the knowledge-based and responsibility-based wish of somebody higher up in the hierarchy. It is commitment to the same organizational goals that makes the receiver adopt this wish, not the fact that he or she, in return for payment, has traded the right to be told what to within a prearranged “zone of indifference” in which own intelligent effort is suspended. In fact, in functional hierarchies directives must be seen as setting goals and as “promulgating standards and rules of the road” (as Simon, 1997, p. 233 observed). A functional hierarchy still offers incentives for promotion. Higher positions carry more prestige and more pay. However, as regards relational concerns, the negative sides of a strong emphasis on status and on rank have been removed. In order to facilitate functional hierarchies, organizations must be sure that employees have a good understanding of the functions and the entire workings of the organization. This can be achieved for example by job rotation and/or by periodic instruction.
The Cognitive Side of Governance
69
Not everybody can advance up the hierarchy. Disappointed expectations create the experience of loss which, in turn, strengthens hedonic frames and thus stronger influences of moods. The organizational commitment to individual improvement can thus only be loosely coupled to advancement up the hierarchy. Instead, the organization can offer a great number of alternative ways to improve. Seniority pay is one prominent solution. Other solutions include the creation of personal advancement schemes (without positional change), as in the Japanese merit evaluation (satei), and training for possible advancement outside the organization. Improvement may also take the form of reduction in tasks that are experienced as burdens (such as teaching load in some universities), or increase in flexibility. In sum, the more ways of improvement there are and the clearer the commitment of the organization to individual improvement, the fewer are the negative consequences of not being promoted to a higher position, and the greater the chance of a dynamic interdependence between a gain orientation and relational concerns as frames. With regard to the commitment to relational quality, it is most important that the organization itself be committed and not just encourages good relations among its employees. Investigating the role of commitment to relational quality with data on Japanese and American companies, M¨uhlau (2000) found that it is indeed the commitment of the central organization to relational quality that makes the difference. High relational quality among peers in the absence of this central commitment does not create commitment to the organization. The reciprocating commitment of employees to relational quality can be achieved by the organization through a number of instruments that can be gleaned from the literature on relations and on organizational commitment (see, for example, M¨uhlau, 2000 for a discussion of this literature). For reasons of space, I will simply list them. First of all, the organization must be perceived by employees as an organization that communicates clearly, honestly, and with respect for its employees in the way language is chosen. Second, the organization must be reliable, that is, it has to keep to agreed rules and promises. Third, the organization must be committed to standards of fairness and equity. Fourth, employees must feel that their participation in the organization is encouraged. Fifth, the employees must have the right to appeal decisions and to organize collectively if they so desire. Sixth, the more dependent the employee, the more the employee must have the feeling that the powerful organization is concerned about his or her well-being. For example, the more dependent the employee, the more the employer must ensure health and safety of the work practices and make sure that the employee is insured against important risks and provided with retirement schemes. Seventh, the employee is held accountable for his or her commitment to the goals of the organization and to relational quality. Failure to show commitment to these seven
70
SIEGWART LINDENBERG
points will lead to a failure of the governance of motivation for adaptive joint production in the sense that the reciprocal commitment by employees will not be stable. With regard to task enjoyment without a hedonic frame (i.e. enjoyment as an extra stabilizer of the dynamic interdependence of gain and normative frames), it can be said that it is likely that over time, enjoyable tasks will become less enjoyable due to decreasing stimulation, pride and praise, and increasing perception of the costly side of the activities. This means that the governance of motivation must also be directed at periodic realignment of the tasks and enjoyment, either through task rotation, task redesign, changing task sharing or new task evaluation. Such periodic realignment is important but it should not be accompanied by an official company standpoint that jobs should be enjoyable in order to create commitment by employees. As mentioned before, such an emphasis on enjoyment would displace both gain and normative frames by hedonic frames, to the detriment of task related motivation. Thus, there is a fine line between being concerned about the enjoyability of tasks and yet not making that a central issue, not even in advertising, such as “the best guarantee of quality is that our employees thoroughly enjoy what they are doing.”
Relational Signals, Control, and Frame Stabilization The fundamental alignment is aided by structural measures, such as the right to appeal and organize, but for most governance tools that have just been described, it is true that they will only be as stable as the frames that generate them. To the employee, it is by no means taken for granted that the organization will remain committed to individual improvement and to relational concerns. People are generally aware that these commitments must be generated by orientations (or what we call frames) that may or may not be stable. The “right” orientation in this sense signifies relative lack of strategic behavior. For this reason, employees will interpret the organization’s actions as relational signals, as telltale signs of the “true” orientation (frame). For example, M¨uhlau (2000) found that an employee also looks at how the employer treats other employees in order to judge the organization’s commitment to relational concerns. Strategic as if relational campaigns that actually only try to create the appearance of relational concern and concern for individual improvement, will not be effective for long and will ultimately cost the organization dearly in terms of lost reciprocal commitment of employees. This actually helps managers to take relational concerns seriously and thereby set up dynamic frame interdependence. One can say that the commitments to individual improvement and relational quality must be continuously signaled by the organization, including those cases where there is bad news. If reorganization is necessary, then
The Cognitive Side of Governance
71
it is even more necessary to communicate openly, show commitment to fairness etc. Once the right governance tools are set up, governance in the daily interactions in an organization is equal to the exchange of relational signals. There are situations when control has to be exercised. When things have gone wrong, supervisors or colleagues are likely to be negatively affected and develop a regulatory interest. The question is how this regulatory interest can be satisfied within the context of dynamic frame interdependence. The crucial point is: how can control be exercised in such a way as not to be a negative relational signal? If control does become a negative relational signal (as it might have to if previous attempts did not result in changed behavior), it is likely to create a spiral of deterioration that renders dynamic frame interdependence virtually impossible. Fear of strategic behavior of the employer will make it more likely that the employee at first will respond by being strategic him- or herself (in a gain frame). However, the experience of frustrated expectations is likely to lead to the experience of loss and thus to a hedonic frame. Relational concerns (in a normative frame) are then virtually beyond repair. This is clearly a failure of the whole set up. An organization that never began with relational concerns will not have this kind of mishaps but then it will also be stuck with a high level of strategic and myopic opportunism. So, what can be done to keep these kinds of failures to a minimum? The answer is again to be found in relational signaling itself. There are some positions in the organization that can exercise control without ambiguity of signal. Control has to be exercised by these positions. This has been shown by Lazega and by Wittek. Lazega’s (2001, p. 201ff.) studies on collegiate control show that informal control only works if the act of control does not simultaneously signal possible strategic behavior in the guise of relational interest. And he shows that control tends to shift to those who can exercise it without ambiguous or negative relational signal. Wittek (1999) describes how a technological change (the addition of a large machine in a paper factory) created a breakdown in relational signaling due to ambiguity in control, and that after that, tasks had been redesigned and both control and relational signaling were restored. In other words, in organizations, there are clear efforts to make relational signaling work and to avoid the negative spirals that occur when control is not well aligned with relational signals (see also Wittek et al., 2003). Governance is thoroughly interwoven with relational signaling because it is all about managing motivation through frames in this final analysis.
CONCLUSION Governance in organizations has traditionally been associated with lowering transactions costs between contracting parties and arranging incentives in such a way that the employee works in the interest of the organization. From the point of
72
SIEGWART LINDENBERG
view of organizational behavior approaches and human resource management, it became clear that what really matters in organizations is the management of motivation, especially in organizations in which employee’s intelligent effort is required for adaptive behavior. It is not quite clear how the two approaches link up. They should link up because the workings of incentives are actually dealing with motivation. In the literature, the two approaches run side by side and at times, they are put together, each offering some important factors, but no common theory. The result is that many ad hoc arguments are used and many problems are covered up or ignored. Very likely, this state of affairs is due to the fact that both approaches pay too little attention to the cognitive aspects of governance and of motivation. If one has a serious look at the cognitive aspects, one discovers that governance is first and foremost governing cognitive processes that, in turn, are vital for motivation. The way individuals define (“frame”) a situation is crucially important for what they consider and what they ignore. People influence each other’s frames and on this basis, cognitive coordination is possible. Three basic frames were distinguished in this chapter: hedonic, gain, and normative frames. A priori, the hedonic frame is stronger than the gain frame, which, in turn, is stronger than the normative frame. It was argued that governance would have to bring about that hedonic frames do not displace gain and normative frames. It was also argued that governance on the basis of either gain frames or normative frames is likely to run into very large trouble. The solution lies in the establishment of a dynamic frame interdependence between gain and normative frames. In such a situation, relational concerns keep the pursuit of gain in check and, conversely, the pursuit of gain keeps excesses of relational concerns in check. The basis for achieving such a dynamic interdependence is the organization’s ability to convey convincingly that it is all about joint production (rather than labor versus capital, leaders versus followers, superiors versus subordinates). For this to work, the organization must show commitment to the individual employee, in terms of task enjoyment, in terms of individual improvement, and in terms of quality of its relation with the employee and of the relations between employees. The interest alignment through credible commitments, argued for by transaction cost economics, is an important part of this commitment to the individual employment and it is likely to work only if it is not interpreted as a strategic move by the organization to keep the employee from shirking or malversation. Interest alignment is then flanked by the legitimate expectation by the organization that the employee is committed to the goals of the organization and to the quality of relations within it. It is this legitimate expectation that enables the organization to make the employee accountable for this commitment without negative relational consequences. This mutual commitment is quite resistant to problems of measurement of performance. However, because the frames on which the commitment rests are in principle precarious, the
The Cognitive Side of Governance
73
commitment has to be gained and regained in the daily interaction by relational signals. Governance in the day to day interactions in organizations is likely to be to a large degree seeing to it that the “right” relational signals are given and that control efforts are well in tune with these signals. This chapter contains many very concrete suggestions on governance that come out of the closer consideration of the cognitive impact on motivation. Most of these could not have been made in any systematic way without giving the cognitive side of governance a prominent place.
NOTES 1. Tractability refers to the ease with which one can trace the consequences of making certain assumptions. 2. Because Baron and Kreps use game theory in their attempt to work out the interdependencies among actors more clearly, I will refer in this paper at times specifically to the difference between the cognitive approach presented here and the game theoretic approach. 3. This has been clearly seen by Frey and Osterloh (2000). However, they do not yet use the cognitive aspects of governance for their analyses. 4. According to the Oxford English Dictionary, “adaptive” means that one fits one’s actions to changing circumstances, guided by a purpose. 5. In game theory, the question of possibly competing goals is side-stepped by assigning payoffs to particular event combinations, thereby seemingly concentrating on the combined structure of events and payoffs of the interactants, rather than their goals. No attention to goals seems necessary. However, in all applications to real life situations, goals will have to be made explicit to guide the assignment of payoffs. 6. Foss and Lorenzen (2001) also deal with issues of cognitive coordination albeit not from a framing point of view. Game theory assumes common knowledge and consistently aligned beliefs before the game starts. Neither the achievement of common knowledge nor the alignment of beliefs (and expectations) are part of the game itself. By this assumption, we are encouraged not to ask the question by what processes cognitive coordination is brought about and what processes work against cognitive coordination. What would a game look like that was cognitively uncoordinated (say one would see it as a Prisoner’s Dilemma, the other as a Chicken Dilemma, see Boudon, 1981)? What social processes are at work to get cognitive (and thereby motivational) coordination established? 7. These effects need not be conscious or work via prior intention. For example, in a situation in which others speak highly of the value of achievement a person can get “primed” to focus on achievement without being aware of it (see Bargh, 1997). 8. Simon (1997) has made a related point when he drew our attention to the fact that governance is in part equal to influencing the premises of decisions. 9. In addition, the explanation of myopia as discounting (say, related to the interest rate when money is concerned) does not hold up to empirical tests. Rational discounting (i.e., discounting related to the interest rate) does very badly in empirical tests and other discounting functions (such as hyperbolic) do better but still not very well (see Gattig, 2002). “Information impactedness,” as Williamson (1985) calls his version of the boundedness of rationality, is a matter quite unrelated to individuals’ neglect of long-term aspects they are well acquainted with.
74
SIEGWART LINDENBERG
10. Of course, there can be direct effects of, say, your efforts to increase your earnings status and comfort. Notice that we are not talking about effects of having money (even without spending it) but of improving one’s earnings. For most resources this hedonic link to efforts of improvement is lacking. 11. Motivaction uses a “mentality model.” These four types have recently been extracted from a larger set of “mentality styles” reported in Spangenberg et al. (2001). 12. Williamson states that people will make “calculated efforts to mislead, distort, disguise, obfuscate, or otherwise confuse” (Williamson, 1985, p. 47). 13. Why role theory is not a good theory of action is by now subject of many publications (see, for example, Lindenberg, 1990). 14. Interestingly, March and Olsen also see the necessity for a balance between what they call “efficiency” and “adaptiveness” (1995, p. 213ff.). Only they cannot link this balance to frames because they admit only a normative frame. 15. This view of weak versus strong solidarity fits with Fiske’s (1991) distinction between “equality matching” and “communal sharing” relationships. Observe, though, that authority relations can also be governed in terms of weak solidarity. 16. Rousseau (1995) speaks of a “psychological contract” in this regard. However, Rousseau sees such a contract leading to “scripted behavior,” routine role fulfillment. This view completely misses out on the precariousness of the frames and, in consequence, it also misses out on the governance structures and patterns of interaction in organizations that deal with this precariousness.
ACKNOWLEDGMENTS This paper was written while the author enjoyed a stay at the Netherlands Institute for Advanced Study in the Humanities and Social Sciences (NIAS) in Wassenaar. The author gratefully acknowledges the support of NIAS. In addition, the author is grateful for the constructive suggestions for improvement made by the editors of this volume.
REFERENCES Akerlof, G. A. (1982). Labor contracts as partial gift exchange. Quarterly Journal of Economics, 97, 543–569. Andreoni, J. (1988). Why free ride? Strategies and learning in public goods experiments. Journal of Public Economics, 37, 291–304. Bargh, J. A. (1997). The automaticity of everyday life. In: R. S. Wyer (Ed.), The Automaticity of Everyday Life. Advances in Social Cognition (Vol. 10, pp. 1–61). Mahwah, NJ: Lawrence Erlbaum. Baron, J., & Kreps, D. M. (1999). Strategic human resources. New York: Wiley. Boudon, R. (1981). The logic of social action. London: Routledge & Kegan Paul. Carnevale, P. J. D., & Lawler, E. J. (1986). Time pressure and the development of integrative agreements in bilateral negotiations. Journal of Conflict Resolution, 30, 636–659.
The Cognitive Side of Governance
75
Clark, M. S. (1981). Noncomparability of benefits given and received: A cue to the existence of friendship. Social Psychology Quarterly, 44, 375–381. Deci, E. L., & Ryan, R. M. (1985). Intrinsic motivation and self-determination in human behavior. New York: Plenum Press. Dhar, R., & Wertenbroch, K. (2000). Consumer choice between hedonic and utilitarian goods. Journal of Marketing Research, 37, 60–71. de Dreu, C. K. W., & Boles, T. L. (1998). Share and share alike or winner take all? The influence of social value orientation upon choice and recall of negotiation heuristics. Organizational Behavior and Human Decision Processes, 76, 253–276. Durkheim, E. (1964). The division of labor in society. New York: Free Press. Fiske, A. P. (1991). Structures of social life. New York: Free Press. Fiske, S. T., & Taylor, S. E. (1991). Social cognition (2nd ed.). New York: McGraw-Hill. Folger, R (1998). Organizational justice and human resource management. Thousand Oaks: Sage. Folger, R., Rosenfield, D., & Hays, R. (1978). Equity and intrinsic motivation: The role of choice. Journal of Personality and Social Psychology, 36, 557–564. Foss, N., & Lorenzen, M. (2001). Cognitive coordination and economic organization: Analogy and the emergence of focal points. Unpublished manuscript, Copenhagen Business School. Frey, B. S., & G¨otte, L. (1999). Does pay motivate volunteers? Working Paper Series No. 7. Institute for Empirical Research in Economics. University of Z¨urich. Frey, B. S., & Osterloh, M. (2000). Managing Motivation. Wie Sie die neue Motivationsforschung f¨ur ihr Unternehmen nutzen k¨onnen. Wiesbaden: Gabler. Gattig, A. (2002). Intertemporal decision making. Amsterdam: Thela Thesis. Gollwitzer, P. M., & Moskowitz, G. B. (1996). Goal effects on action and cognition. In: E. T. Higgins & A. W. Kruglanski (Eds), Social Psychology. Handbook of Basic Principles (pp. 361–399). London: Guilford Press. Kahneman, D., Knetsch, J. L., & Thaler, R. H. (1991). The endowment effect, loss aversion, and status quo bias. Journal of Economic Perspectives, 5, 193–206. Kim, W. C., & Mauborgne, R. (1998). Procedural justice, strategic decision making, and the knowledge economy. Strategic Management Journal, 19, 323–338. Kochanska, G., Murray, K. T., & Harland, E. T. (2000). Effortful control in early childhood: Continuity and change, antecedents, and implications for social development. Developmental Psychology, 36, 220–232. Kruglanski, A. W. (1996). Motivated social cognition: principles of the interface. In: E. T. Higgins & A. W. Kruglanski (Eds), Social Psychology. Handbook of Basic Principles (pp. 493–520). London: Guilford Press. Lazega, E. (2001). The collegial phenomenon: The social mechanisms of cooperation among peers in a corporate law partnership. Oxford: Oxford University Press. Ligthart, P. A. M. (1995). Solidarity in economic transactions: An experimental study of framing effects in bargaining and contracting. Amsterdam: Thela Thesis. Lindenberg, S. (1990). Homo socio-oeconomicus: The emergence of a general model of man in the social sciences. Journal of Institutional and Theoretical Economics, 146, 727–748. Lindenberg, S. (1993). Club hierarchy, social metering and context instruction: Governance structures in response to varying self-command capital. In: S. Lindenberg & H. Schreuder (Eds), Interdisciplinary Perspectives on Organization Studies (pp. 195–220). Oxford: Pergamon Press. Lindenberg, S. (1998). Solidarity: Its microfoundations and macro-dependence. A framing approach. In: P. Doreian & T. J. Fararo (Eds), The Problem of Solidarity: Theories and Models (pp. 61–112). Amsterdam: Gordon and Breach.
76
SIEGWART LINDENBERG
Lindenberg, S. (2001a). Social rationality versus rational egoism. In: J. Turner (Ed.), Handbook of Sociological Theory (pp. 635–668). New York: Kluwer/Plenum. Lindenberg, S. (2001b). Intrinsic motivation in a new light. Kyklos, 54, 317–342. Loewenstein, G. F., & Elster, J. (1992). (Ed.) Choice over time. New York: Russell Sage. Loewenstein, G. F., & Thaler, R. (1989). Anomalies: Intertemporal choice. Journal of Economic Perspectives, 3, 181–193. Mano, H., & Oliver, R. (1993). Assessing the dimensionality and structure of the consumption. Experience: Evaluation, feeling, and satisfaction. Journal of Consumer Research, 20, 451–466. March, J. G., & Olsen, J. P. (1989). Rediscovering institutions. The organizational basis of politics. New York: Free Press. March, J. G., & Olsen, J. P. (1995). Democratic governance. New York: Free Press. McGregor, D. (1960). The human side of enterprise. New York: McGraw-Hill. Meyer, M. W. (1994). Measuring performance in economic organizations. In: N. J. Smelser & R. Swedberg (Eds), The Handbook of Economic Sociology (pp. 556–578). Princeton, NJ: Princeton University Press. Moorman, R. H. (1991). Relationship between organizational justice and organizational citizenship behaviors: Do fairness perceptions influence employee citizenship? Journal of Applied Psychology, 76, 845–855. M¨uhlau, P. (2000). The governance of the employment relation. A relational signaling perspective. Amsterdam: Thela Thesis. Nee, V. (1998). Norms and networks in economic and organizational performance. American Economic Review, 2, 85–89. Ormel, J., Lindenberg, S., Steverink, N., & Verbrugge, L. M. (1999). Subjective well-being and social production functions. Social Indicators Research, 46, 61–90. Osterloh, M., & Frey, B. S. (2000). Motivation, knowledge transfer, and organizational form. Organization Science, 11, 538–550. Raven, B. H., & Rietsema, J. (1957). The effects of varied clarity of group goal and group path upon the individual and his relation to his group. Human Relations, 10, 29–45. Ridgeway, C. L., & Walker, H. A. (1995). Status structures. In: K. S. Cook, G. A. Fine & J. S. House (Eds), Sociological Perspectives on Social Psychology (pp. 281–310). Boston, MA: Allen and Bacon. Rousseau, D. M. (1995). Psychological contracts in organizations. Understanding written and unwritten agreements. Thousand Oaks, CA: Sage. Spangenberg, R., Lampert, M. A., Moerland, M., Boerboom, H., & van der Lelij, B. (2001). Burgerschapsstijlen en overheidscommunicatie. Rapport commissie overheidscommunicatie. Amsterdam: Motivaction. Simon, H. (1997). Models of bounded rationality. Volume III: Empirically grounded economic reason. Cambridge, MA: MIT Press. Williamson, G. M., & Clark, M. S. (1989). The communal/exchange distinction and some implications for understanding justice in families. Social Justics Research, 3, 77–103. Williamson, O. E. (1985). The economic institutions of capitalism. New York: Free Press. Wittek, R. (1999). Interdependence and informal control in organizations. Amsterdam: Thela Thesis. Wittek, R., van Duijn, M. A. J., & Snijders, T. A. B. (2003). Frame decay, informal power, and the escalation of social control in a management team: A relational signaling perspective. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 355–380). Oxford: Elsevier Science.
QUALITY, EXCHANGE, AND KNIGHTIAN UNCERTAINTY Joel M. Podolny and Greta Hsu ABSTRACT Sociologists have long recognized that stable patterns of exchange within a market depend on the ability of market actors to solve the problem of cooperation. Less well recognized and understood is a second problem that must be solved – the problem of Knightian uncertainty. This chapter posits that the problem of Knightian uncertainty occurs not only in the market; it underlies a variety of exchange contexts – not just markets, but art worlds and professions as well. These three exchange contexts are similar in so far as a generally accepted quality schema arises as an important solution to the problem of Knightian uncertainty; however, the quality schemas that arise in these three contexts differ systematically along two dimensions – the complexity of the schema and the extent to which the “non-producers” have a voice in the determination of the quality schema. By comparing and contrasting the way in which quality schemas arise in these three domains, this chapter (1) gives some specificity to the notion of quality as a social construction; (2) provides some preliminary insight into why a particular good or service becomes perceived as a market, artistic, or professional offering; and (3) offers an imagery for conceptualizing the mobility of goods and services between these three domains.
The Governance of Relations in Markets and Organizations Research in the Sociology of Organizations, Volume 20, 77–103 © 2003 Published by Elsevier Science Ltd. ISSN: 0733-558X/PII: S0733558X0220004X
77
78
JOEL M. PODOLNY AND GRETA HSU
INTRODUCTION Sociologists have long recognized that stable patterns of exchange within a market depend on the ability of market actors to solve the problem of cooperation or – as Parsons labeled it – the Hobbesian problem of order. Indeed, sociological inquiry into this problem can be traced at least to the writings of Durkheim and Weber. Durkheim emphasized the non-contractual elements of contract. Weber highlighted the institutional and reputational underpinnings of the transactions on equity exchanges. Modern sociologists have elaborated upon and refined these ideas (see the review by Voss, 2003 in this volume). For example, in his work on business groups, Granovetter (1995) offers some grounding to Durkheim’s conception of the non-contractual elements of contract. Considering a diverse array of economies, Granovetter observes that business groups function as “moral communities,” fostering cooperation in the flow of resources and goods. By focusing on the role of personal networks in commodity and equity exchanges, Baker (1984) and Abolafia (1997) have added much to Weber’s (1978) initial observation on the topic, and at a more general level, Raub and Weesie (1990) and Kollock (1994) demonstrate the role that reputation can play in fostering and sustaining cooperation among exchange partners. In this volume, Buskens et al. (2003) as well as Stuart (2003) offer empirical evidence how different forms of governance depend on networks and reputational concerns induced through network ties. There is parallel work within the field of economics. McMillan (2002), for example, considers the institutional underpinnings of competitive markets. Gametheorists explore coordination problems that arise in markets where, for example, producers can choose quality levels when manufacturing products (e.g. Stiglitz, 1987; Tirole, 1996). As an illustration, consider a market in which there are many producers and consumers. All producers are capable of manufacturing a low quality product, but one producer has the capability to produce either a low quality product or a high quality product. The cost of producing the higher quality product is greater than the cost of producing the lower quality product. The producer is willing to invest in manufacturing the high quality product if she can be assured that the rents from doing so are greater than the rents that she derives from producing the low quality one. In this market, there is at least one consumer who is willing to pay a price for the superior quality good that will provide the producer with greater rents than the producer could obtain from selling the lower quality good at the lower price. However, the consumer does not know the quality of the good until after he has purchased it, and since the cost of producing the inferior good is lower than the costs of producing the superior one, the producer has an incentive to manufacture and sell the inferior good even if the consumer pays the higher price. Knowing
Quality, Exchange, and Knightian Uncertainty
79
that the producer has such an incentive, the consumer refuses to pay a higher price. Thus, while both the consumer and the producer would be better off with the producer selling the superior quality good for the higher price, the superior good is not produced. Repeated transactions between a consumer and producer, reputational information passed between consumers, and signals of quality such as warranties all are mechanisms that help solve the cooperation problem, but absent such mechanism, the low quality good is produced. This issue is discussed in detail in Buskens et al. (2003) as well as Gulati and Wang (2003) in this volume. Yet while sociologists and economists continue to elaborate on the classic concern with problems of cooperation and coordination, the premise of this chapter is that the problem of “Knightian uncertainty” is an equally fundamental problem in market contexts. In his classic work, Uncertainty, Risk, and Profit (1921), Knight distinguished uncertainty from risk. Risk refers to a situation in which an actor knows the possible consequences of some action that he or she will undertake and is able to estimate the likelihood of those outcomes occurring. Uncertainty refers to a situation in which an actor foresees the possible outcomes but cannot reliably estimate the likelihood of those outcomes occurring. Economists, most notably Savage (1954), developed axiomatic rules of subjective expected utility that assumed away the distinction between risk and uncertainty. However, after a seminal paper by Ellsberg (1961), experimental economists began to demonstrate that actors did behave as if there was a distinction between risk and uncertainty, and in a broad class of situations, actors are uncertainty averse. For example, Becker and Brownson (1964) presented subjects with two urns: one contained 50 red balls out of 100, and the other contained a specified range of red balls with a center-point of 50. Across different subjects, Becker and Brownson manipulated the specified range. The subject received $1 if she drew a red ball. Becker and Brownson asked the subjects to choose one of the two urns from which to draw; they also asked the subjects how much they would pay to draw from the preferred urn. Becker and Brownson found that subjects always preferred the urn with the known number of red balls, and they were often willing to pay substantial premiums to draw from their preferred urn. So, when the range in the second urn was 0 to 100 red balls, subjects were willing to pay 36 cents, more than two-thirds of the expected value of the draw. Reviewing experiments like the Becker and Brownson study, Camerer (1995) argues that subjects are typically willing to pay 10–20% of expected value to avoid a situation of Knightian uncertainty. In an interesting variant on the basic experimental design, Sarin and Weber (1993) auctioned off urns with 10 balls inside. After winning the auction, the subject draws a ball, with a winning ball paying 5 marks. Some of the urns had a known distribution of five winning balls and five losing balls. Others had an
80
JOEL M. PODOLNY AND GRETA HSU
unknown distribution of winning and losing balls. The urns with a known distribution commanded a higher value. Such experimental results have powerful implications for how actors will allocate their time, energy, and resources across different markets. Indeed, these results have implications not only for the market – extending to practically any domain that can be understood in terms of exchange. Think of different exchange contexts as in some sense representing different urns. Actors give up time, energy, and/or resources to draw from any given urn, and therefore they must make some choices about which contexts/urns they wish to draw from and which they do not wish to draw from. In some particular exchange contexts, actors will have a strong basis on which to make inferences about the likely consequences of their “draw” in that context – for example, a certain rate of return for their time and effort. In other exchange contexts, actors will not have a particularly strong basis on which to make such inferences. The just discussed experimental results suggest that actors will be reluctant to make resource commitments to those exchange contexts that do not provide a tangible basis on which to infer the returns to action. Or, stated another way, Knightian uncertainty, like the Hobbesian problem of order, is a problem whose solution impacts on the level of resource commitments that are made to an exchange context. White (2002) explicitly recognizes the problem of Knightian uncertainty in elaborating the epistemological foundations of his model of production markets, and he argues that a collectively understood and enacted quality schema or array represents a socially constructed solution to the problem of Knightian uncertainty. We contend that White’s argument applies more broadly than production markets; it also applies to the domain of arts and professions. At a high level of generality, all three domains share a common social context – that of exchange. One set of actors – be they producers, artists, or professionals – aspire to offer something of value to another set – be they consumers, an audience, patients, or clients – with the hope of receiving financial and other forms of resources in return. Regardless of whether an actor is in the domain of art, professional jurisdiction, or a market, a collectively understood quality array provides a basis for plausible albeit implicit “if-then” statements regarding alternative allocations of resources. By “if-then” statements, we mean statements like the following: “If this piece of art has attribute X, then its value should be at least Y”; “if this provider of this service has credential A, then his efforts on my behalf are worth B”; “if I invest in equipment that allows me to produce a good with quality Q, then I should be able to command price P.” The more private the conceptions of quality, the less certainty that actors will have about the returns for different resource allocation decisions. To be sure, the fact that quality schemas represent a solution to the problem of Knightian uncertainty does not by itself imply that there will be a
Quality, Exchange, and Knightian Uncertainty
81
dominant, collectively supported schema in every context. It simply implies that ceteris paribus, more resources will be devoted to those contexts that do possess such an overarching quality schema. However, while a collectively understood quality schema represents a solution to the problem of Knightian uncertainty in each of these domains, we will use the rest of this chapter to hopefully convince the reader that there are fundamental and systematic differences across the domains in the nature of the schemas; first, in terms of the complexity of the schemas and, second, in terms of the extent to which those creating the objects or acts of value are able to exclude others from determination of the schema. There will be two payoffs from this consideration of the systematic ways in which quality schemas differ across these contexts. First, in undertaking a comparative analysis, each context represents a backdrop against which to evaluate the way in which the problem of Knightian uncertainty is solved in the other domains. Or, stated another way, the term “social construction” is an extremely vacuous one unless the distinct features of the construction become clear, and the comparative analysis therefore gives some analytical specificity to the claim that quality is socially constructed in each of these domains. A second payoff that accrues from an elucidation of these differences is a clearer understanding of why one valued object or act is considered art, another is considered a professional activity, and a third is simply a market object. There is an inherent flexibility in the domain to which a particular activity becomes assigned. Innate, objective aspects of an activity do not determine whether it comes to be defined as art, as an integral element of a profession, or something whose value is most appropriately determined by the forces of supply and demand. While a complete theory for the categorization of activities into these different domains is beyond the scope of this chapter, we hope to offer some initial insight into this question by sketching out what the schema-based solutions to the Knightian problem of uncertainty “look like” in each of these domains.
ART WORLDS We begin our investigation of the significance of quality schemas underlying exchange contexts by focusing on the artistic arena. In his influential text Art Worlds (1982), Howard Becker argues that art does not exist independently of an aesthetic that separates “good art” from “bad art” and even more fundamentally that distinguishes what is art from what is not. A creator or performer cannot convince others that the results of her effort merit the art label – let alone merit the claim of being good art – without some collectively accepted aesthetic or quality schema
82
JOEL M. PODOLNY AND GRETA HSU
to support the claim.1 A quality schema may arise before, during, or after the existence of works that epitomize it. However, regardless of the temporal sequencing, art is only manifest when the schema is manifest. For a creator or performer, the ability to produce work that others acknowledge as art is not simply a matter of self-esteem or bragging rights. Using jazz as an illustration, Becker argues that the ability to produce what others acknowledge as art has a tangible impact on the resources that the anointed artist can acquire. If a jazz performer can rightly claim to be creating art, then he or she can more effectively compete for grants from foundations and government agencies, positions within schools of music, and the right to perform in distinguished venues. For our purposes, one aspect of Becker’s understanding of the role of aesthetics in art worlds is especially noteworthy. The aesthetic underlying an art world is a collective construction that does not simply emanate from a single artist or even a group of artists. Rather, the aesthetic arises from the discourse of the disparate constituencies that comprise the art world. Certainly, there will be some art worlds in which some individuals take on the role of critics or aestheticians whose sole function in the art world is to elaborate the aesthetic. However, even if there are professional critics in a particular art world, these critics are not able to act as the sole arbiters of quality. Audiences, purveyors of art such as gallery owners or theater owners, and the artists themselves can impact the aesthetic, and the schemas of critics are as often a reaction or rationalization of these constituencies’ views as a determinant. There is a second noteworthy aspect of the quality schemas in artistic domains: they tend to be relatively complex. This second aspect does not figure as prominently in Becker’s characterization of aesthetics; however, it is implicit. For example, when Becker describes the aesthetic underlying photography, he points out that there are many aspects of quality that can be identified. Or, when Becker distinguishes arts from crafts, a major distinction is that crafts tend to be defined in terms of a relatively straightforward utilitarian objective. Art is completely decoupled from well-defined utilitarian objectives, and as a consequence, artists have considerably more freedom to produce works that exemplify a richer set of criteria than the meeting of a well-defined need. It is important at this point to define what we mean by complex. Our conception of complexity borrows heavily from NK fitness landscape models (Kauffman, 1993; Levinthal, 2001). In these models, complexity is a function, first, of the number of dimensions (N) of a space – which, in this case, is a quality-space. Second, and perhaps most importantly, complexity is a function of the interdependence of those dimensions (K) in determining the “fitness” or “perceived value” of the objects located in that space. If the dimensions of quality are completely independent, then the perceived value of an object is simply a linear, additive function of its
Quality, Exchange, and Knightian Uncertainty
83
attributes. Such a conception of quality-space is implicit in hedonic pricing models (e.g. Griliches, 1971; Rosen, 1974), which specify price as a linear combination of the attributes of a good. One can think of the level of interdependence as denoted by the underlying functional form linking the different attribute dimensions to the perceived value of the object or service offered in an exchange context; a simple schema is one denoted by a simple function. So, for example, a diamond’s value is determined by four factors: carat, color, clarity, and cut. Each of these dimensions is a relatively straightforward uni-dimensional construct with the exception of cut, which can be further subdivided into numerous sub-dimensions, such as depth, “girdle thickness,” and “crown angles.” If we could ignore the intricacies involved in determining how good the cut is, then the price of a diamond is a relatively simple function of these four attributes. The price of a diamond is not an additive linear function; a diamond that is high on all four dimensions is worth more than four diamonds that are each high on a single dimension and low on others. However, we speculate that one could explain considerable variance in the price of a diamond with a simple multiplicative function of the four attributes like the following: P = eB1(CARAT)+B2(COLOR)+B3(CLARITY)+B4(CUT) , where P denotes the price of the diamond and B1 through B4 could be estimated as regression coefficients. While the effects of each attribute on price are not linearly independent, the equation can obviously be converted into a simple linear function by taking the logarithm of both sides of the equation. In comparison, in the vast majority of art worlds, it is going to be very difficult to write a simple equation linking the features of an artistic object to their value or worth. Even if one can specify that the value of a painting is a function of its size, technique, use of color, subject matter, and so on, it is highly doubtful that there is a simple function linking such attributes to the perceived value of the artistic object. The consequence is that the artist’s search for higher quality occurs in a very rugged search landscape. The potentially large number of interdependent attributes underlying value means that there is no simple production function to guide the artist; minor differences in one attribute of quality can have huge implications for the over-all value because of the complex ways in which that one attribute interacts with others. If an artist finds a combination of attributes that are perceived to be of tremendous value (i.e. a peak in the rugged landscape), a minor shift of one attribute or the other can engender a significant drop in value. As a consequence, artistic reputations tend to become defined by a tightly circumscribed – albeit richly described – location in the quality space. A Monet, DeKoonig, or Pollock comes to stand for something very specific.
84
JOEL M. PODOLNY AND GRETA HSU
In short, in order for an act or object to be classified as art, we would contend that it must be collectively associated with a complex quality schema – as we have defined complexity. When the quality schema is simple, the object can at best merit the label of “pop art” or “low art,” but such adjectives as “pop” or “low” necessarily imply that the object is not “true art.” An often-heard criticism of low-brow movies is that they are formulaic – which seems nothing more than an assertion that the connection between attributes and perceived value is too straightforward. Importantly, while a schema helps to reduce the problem of Knightian uncertainty in art worlds by providing a basis for assessing the returns to some action, it tends to do so more for buyers than for sellers. Because of the complexity of the quality schemas in art worlds, artistic competence is an ephemeral attribute. However, complex quality schemas still allow for ex post rationalizations; they allow intermediaries to transactions such as critics or gallery owners to spin tales about why an artistic object is worth what it is worth and, in so doing, remove at least some key element of the Knightian uncertainty underlying this exchange context. Before concluding this discussion of the quality schemas in art worlds, it is important to acknowledge that not every participant in an art world necessarily appreciates the complexity of the schema underlying the offerings in that world. Indeed, there are typically some peripheral members of the audience that do not invest the time or effort to appreciate the world. Such peripheral members might choose to enter into exchange relations only on the basis of the status of those creating artistic objects. Or, they might adhere to simpler schemas than those who are most intensively involved members of the art world. How does the presence of such peripheral members impact on our observations? We would argue that to the extent that such peripheral members play an important and even a dominant role in exchanges within the art world, then the art world in fact becomes less an art world and more a market context. However, to understand why this is the case, it is necessary to discuss the market context in more detail.
MARKETS If a complex schema is one in which there are a large number of intricately related dimensions, then a simple schema is one in which there is a small number of separable dimensions. Whereas simple additive or multiplicative scales yielding some summary number to indicate quality are not appropriately applied to “real art,” such scales are extremely prevalent in market contexts. The magazine Consumer Reports rates practically all market products – from appliances to cars to batteries – on such scales. Indeed, pick up any computer or car magazine on a magazine
Quality, Exchange, and Knightian Uncertainty
85
stand, and there will typically be a section in which editors evaluate products by assigning numbers to some limited number of attributes and then aggregating up across those numbers to yield a single number denoting worth. The simplicity of quality schemas in markets and the fact that such rating schemes are easily applied to market products and not to high art has obvious implications for the classification of film as art or market product. We would argue that the general acceptance of film ratings on a uni-dimensional scale is by itself evidence that the vast majority of films are not appropriately categorized as art but are more appropriately considered market objects. Moreover, in order for films to acquire the status of art, a commonly perceived quality schema must arise that is sufficiently complex that any additive uni-dimensional rating would be a poor reflection of that schema. Sociological conceptions of markets either implicitly or explicitly assume that quality is of sufficiently low complexity that producers can be easily indexed or ranked. So, in Podolny’s (1993) status-based model of market competition, status, a uni-dimensional construct, is a signal for quality, which is also uni-dimensional. In White’s W(y) model (2002), a market can only exist to the extent that producers are able to index themselves in a uni-dimensional quality array that is isomorphic with the perceptions of those on the consumer side of the market. Because the solution to the problem of Knightian uncertainty in the market is simpler than the solution to the in the artistic context, there is a tendency to believe that the problem is either relatively minor or nonexistent. Put in terms of Becker’s art world, there is a tendency to assume that there is no aesthetic ambiguity that must be resolved in the context of the market. However, two works indicate quite strongly that this is not the case. One piece of research comes out of the field of marketing on what is known as the “pioneering advantage.” Carpenter and Nakamoto (1989) consider the situation in which an individual has no familiarity with a set of products in a market and then over time is exposed to those different products. Such a situation will of course arise for all consumers in a new market. Using an experimental research design, Carpenter and Nakamoto show that individuals’ beliefs about what constitute high quality are strongly influenced by the order in which they experience goods in a market. Attributes associated with the earliest goods experienced are more likely to come to be perceived by the individual as defining high quality. For example, suppose that when an individual enters the soft drink market, she first drinks a clear soda and generally likes the taste. The individual will now come to perceive “clear” as an important underlying dimension of quality even after the individual samples darker colas. If the individual had experienced one of the dark cola’s first, then Carpenter and Nakamoto’s work implies that the individual would be less likely to associate “clearness” with quality.
86
JOEL M. PODOLNY AND GRETA HSU
Carpenter and Nakamoto’s work provides a vivid illustration of the problem of Knightian uncertainty in markets. Absent the interaction with producers in the market, consumers have extremely ill-defined preferences about what constitutes high quality. A producer confronting such ill-defined preferences is in the same situation as the individual who draws from an ill-defined distribution of balls within an urn. However, as a producer enters a nascent market, the attributes of the producer provide an initial basis for a quality schema, and this quality schema then provides guidance for consumers and producers to select among alternative resource allocation commitments. Another relevant piece of research is Favereau et al.’s (2001) analysis of White’s W(y) model. An essential feature of White’s W(y) model is that all producers line up on a curvilinear array of volume-revenue choices. If a producer knows his quality, then the producer can use the information contained in the array to guide his pricing and volume decisions. However, the only way that the producer can know his quality is through observing his location in the array. In effect, the problem of Knightian uncertainty in White’s model is manifest as a “Catch-22.” If the producers know the quality distribution, they can choose the appropriate volume and price. However, they can only know their quality when the array emerges from all producers choosing the appropriate volume and price. How the problem becomes solved is of course outside the bounds of White’s equilibrium model, but for our purposes, what is important is that in the absence of a quality array, producers lack a tangible basis on which to assess the consequences of their resource allocation decisions and, in the presence of the quality array, such calculation is possible. If quality arrays provide solutions to the problem of Knightian uncertainty in both markets and art worlds, why is it that the products in “market worlds” conform to a simpler quality schema than products in art worlds? One important driver of the difference is the tremendous functionality of a simple quality schema in markets. In most market worlds, there is a purchase decision in which the selection of one object implies that another object will not be selected. In contrast, in many art worlds, purchasing is not the central act. For example, in the world of painting, there are many more who appreciate a given painting than purchase that painting. Even in those art worlds in which purchasing is a central decision, those creating objects or acts of value do not offer their objects with the hope or aspiration that their object or act of value will be chosen over another’s. So, for example, in a city with multiple dance companies, one local dance company does not aspire to be the sole provider for dance in the city. Rather, the aspiration of the dance company is that their performances will engender a sufficiently positive reaction that the audience attends not only performances of that particular dance company but the performance of other dance companies as well.
Quality, Exchange, and Knightian Uncertainty
87
Put simply, the more that the fundamental activity in an exchange context is that one set of actors must choose between the actors on the other side of the market, the more that those choosing will find a simple schema to be extremely functional; the simple schema enables competing offerings to be more easily ranked. To return to the example of diamonds, there is nothing inherent in diamonds that allows them to be rated on four independent and – with one exception – uni-dimensional attributes. Yet, because DeBeers, a vast network of distributors, and jewelers reify this particular quality schema through continued utilization, this particular quality schema directly helps provide a rational basis for buyers to discriminate among purchase options and, in so doing, helps provide suppliers with a rational basis for their own resource allocation decisions. Given the functionality of a simple quality schema in the market context, an important question arises: what mechanism or mechanisms facilitate the simplification of quality schema? Zuckerman (1999) highlights one mechanism – the sorting of market objects into exclusive categories that, in turn, enable simple rank-order comparisons. Zuckerman provides evidence of this sorting mechanism in his analysis of equity markets. On one side of the market are investors; on the other side are firms. The investors seek to provide capital to firms that will maximize the return on their investment. Just as moviegoers look to film critics in deciding between films to attend, investors look to the analysts of investment banks for guidance in choosing between firms to whom the should provide capital. In order to be able to compare firms, the analysts must first sort the firms into clear reference groups. One group, for example, might be health care firms; another might be entertainment companies. Importantly, while firms within a given group will tend to participate in the same markets, there is no overriding objective determinant of the actual boundaries. An analyst within a firm might be assigned multiple areas to track, but analysts only track those firms that fall neatly within a group. For the analyst, the focus on particular groups facilitates the construction of personal formulas or heuristics for the determination of firm value. As a result, firms that cut across reference groups tend to receive less attention from analysts than firms that fit neatly within one. The impact of these categorical schemas on market dynamics is significant. Zuckerman shows that, controlling for a variety of factors, firms that cut across reference groups have a lower market valuation than firms that fit neatly within a reference group. In effect, the lack of attention from analysts translates into a lack of dollars from investors. Moreover, firms that cut across reference groups are more likely to divest assets, presumably because they feel implicit or explicit pressure to be part of a reference group that a given analyst can evaluate. In short, in the equity markets, there is an analytical sorting of firms into distinct types that in turn facilitates a simple within-type ranking. This mechanism clearly exists in other markets. For example, if one
88
JOEL M. PODOLNY AND GRETA HSU
peruses automobile magazines, one observes that cars are first sorted into distinct types and then rated by type. To be sure, in art worlds, there are some sorting dynamics akin to the ones that Zuckerman observes. As White (1993) notes, a fundamental role of critics is to partition, and a clear implication of Becker’s work is that those who do not neatly fit into any particular art world tend not too receive as much attention. However, even if partitioning is a fundamental task of critics in art worlds, the fact remains that there is still greater heterogeneity within categories. Even though Cezanne, Monet and van Gogh could all be categorized as impressionists, it would be much more difficult to rank the three in terms of some parsimonious and independent measures of quality than to rank Pfizer, Eli-Lilly, and Merck as investment opportunities. So, to summarize up to this point, art worlds and markets are alike in so far as the problem of Knightian uncertainty must be solved in each, and the solution to the problem – a quality schema – involves the joint effort of potentially diverse actors in the “world” of interest. This joint effort results in general consensus among the various constituencies in both contexts regarding quality schemas. Art worlds and markets differ fundamentally, however, in the complexity of that solution to the problem. In the former, the solution is complex while, in the latter, the solution is relatively simple.
PROFESSIONS We now turn to a third category of exchange contexts – professions. In so doing, we draw strongly on the work of Abbott (1988). According to Abbott, the essence of professionalization is jurisdiction, where jurisdiction refers to the exclusive province of an occupational group over some particular class of problems. An important element of jurisdictional control is defining problems as amenable to the techniques of the occupation. Using the example of “drunkenness,” Abbott illustrates how different occupational groups – psychologists, medical doctors, clergy – have sought to define this particular problem as being in their particular sphere of competence. Another important element of jurisdiction is defining the means for approaching the problem. In effect, as an occupational group searches for problems over which it can claim jurisdiction, it also searches for means for approaching the problem. Abbott argues that the means for addressing the problem can be broken down into three different modalities: diagnosis, inference, and treatment. If an occupation has the status of a profession, it has the right to define its own unique pattern of diagnosis, inference, and treatment for addressing a problem. The English legal profession, for example, has formally set up a system where a client seeking legal help must first go to a solicitor for a diagnosis of
Quality, Exchange, and Knightian Uncertainty
89
her legal problem; if formal litigation is deemed necessary, the solicitor refers the case to a barrister who litigates the case within the professional system, producing a variety of possible treatments for the problem. The case then goes back to the solicitor and client, who reach a decision on treatment together. A third and final element of jurisdictional control is the right to define when the problem is – in effect – solved. During the early nineteenth century, the public accepted without question the medical profession’s decree that a treatment’s success should be measured by the amount of shock it produced in a patient. Homeopathic doctors, whose treatments produced relatively mild reactions, were thus at a disadvantage even though their treatments were less likely to kill patients than those of “regular” medicine. In short, jurisdiction implies the right to define the nature of the problem as amenable to the profession’s techniques, to decide upon the appropriate techniques for addressing the problem, and to decide when the problem is truly solved. To be sure, a profession cannot modify all aspects of its jurisdiction simultaneously. It cannot simultaneously claim new problems, develop new techniques, and offer new definitions for solutions without seriously threatening whatever legitimacy it has attained. However, if we compare professions to either the idealtypical providers of market goods or to artists, one salient difference arises: in the professional context, the non-producers (i.e. those who are recipients or observers of the profession’s services) have effectively no role in determining the value of a producer’s contribution. As we have just observed, in art worlds and in prototypical market contexts, consumers, critics, and other third parties to exchanges play a central role in judging the value of what a producer offers up for exchange. In contrast, the concept of jurisdiction implies no corresponding role for nonproducers in defining the value or quality of what a profession offers for monetary exchange. Interestingly, the idea that a profession should have the exclusive right to decide on whether a service is of value manifested itself early on medicine – in the writings of Hippocrates. In the first of his Aphorisms, Hippocrates wrote, “Life is short, and Art long; the crisis fleeting; experience perilous, and decision difficult. The physician must not only be prepared to do what is right himself, but also to make the patient, the attendants, and externals cooperate [italics added].” The implication is straightforward: the physician should have exclusive province to decide on what is most valuable for a patient and potentially use force if necessary to ensure that the path is followed. In drawing this distinction, it is worth observing that the most successful professions scrupulously avoid using the labels of “customer” or “buyer” in referring to recipients of their services. Doctors provide service to patients. Lawyers and accountants provide services to clients. The nomenclature is important in both
90
JOEL M. PODOLNY AND GRETA HSU
defining and reflecting a role for the recipient of services that carries relatively little voice in deciding what constitutes service of high value or high quality. A “customer” has a say in deciding whether or not the good that he or she has purchased is of the highest quality; a “patient” does not. If an individual or even corporation believes that its accountants have not provided the best tax advice, the only remedy is to consult another member of the profession and essentially get a second opinion. Perhaps no example better illustrates how the nomenclature for buyers relates to the jurisdiction of sellers than that of U.S. business schools. One of the most significant battle grounds in business schools from the late 1980s through today has revolved around the question of who gets to decide what constitutes quality teaching. On one side, a significant fraction of business school professors assert that they have the right to decide what constitutes quality, and their “students” should have at most a limited input into the assessment of how well they have been educated. On the other side, a significant fraction of those attending business schools assert that they are “customers” who are fully capable of deciding what constitutes quality. For the purpose of this chapter, we are not interested in whose view is correct. Rather, we only wish to observe that the “student” label is employed by those who reject the claim that the recipients of a service should be able to play a substantial role in assessing its quality, whereas the “customer” label is employed by those who support this claim. In effect, this is a context in which the professionalization of an area of knowledge is at most incomplete, and as a result the claims of business school professors to possess jurisdiction over the problems of management is highly contested. What factors determine the success of a profession in claiming jurisdiction? In order to answer this question, Abbott argues that one first needs to adopt a “system” perspective; each profession exists alongside others as part of an ecological system. Different professions often engage in competition for the jurisdiction over the same problem areas. Again, to return to the example of alcoholism, one profession’s success in claiming jurisdiction over a problem area is intimately bound to another’s failure. The question therefore can be reframed: what determines the extent to which one profession drives others from a particular problem area? Abbott draws attention to a number of factors. For example, he highlights the extent to which the objectives and activities of the profession further values that are central in the broader culture of society. Another obvious factor is the extent to which the profession can demonstrate its efficaciousness in addressing the problem area. Initial demonstrations of efficaciousness result in a public and potentially legal approbation of jurisdiction, and this public and legal granting of jurisdiction can be used by the profession in a number of ways. For example, the profession might seek to extend its jurisdiction into closely related areas. Abbott describes
Quality, Exchange, and Knightian Uncertainty
91
how, after its prominent “successes” with psycho-military problems during the First World War, psychiatry was able to expand its intellectual jurisdiction to adjacent social-problems areas such as juvenile delinquency. Or, the profession might seek to modify the definition of a solution in a way that further consolidates its jurisdiction. As the medical profession became increasingly successful in developing drugs that would eliminate diseases and infections, health increasingly became redefined as not having a disease or infection rather than as some general state of well-being. However, alignment with cultural values and efficaciousness are not the only factors that are relevant to jurisdiction. Indeed, for the purpose of this chapter, the most pertinent attributes are those relating to the profession’s theory underlying the connection between problem and solution. If an occupation is to be able to claim the stature and hence jurisdiction associated with being a profession, the occupation must offer a formal theory at a reasonably high-level of abstraction tying diagnosis, treatment, and inference to the problem, on the one hand, and the solution, on the other. Thus, one of the significant impediments to auto repair becoming a profession is the lack of a formal theory; in the absence of theory, practitioners simply seem to be learning from experience and following intuition; they are performing a craft rather than enacting a profession. What attributes of a formal theory foster acceptance of a profession’s jurisdiction? First, there must be a strong but still incomplete connection with the actual practice of the profession. Why incomplete? The reason is that the profession could then be routinized, which in turn would threaten the profession’s jurisdiction. As a consequence, the theory of what the profession does will typically be simpler than the practice of what the profession does even if there are significant touch-points between the theory and the practice. More generally, Abbott seems to conclude that the parsimony of a profession’s theory linking problem, diagnosis, treatment, inference and solution increases the profession’s ability to claim jurisdiction and exclude others from a particular problem area. A particularly compelling illustration comes from Abbott’s analysis of the “personal problems jurisdiction” from 1860 through the 1940s. Today, we perceive such problems as general angst, marital difficulties, trepidation surrounding work, and chronic albeit minor ailments like a “nervous stomach” as problems that must be addressed if an individual is to lead a normal, healthy life. In the mid 19th century, however, this was not the case. The first profession to claim jurisdiction were the clergy who addressed what they categorized as “personal problems” with a treatment of sympathy and support. However, the jurisdiction of the clergy was usurped by neurologists, who at the turn of the century formulated a category of problems referred to as “general nervousness,” later to be labeled “neurasthenia” and “psychasthenia.”
92
JOEL M. PODOLNY AND GRETA HSU
Neurologists were no more effective than the clergy in treating personal problems. However, neurologists won jurisdiction from the clergy for reasons that Abbott elaborates: [The neurologists’] construction of personal problems differed from the clergy’s in essential ways. First, it was narrower. Many personal problems of interest to clergymen had no direct interest for neurologists. Financial difficulties, marital problems, and the like became relevant only when they generated or appeared to generate medical symptoms. Second, it was more specific than any clergy definition. There was a syndrome of general nervousness, with signs and symptoms, with causes and mechanisms. If the concept lacked the specific, detailed understanding that came later, neither was it simply a general category of personal problems . . . . Finally, the neurologist had concrete therapies – the rest cure, electrotherapy, psychotherapy – that attacked this problem in particular ways . . . . [D]isciplines with specific therapies will generally defeat those with vague therapies when results are equal. Thus, the neurologists, who in fact accomplished little more for general nervousness than did the clergymen, won the day (Abbott, 1988, p. 290).
Of course, neurology would ultimately have to cede jurisdiction over personal problems to other professions, most notably psychology. The reason was that neurology’s theory for linking problem to solution, while well defined in comparison to that of the clergy, still lacked clarity and parsimony. Because neurologists’ patients had relatively few symptoms, the neurologists’ system of diagnosis was necessarily subtle – appearing extremely idiosyncratic in practice. The subtlety and complexity of the theory made neurology extremely susceptible to a loss of jurisdiction to other professions. Abbott writes, The complexity of [neurology’s system of diagnosis] made neurology as much a hermeneutic art as a natural science. As a result, the neurological jurisdiction proved a fertile ground for interprofessional poaching (Abbott, 1988, p. 289).
Freudianism – and other similar paradigms – that arose in the field of psychology had at least the appearance of a much more rigorous methodology for diagnosis, treatment, and inference. The problem, of course, with psychology – or more specifically psychotherapy – was (and is) the lack of clarity around the problem has been solved. At various points, Abbott emphasizes the importance of relatively straightforward measures of success. As alluded to above, the most successful professions seek to simplify the definition of success in a way that is amenable to the techniques of the profession. The legal profession in the U.S. redefines the adversarial system as justice (i.e. a success) rather than a means to justice. The medical profession provides an even better example. As Abbott writes, “While direct medical interventions in disease contribute relatively little to the net health of modern societies, the medical profession has until recently successfully defined health as ‘not needing a doctor”’ (Abbott, 1988, p. 186).
Quality, Exchange, and Knightian Uncertainty
93
The careful attention to the environment, nutrition, and general health that had characterized mid-nineteenth century medicine disappeared (Abbott, 1988, p. 190).
The implication is clear: a lack of formal clarity and parsimony in the abstract theory underlying how the profession solves the problem makes the profession susceptible to “interprofessional poaching.” If we frame this discussion of professions using the language of quality schemas, we can think of the abstract theory underlying each profession’s activities as denoting a particular quality schema. When a profession has jurisdiction over a particular problem area, quality comes to be defined in terms of activity that at least has the appearance of conforming to the abstract theory. Efficaciousness helps, but as Abbott’s discussion of the personal problems area illustrates, efficaciousness is neither a necessary nor a sufficient condition for being perceived as offering a high-quality solution. To the extent that medical doctors have jurisdiction over health, acupuncturists, chiropractors, and therapists are all of lower quality in so far as they do not follow the same legitimated techniques of diagnosis, treatment, and inference. Moreover, to the extent that a doctor wishes to be perceived as high quality, he or she must conform to the abstract theory. Suppose, for example, that a heart surgeon decides that he will not perform a coronary bypass operation without first consulting with his or her patient about how a change in diet could lower the level of cholesterol and hence need for a bypass. Suppose furthermore that the surgeon compiles a cookbook of good tasting but low-cholesterol foods to help his patients avoid the need for a bypass operation. On some objective grounds, one could clearly argue that the surgeon is providing higher quality service by giving patients the option of avoiding a painful and potentially dangerous surgical procedure. However, the surgeon’s colleagues and perhaps some of his patients are likely to perceive the surgeon as “strange” or “weird” for offering a cookbook solution to a surgical problem. As in the market context, consumers’ reliance on a simplified quality schema for solutions to a particular problem area facilitates the seller’s (i.e. profession’s) ability to cultivate and consummate exchanges. If recipients of a profession’s services perceive its quality schema as too elaborate, then the profession runs the risk of being usurped by another occupation offering a substitute service with a more formal, parsimonious linking of a solution to the general problem – even if a redefinition of the solution is an important contributing element of the formalism and parsimony. It is important to note that, unlike markets and art worlds, among professions there exists a fundamental divergence in the quality schemas used by producers and recipients of professional services. Internally, a profession is likely to develop a complex quality schema. The profession of medicine, for example, houses a
94
JOEL M. PODOLNY AND GRETA HSU
clear status differentiation among its various members. A doctor’s educational background, specialization, participation in academic research, and institutional affiliations, for example, all contribute to how he or she is regarded relative to others within the medical profession. Among clients, however, such distinctions often go unnoticed. Professions attempt to minimize external perceptions of internal differentiation because the appearance of homogeneity increases a profession’s claim to comprehensive jurisdiction. The presentation of a simplified quality schema (made possible in large part by the exclusion of non-producers from the determination of quality) thus allows professions to defend their jurisdictional claims. Art worlds provide a noteworthy contrast here. In art worlds, both the creator of art and the recipient/receiver of art delight in the complexity of the schema and in the development of a special language for discussing distinctions between different artistic objects. In contrast, in the domain of professions, the recipients of professional services are not supposed to speak the language of the professions. The link to the experimental work on Knightian uncertainty should be clear. In a world in which individuals are uncertainty averse, individuals will be more likely to accept the offers of professions that can provide them with some clear probability distribution for the outcomes that can accrue from entering into the exchange with the profession. Individuals will even be willing to accept a redefinition of the state space for outcomes if they are promised some basis for assessing the link between various courses of action and the likelihood of different alternatives within that state-space. Thus, like the market context and unlike the artistic context, it seems reasonable to imagine a rating of members of a profession on some quantitative scale based on some separable set of attributes. However, the difference is that to the degree that a profession has jurisdiction, the members of the profession can legitimately claim the exclusive right to determine the criteria along which they should be ranked. So, one of the clearest signs that an occupation is losing its professional status is that it loses control over its ability to set rankings. A sign that medicine is losing some of its status is that outsiders are now ranking medical professionals on criteria such as “customer satisfaction” and “responsiveness,” and these rankings in turn are impacting on the financial success of those professionals.2
A TYPOLOGY This discussion of the three domains – art, the professions, and the market – suggests two dimensions along which exchange contexts can be differentiated. The first dimension is whether quality is simple or complex. The second dimension is
Quality, Exchange, and Knightian Uncertainty
95
Fig. 1. Quality Schemas and Exchange Domains.
whether non-producers have a voice in defining quality. Given these two dimensions, the three domains can be arrayed in a typology like that in Fig. 1. It is notable that the upper right hand quadrant is not a stable context for exchanges. The recipients of objects or services of value seem to have little affinity for providers who insist on the right to define quality and simultaneously insist that quality is extremely complex. As Abbott’s work implies, buyers will be looking for substitute products or services with a simpler, more tractable formalization of quality. In effect, the nature of competition between professions ensures that the upper right quadrant is not stable. At least viewed historically, recipients seem to make only one exception: religious orders. Though religious orders have had to cede considerable ground to other professions, religious orders have nonetheless remained quite viable given the highly complex theoretical formulations and relatively intangible measures of success. Buyers have perhaps been willing to tolerate such complexity given the promise of “eternal salvation,” but one could at least question how long such a tolerance will last. As Abbott observes, the cultural concern with eternal salvation has declined considerably over the last century. Some other professions may be able to occupy that space for a brief time period. Psychotherapy – due largely to its murky conception of salvation for mental and emotional distress – probably can be coded as belonging in that upper right quadrant. However, unless the goal of emotional and mental stability can acquire the same cultural significance as that possessed by the goal of eternal salvation in years past, it is difficult to see how psychotherapy can preserve its jurisdiction. One of the advantages of the dimensionalization of the three exchange domains in Fig. 1 is that it provides an image for considering the mobility of different activities between domains. Activities clearly shift domains. For example, teaching at the elementary and high school levels has over the last several decades lost its
96
JOEL M. PODOLNY AND GRETA HSU
status as a profession in the U.S. Parents no longer accord teachers jurisdiction over how their children should be educated; through pressure on school boards, ballot propositions, and direct feedback, parents perceive themselves as being in the role of customer and – as a consequence – shift the exchange from the professional domain to the market domain. Notably, as the domain has shifted, there has been increasing interest in easily quantifiable metrics for evaluating the educational attainment of children. Again, we would observe that cause and effect are difficult to disentangle. The championing of ratings and rankings in the educational domain has probably been an important cause in undermining the professional status of teachers; at the same time, the legitimacy of such ratings is enhanced to the degree that teachers are not perceived as being satisfactory arbiters of quality. More generally, the inter-professional competition elucidated by Abbott constitutes a set of mechanisms for the vertical movement between the market and professional domains. For example, a relatively diffuse theory, linking problem to solution, may cause an occupation to shift from profession to purveyor of market good. Or, if a profession over-reaches and tries to apply its theory and technique to too many problems, a similar shift may occur. Conversely, the articulation of a relatively well-defined theory connecting a problem to its solution may enable a profession to raise itself from the market to the professional context. Following Becker, it is clear that the shift from craft to art can be understood at least in part as a shift in the exchange context from the market to the artistic domain. For example, as the value of a quilt derives less from its utility and more from a multi-dimensional conception of its “beauty,” quilting ceases to be a craft and becomes an art form. It is difficult to imagine an activity shifting from professional to artistic or vice-versa. A movement from professional to artistic implies a ceding of control to non-producers in the determination of quality and increasing complexity in the conception of quality. Such an activity is an anathema to an occupation that seeks jurisdictional control. Perhaps the only cases of such movement would be weaker professions that lack considerable jurisdiction and are therefore at the border between the market and professional domains. In Europe, such a transition arguably occurred around the time of the Baroque period, as music was decreasingly used as an accompaniment to a religious liturgy. While we may look back and perceive Bach as a true artist, we are doing so because we assume that Bach lived at the center of an art world. However, such a world unfolded after Bach’s time, and Bach mostly likely perceived himself as a professional for hire. Under our framework, Bach’s profession would be classified as a weak one because the patrons of his compositions were less likely than a doctor’s patients, for example, to unquestioningly accept the profession’s definition of quality. Today the world of computer graphics design seems on the verge of giving rise to a group of artists who
Quality, Exchange, and Knightian Uncertainty
97
are at the center of their own art world, with critics, gallery owners, and museums contributing to the definition of this new art form. But, again, we would assert that computer graphics design has only weak jurisdiction – being unable to impose its own standards of quality – on the consumers of that design work. So, the move from professional to artistic is possible – but only if the profession is relatively weak. Movement from the lower-left quadrant to the upper-right seems equally implausible. It is difficult to conceive of market producers simultaneously excluding non-producers from the definition of quality and legitimately claiming a complex dimensionalization of quality. However, the reverse move seems possible. The market may be the absorbing state for professions that develop a complex schema in their attempt to overreach. One interpretation of the path taken by business schools since the 1960s is that they initially tried to professionalize management by a strong reliance on economics and decision sciences, but they gradually came under criticism for a lack of relevance of the schemas to the problems of day-to-day management. In response, business schools offered “softer” courses in strategic management and especially organizational behavior. While these softer courses were typically more relevant to the daily problems of management, the introduction of these softer courses into the curriculum only furthered the incompleteness of management professionalization. The theory linking management problem to solution now became little more than a cacophony of different intellectual approaches. This cacophony has undermined the legitimacy of business schools as “professional schools” and, as a result, business schools find themselves in a market context facing intense pressure from ratings they do not control. While our observation is admittedly quite cursory, we would argue that business schools have come full circle. They started in the market context; the reliance on economics and the decision sciences raised their teachings to a professional level, but the poor fit of the theory to management problems forced a move to the inherently unstable upper-right quadrant. They have now fallen down to the market domain once again. One more example illustrates how a given exchange relation can shift domains. Consider winemaking in Europe and the U.S. beginning in the 1960s and 1970s. In Europe, the buying and selling of wine in Europe had many characteristics of an art world. Wine critics promulgated rich multi-dimensional schemas to characterize the distinctions between wines. They used adjectives like “dry,” “sweet,” “smooth,” “rich,” “complex,” and “full-bodied.” They used the tastes of fruits as points of comparison. Both producers and consumers appreciated these distinctions. In the U.S. in the 1960s and 1970s, winemaking techniques and the quality of vines had developed to a level where U.S. winemakers – especially those in California – were able to produce wines that possessed many subtle characteristics
98
JOEL M. PODOLNY AND GRETA HSU
that would be appreciated by sophisticated wine drinkers. However, in comparison to European consumers, the overwhelming mass of U.S. consumers was not particularly sensitive to such distinctions. Absent some way of instilling within U.S. consumers a sensitivity to finer distinctions between American wines, it did not make sense for U.S. producers to invest in vines, equipment, or winemaking techniques that would yield the subtler characteristics perceived by more sophisticated drinkers. Moreover, the small number of wine connoisseurs within the U.S. had developed long-standing prejudices against American wines. This community of U.S. connoisseurs regarded American wines as simply inferior to their European counterparts. As a result, even those who might be sensitive to subtle distinctions did not perceive them to be a relevant basis on which to classify American wines. Adams (1973) discusses how the situation of U.S. winemakers became transformed. First came the popularization of varietal labels for American-produced wines. The traditional practice of U.S. wine producers had been to mimic European producers by adopting old generic names of European geographic origin when naming their products. Instead of continuing to imitate their European counterparts, American winemakers increasingly used the variety of grape from which their wines were made to name their wines. Federal regulations required that 51% of the grapes used to make the wine come from the varietal listed on the bottle (Peters, 1997). Initially, in California, the two most prominent varietals were Chardonnay and Cabernet Sauvignon. Consistent with Zuckerman’s work, such a classification scheme helped to create clear reference groups. Varietal labels were not the only nomenclature introduced to characterize wines. In addition, critics introduced uni-dimensional numerical scales for the rating of wines. Robert Parker, a particularly influential wine critic in the U.S. whose writings appeared in the Wine Advocate, was the first to do so; he introduced a 100-point scale to rate wines (Lukacs, 2000). Parker used this numerical scale to, in effect, “rank” wines; the better he perceived the wine to be, the higher would be the numerical rating that he would he would assign to that wine. Wines were grouped by varietal for the purpose of rating because varietals were presumed to be sufficiently different in taste that cross-varietal comparisons were uninformative. So, one could not say that a Chardonnay with a rating of 90 was better than a Cabernet with a rating of 85. Elaborated in this way, the ratings were complements rather than substitutes of the varietal scheme. When first introduced, Parker’s numerical scheme was quite controversial. Other critics regarded the numerical rating scheme as inadequate for conveying the subtle differences between wines. Parker agreed, but he felt that the U.S. consumers would nonetheless find them more valuable for making purchasing decisions than subtler, more complex adjectives. In effect, the premise underlying the initial negative
Quality, Exchange, and Knightian Uncertainty
99
reaction of other critics was that winemaking is an art and a numerical rating scheme is inconsistent with an art. However, in the U.S., there was not a sufficiently broad, invested community to sustain a general appreciation of a rich quality schema. Parker effectively presented the wine community in the U.S. with a market – rather than artistic solution – to the interpretive problem. This market solution worked; it created a tangible basis for convergence of expectations of buyers and sellers. This example is especially noteworthy since it seems to illustrate some of the inherent indeterminacy in which region of Fig. 1 a particular product finds itself.
CONCLUSION We began with a discussion of the problem of Knightian uncertainty underlying exchange. For actors to interact in a productive manner, there must be some agreement within an exchange context regarding the constituent elements of quality. We ended up concluding that the solutions to the problem can be arrayed along two dimensions – the extent to which the conception of quality is of high or low complexity and the extent to which the non-producer in the exchange has a voice in conveying quality. This dimensionalization in turn gives rise to three viable domains of exchange: artistic, professional, and market. Accordingly, how the problem of Knightian uncertainty is solved determines whether the producer of the good is perceived as artist, professional, or simply a market actor. We also observed that mobility across the domains is possible. However, we have no illusions about this chapter; it clearly raises more questions than it answers. Some of the questions that the chapter raises are methodological. If mobility is possible, an important challenge is developing a methodology that enables one to specify the extent of movement. In another paper (Hsu & Podolny, 2002), we develop a methodology for dimensionalizing the quality schemas of film genres. The methodology uses the adjectives employed by film critics in their reviews to develop a characterization of the quality schemas underlying particular genres. The methodology, for example, enables us to assess whether the New York Times critics’ schema for the quality of comedies is more or less complex than the Variety critics’ schema for drama. While this methodology may provide some guidance in this particular context, it is not clear that the methodology is sufficiently general to be of help across different types of our activities. In our view, some of the most interesting questions are substantive and pertain to the causes and consequence of shifts between domains. While we have distinguished the three domains in terms of how the Knightian problem is solved, we
100
JOEL M. PODOLNY AND GRETA HSU
have not given much systematic consideration to the causes of the shifts and how the nature of the solution to the interpretive problem impacts on the fate of those in the particular exchange context. It is certainly possible to make some cursory observations. For example, with respect to causes of movement, it is probably important to understand how much an individual’s activities are constrained by a division of labor. One of the factors that probably pushes car designers into the market – rather than artistic or professional – quadrant is that their designs are significantly constrained by the fact that their designs must respond to the imperatives of others involved in the manufacture of a vehicle. By comparison, clothing designers are relatively unconstrained. As this example illustrates, we would expect that changes in the division of labor can shift the domain of the exchange. Cultural values also probably impact on the position and movement of exchange relations across domains. For example, in the U.S., there is a comparatively strong belief in the superiority of the market as an allocative mechanism; we would therefore expect that many activities coded as “professional” or “artistic” in Europe will simply be seen as market objects or services in the U.S. Again, winemaking represents on example of this difference; filmmaking perhaps represents another. Just as one can make cursory observations about the causes of mobility between domains, so one can make cursory observations about the consequences. For example, the artistic domain is the one in which the producers can most legitimately call on the state for support; this is the only legitimately publicly-subsidized domain. Yet, whereas the art domain can attract state subsidies, we would expect that the market domain is the one that is likely to attract the greatest flow of resources from buyers. The reason is simple: to cite a clich´e, “the customer is king” in the market. The buyer has a strong say in what constitutes quality, and the simplicity of the quality ordering facilitates comparisons that enable a buying decision. In our view, one of the most interesting outcomes of interest is the inequality in the distribution of rewards among producers. We suspect that art worlds are likely to be especially high in variance; that is they are likely to be “winner-take-all” markets. The high complexity of quality implies an inherent inability to routinize the process of creating new works at the highest quality standards. Even the most prolific artists – Picasso, for example – could never provide a codifiable set of rules for artistic creation at the highest quality levels. Because the artistic process cannot be routinized, there will inevitably be considerable limits on the supply of works that conform to the highest standards of quality, and this scarcity will engender a tremendous inequality in the distribution of rewards. Professions – by comparison – will generally be characterized by relatively low inequality in rewards. We believe that there are two reasons that this is the case.
Quality, Exchange, and Knightian Uncertainty
101
First, a relatively homogenous distribution of rewards follows directly from the concept of jurisdiction. The more that the members of a profession jointly adhere to a common set of practices for addressing the needs that arise in a particular problem area, the harder it will be for some subset of that profession to claim that they are deserving of considerably greater rewards than the other members of the profession. A second reason that we would expect a relatively homogenous distribution of rewards within a profession is that variance in rewards can become a signal of quality. That is, if there are considerable differences in the pay within a profession, the recipients of the professionals’ service can use those pay differences as a tangible basis for making inferences about quality. In so doing, they can wrest control of quality determination from the profession. If the reward variance among professions is generally low and the reward variance among artists is generally high, then we expect the reward variance in markets to be generally indeterminate. That is, in markets, the reward variance can be either high or low. The low complexity of the market context implies the possibility for routinizing the creation of objects of a given quality; such routinization should facilitate imitation, which should in turn undercut the variance in rewards, as producers seek to imitate those who provide the highest rewards. However, stable positional differences deriving from differences in identities – such as those emphasized by White (2002) in his W(y) model or by Carroll in his resource partitioning model (Carroll & Swaminathan, 2000) – can thwart such imitation. While cursory speculations such as these are possible, it obviously is important be able to draw some formal conclusions about how movement along the two dimensions impacts on the distribution of rewards going to the producers. Do high-status producers have an interest in some types of shifts but not others? It would also be valuable to be able to draw some formal conclusions about the relative inertness of the three domains. Is one of the three domains more stable than the other two? These are the types of questions that require more systematic analysis. However, our hope is that this framing takes us an initial first step toward the answering of such questions.
NOTES 1. Because we seek to compare the art domain to others, we will use the phrase “quality schema” synonymously with the concept of an aesthetic. 2. Cause and effect probably runs both ways here. As an occupation becomes externally ranked, it starts to appear more like a service provider in the market; conversely, the more it appears like a service provider in the market, the more that it seems appropriate to be externally ranked.
102
JOEL M. PODOLNY AND GRETA HSU
ACKNOWLEDGMENTS The authors wish to thank Bill Barnett, Jim Baron, Michael Hannan, Chip Heath, Jim Phills, John Roberts, and the editors of this volume for helpful comments on earlier drafts of this chapter.
REFERENCES Abbott, A. (1988). The system of professions: An essay on the division of expert labor. Chicago, IL: University of Chicago Press. Abolafia, M. Y. (1997). Making markets: Opportunity and restraint on Wall Street. Cambridge, MA: Harvard University Press. Adams, L. D. (1973). The wines of America. Boston, MA: Houghton Mifflin. Baker, W. E. (1984). The social structure of a national securities market. American Journal of Sociology, 96, 589–625. Becker, H. S. (1982). Art worlds. Berkeley, CA: University of California Press. Becker, S. W., & Brownson, F. O. (1964). What price ambiguity? Or the role of ambiguity in decisionmaking. Journal of Political Economy, 72, 62–73. Buskens, V., Batenburg, R. S., & Weesie, J. (2003). Embedded partner selection in relations between firms. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 107–133). Oxford: Elsevier Science. Camerer, C. (1995). Individual decision making. In: J. H. Kagel & A. E. Roth (Eds), Handbook of Experimental Economics (pp. 587–703). Princeton, NJ: Princeton University Press. Carpenter, G. S., & Nakamoto, K. (1989). Consumer preference formation and pioneering advantage. Journal of Marketing Research, 26, 285–298. Carroll, G. R., & Swaminathan, A. (2000). Why the microbrewery movement? Organizational dynamics of brewing in the U.S. brewing industry. American Journal of Sociology, 106, 715–762. Ellsberg, D. (1961). Risk, ambiguity, and the savage axioms. Quarterly Journal of Economics, 75, 643–669. Favereau, O., Biencourt, O., & Eymard-Duvernay, F. (2001). Where do markets come from? From (quality) conventions! Working Paper, University of Paris X. Granovetter, M. S. (1995). Coase revisited: Business groups in the modern economy. Industrial and Corporate Change, 4, 93–130. Griliches, Z. (1971). Hedonic price indexes of automobiles: An econometric analysis of quality change. In: Z. Griliches (Ed.), Price Indexes and Quality Change (pp. 55–87). Cambridge: Cambridge University Press. Gulati, R., & Wang, L. O. (2003). Size of the pie and share of the pie: Implications of network embeddedness and business relatedness for value creation and value appropriation in joint ventures. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 209–242). Oxford: Elsevier Science. Hsu, G., & Podolny, J. M. (2002). Critiquing the critics: A methodological approach for the comparative evaluation of critical schemas. Unpublished manuscript, Stanford Graduate School of Business. Kauffman, S. A. (1993). The origins of order. Oxford: Oxford University Press. Knight, F. H. (1921). Risk, uncertainty, and profit. Boston, MA: Houghton Mifflin.
Quality, Exchange, and Knightian Uncertainty
103
Kollock, P. (1994). The emergence of exchange structures: An experimental study of uncertainty, commitment, and trust. American Journal of Sociology, 100, 313–345. Levinthal, D. (2001). Modeling adaptation on rugged landscapes. In: A. Lomi & E. R. Larsen (Eds), Dynamics of Organizations (pp. 329–348). Menlo Park, CA: AAAI Press. Lukacs, P. (2000). American vintage: The rise of American wine. Boston, MA: Houghton. McMillan, J. (2002). Reinventing the bazaar: A natural history of markets. New York: W. W. Norton. Peters, G. L. (1997). American winescapes: The cultural landscapes of America’s wine country. Boulder, CO: Westview Press. Podolny, J. M. (1993). A status-based model of market competition. American Journal of Sociology, 98, 829–872. Raub, W., & Weesie, J. (1990). Reputation and efficiency in social interactions: An example of network effects. American Journal of Sociology, 96, 626–654. Rosen, S. (1974). Hedonic prices and implicit markets: Product differentiation in pure competition. Journal of Political Economy, 82, 34–55. Sarin, R. K., & Weber, M. (1993). The effect of ambiguity in a market setting. Management Science, 39, 602–615. Stuart, T. E. (2003). Governing strategic alliances. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 189–208). Oxford: Elsevier Science. Savage, L. J. (1954). The foundations of statistics. New York: Wiley. Stiglitz, J. E. (1987). The causes and consequences of the dependence of quality on price. Journal of Economic Literature, 25, 1–48. Tirole, J. (1996). A theory of collective reputations (with applications to the persistence of corruption and to firm quality). Review of Economic Studies, 63, 1–22. Voss, T. (2003). The rational choice approach to an analysis of intra- and interorganizational governance. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 21–46). Oxford: Elsevier Science. Weber, M. (1978). Economy and society (edited by Guenther Roth and Claus Wittich). Berkeley, CA: University of California Press. White, H. C. (1993). Careers and creativity: Social forces in the arts. Boulder, CO: Westview Press. White, H. C. (2002). Markets from networks: Socioeconomic models of production. Princeton, NJ: Princeton University Press. Zuckerman, E. W. (1999). The categorical imperative: Securities analysts and the legitimacy discount. American Journal of Sociology, 104, 1398–1438.
EMBEDDED PARTNER SELECTION IN RELATIONS BETWEEN FIRMS Vincent Buskens, Ronald S. Batenburg and Jeroen Weesie ABSTRACT This chapter addresses how firms buying information technology (IT) products select their suppliers. We argue that social embeddedness, in the sense of own experiences with suppliers and information about experiences of third parties, influences these types of selection decisions. More specifically, we claim that social embeddedness is more important if: (1) the potential damage for the buyer from receiving an inferior product is larger and (2) if it is more difficult for the buyer to monitor the quality of the product. We use large-scale surveys of IT transactions in the Netherlands and Germany to test these hypotheses. In general, our hypotheses about the effects of social embeddedness on partner selection are supported. We find that buyers tend to assign greater weight to product quality if the potential damage for the buyer is larger. Negative third-party information is particularly important if the buyer has large problems to monitor the quality of a product.
INTRODUCTION Choosing a suitable supplier is often a complex decision for buyers. Basically, a buyer wants to optimize quality, price, and service. An important problem is that The Governance of Relations in Markets and Organizations Research in the Sociology of Organizations, Volume 20, 107–133 Copyright © 2003 by Elsevier Science Ltd. All rights of reproduction in any form reserved ISSN: 0733-558X/PII: S0733558X02200051
107
108
VINCENT BUSKENS, RONALD S. BATENBURG AND JEROEN WEESIE
features such as quality and future service are observable only to a limited extent for the buyer at the moment that he buys a product. In addition, the buyer often has to pay the product before it is actually delivered. Because of these information and time asymmetries, the buyer has to trust the supplier that the product will have the properties claimed by the supplier, that it will be delivered in time, and that the supplier will handle contingencies properly. This implies that the buyer not only needs to find a product that suits his needs with a reasonable quality and for a reasonable price. He also has to find a supplier who offers warranties and service as preferred by the buyer, and the buyer has to trust that the supplier will act as promised. To prevent and cope with problems that might occur, the transaction can be governed in different ways. Contracts can be written that prescribe measures to be taken in case of problems such as late delivery or incompatibilities. However, such contracts are costly to design, implement, and enforce. Costs of writing such contracts can be diminished by choosing a supplier who can be trusted to handle contingencies in a way that is acceptable to the buyer, without the threat of contractual sanctions. Therefore, buyers are willing to invest in searching and selecting a supplier with preferable properties. Rather than studying the extent to which a buyer invests in searching a supplier (as is done, e.g. in Blumberg, 2001), we will examine the buyer’s choice of a supplier from the pool of potential suppliers that resulted from the search process. Within marketing research and business economics, there is a large literature on selection processes in which different suppliers are compared on attributes of the supplier and the relevant products (Verma & Pullman, 1998). In this literature, purchasing is part of the strategic processes, often involving many suppliers and buyers, labeled with corresponding modern terminology such as “supply chain management” (Levy & Grewal, 2000). The growing importance of purchase activities can be related to two major business developments. First, organizations increasingly outsource their activities and, consequently, intensify their management of suppliers. Traditional manufacturers are downsized to assembly firms in many industries. Managing suppliers’ networks to market final products has thus become a new core business for these new organizational forms. Second, demands on quality and price of products have increased. Every firm that is part of a production chain has to meet the quality demands of its buyer. It will pass at least the same demands on to its own suppliers. The rapid expansion of quality standards (such as ISO), “total quality management,” and “just-in-time” concepts in the last decades can be largely explained by this phenomenon (Banker & Khosla, 1995). Because firms explicitly internalize the responsibility of the quality of its suppliers’ products, the importance of purchase management has grown considerably. The importance of the supplier selection process is reflected in the rapidly growing literature on various methods to optimize supplier selection (de Boer et al., 2001).
Embedded Partner Selection in Relations Between Firms
109
In the marketing literature, supplier selection is analyzed predominantly by focusing on supplier and product attributes. These include quality, price, flexibility, and delivery performance (Verma & Pullman, 1998). Meta-analyses of these studies show that perceived product quality is one of the most important supplier selection criteria, followed by price and delivery performance history (Mummalaneni et al., 1996; Weber et al., 1991). In contrast with our study, these results are not based on actual decisions of purchase managers, but on fictitious decisions in experimental studies. These experimental studies are mostly based on the perceptions of managers with regard to supplier attributes rather than on actual choices. As Verma and Pullman (1998) demonstrate, the perceived importance of supplier attributes can differ markedly from the importance in the actual selection process. In a survey among 139 managers in different industries, they show that managers actually choose suppliers largely on the basis of price and delivery performance, whereas the same managers assert that product quality is the most important attribute for these decisions. This gap between self-perception and actual choice behavior indicates that managers know that product quality “should” be the dominant strategy in supplier selection, but in practice price and performance are more important. This discrepancy may be a result of incentives of the managers: the purchase managers themselves are primarily evaluated on the money they save and the timely delivery of products. By analyzing actual choices of purchase managers, we aim at getting closer to the real decision process of the purchase managers in this chapter.1 Verma and Pullman (1998) argue that explaining supplier selection within the context of organizations is a complex process. They acknowledge two important limitations of earlier research. First, they suggest that supplier attributes interact strongly. For instance, once a minimal level of product quality is guaranteed (or does not distinguish between suppliers), price becomes a more important selection criterion. The number and shape of possible interactions between supplier attributes, however, remain hard to capture. Second, they indicate that only a limited number of criteria are considered in their study, and that additional criteria might be relevant. In line with these considerations, we will explicitly pay attention to some interaction effects among product and supplier attributes, and we will analyze some new “sociological” criteria related to the possibilities for the buyer and supplier to trust each other. More specifically, we aim to complement the transaction focus of the marketing literature with the relational argument that suppliers are not only selected on the basis of product and supplier characteristics, but that inter-firm relations between buyers and suppliers play a role in the selection process as well.2 We base our research on the notion that inter-firm relations such as buyersupplier relations are built on trust. We conceive of trust as the extent to which
110
VINCENT BUSKENS, RONALD S. BATENBURG AND JEROEN WEESIE
the buyer is convinced that a supplier is not aiming at making additional profits by misinforming or even misleading the buyer. Because performance criteria such as quality and timely delivery are difficult to observe beforehand by the buyer, he has to trust that the supplier is truly committed to delivering in time and to providing adequate quality. If the buyer trusts the supplier less, more extensive contracts will be written to ensure cooperative behavior of the supplier. Trust is often based on information about the supplier’s past performance and on anticipated future transactions. On the one hand, if a buyer has positive experiences with the supplier or if he knows that other buyers have had positive experiences, he will trust the supplier more than if he has no or even negative information. On the other hand, the supplier will be more committed to handle a transaction well if he expects more transactions with this buyer, or with acquaintances of the buyer, because he runs the risk of losing business if he handles the transaction badly. These two mechanisms can be labeled “learning” and “control,” respectively (see Buskens, 2002; Buskens & Raub, 2002) and will be discussed in more detail in the theory section (cf. Stuart, 2003 in this volume for a related discussion about the selection of partners for strategic alliances). Firms that trust each other more can organize their transactions more effectively (Arrow, 1974), because there is less need for costly ex ante negotiating and contracting (Bradach & Eccles, 1989; Macaulay, 1963; Raub & Weesie, 2000). Consequently, the more trustworthy a supplier, the more attractive he is for the buyer firm. Summarizing, this chapter contributes in different ways to the existing literature. First, the data to be analyzed are related to actual supplier choices. We study partner selection within the context of dyadic purchase relations between buyers and suppliers of information technology (IT) products. Buyers from IT products are asked about characteristics of their selected supplier and the most important alternative suppliers. The data are actually from two surveys, one in the Netherlands and one in Germany (Berger et al., 1999; Rooks et al., 1998). The market for IT products is interesting for this study, because this market was growing and technology was changing rapidly especially in the nineties. Therefore, it was difficult for buyers to be informed about new developments, which aggravated the asymmetry of information between buyer and supplier. Second, besides standard attributes such as price, quality, and service, we include attributes about the relationship between the buyer and the supplier and the embeddedness of the buyer and supplier in a network of business relations. This enables us to show that dyadic and network aspects are important for the selection of the supplier, in addition to the more standard attributes such as price and product quality. Third, we develop and test a series of hypotheses about interaction effects between product characteristics and the social embeddedness of buyer and supplier that are mostly absent in research on supplier selection.
Embedded Partner Selection in Relations Between Firms
111
THEORY AND HYPOTHESES When purchasing a product, the buyer is often not completely informed about the quality and fair price of the product he is going to buy. Under such conditions, the supplier has incentives to behave opportunistically and to sell a product that does not correspond with the arranged quality standards (cf. Williamson, 1985). For example, the supplier might use an older specimen of a part within a PC than the part agreed upon. If the supplier is convinced that the buyer is likely not to notice this modification, he might be tempted to do so. Incentives for such opportunistic behavior will be higher for complex products with many features that cannot be directly observed by the buyer. Risks in buyer-supplier transactions can be mitigated in different ways and at different stages of the transaction. Buyers can carefully compare suppliers and select one who is most likely to perform in an acceptable manner before actually entering the transaction. Formal contracts can be written to explicate agreements on what will happen at what time and what has to be done if contingencies occur. During and after the transaction, various strategies for conflict resolution can be used such as delayed payment or going to court. Sometimes a buyer may trust the supplier not to behave opportunistically without such contracts. Then, he will likely proceed with the transaction without drafting a costly formal contract to prevent opportunistic behavior of the supplier. As Williamson (1985, p. 64) acknowledges, trustworthiness of the supplier is not easily observed before a transaction. We argue that trust of the buyer can be based on the embeddedness of the supplier’s transaction in previous and future dealings with this buyer and other buyers (Raub & Weesie, 1990, 2000). Dyadic embeddedness indicates the extent to which the same buyer and supplier are involved with each other over time. This includes past experiences with each other and expectations about the continuation of the relation in the future (Gulati, 1995; Lyons, 1994; see also Batenburg et al., 2003 in this volume). Buyers can learn about a supplier’s trustworthiness from positive experiences with the supplier in the past. This information gives the buyer an idea about the likelihood that a supplier prevents or handles possible conflicts in a satisfactory way even if explicit ex ante arrangements have not been specified. Doney and Canon (1997) find empirical evidence that trust builds up within a buyer-supplier relationship over time and that trust in the salesperson of the supplier plays a key role in this process. Likewise, Gulati (1995) presents evidence that firms prefer forming alliances with partners with whom they had experience in the past (see also Gulati & Wang, 2003 in this volume for a similar discussion related to the value creation of joint ventures). If a buyer and supplier expect to have business opportunities in the future, the buyer has a good reason to expect the supplier to be trustworthy: the supplier wants to ensure that the buyer will continue to do business with him,
112
VINCENT BUSKENS, RONALD S. BATENBURG AND JEROEN WEESIE
or the supplier wants to prevent that the buyer demands lengthy and detailed contracts in future interactions. The buyer’s expectations in this case correspond with the game-theoretic logic of conditional cooperation in repeated interactions (see, e.g. Kreps, 1990 for an illustration based on so called Trust Games). Network embeddedness indicates the extent to which the two firms are embedded in a network of third parties (Buskens, 2002; Granovetter, 1985; Uzzi, 1996, 1997, 1999). The reasoning underlying effects of network embeddedness parallels that of dyadic embeddedness. Trust increases with positive information a buyer receives from other buyers about a supplier.3 Buyers with negative experiences with a supplier can discourage other buyers in this network to do business with this supplier. Grabher (1993) discussed various roles of networks that can affect the governance structure of buyer-supplier relations. Networks can be used to find alternatives if a supplier does not meet the expectations (“exit”). Networks can also play a role in convincing a supplier that his quality and other attributes should be improved (“voice”) (cf. Blumberg, 2001; Hirschman, 1970). Moreover, the argument for repeated interactions can be generalized for transactions of a supplier with a network of buyers. The stronger the ties between the buyers, the larger the incentives for the supplier are to behave trustworthy (Buskens, 2002). In general, the stronger the opportunities for the buyer to sanction the supplier through his network, the more a buyer will be convinced beforehand that a relation with a supplier can be based on trust, and the less contractual arrangements are necessary to prevent the supplier from opportunistic behavior against this buyer (for a more extensive literature overview and related experimental research on this topic we refer the reader to Buskens & Raub, 2002). The implications of these arguments for buyer-supplier relations are relatively straightforward. In correspondence with our data about choices of suppliers from buyers of IT products, we assume that partner selection is controlled by the buyer, and that the supplier will serve all the buyers choosing him as a supplier.4 The assumption that selection is mainly an issue for the buyer does not seem to be very problematic, since our data show that in the vast majority of IT purchases, the buyer is the partner who takes the initiative. As a consequence, we are primarily concerned with the trust problems for the buyers in relation with the suppliers.5 The starting point is that buyers compare different suppliers on a number of criteria and then choose the most attractive supplier. By the embeddedness arguments, we expect that a supplier is selected not only on the basis of price, quality, service and warranties offered by the supplier, and how easily a supplier can be reached (accessibility), but also on the basis of own earlier experiences with a supplier or information about earlier experiences of other buyers with a supplier. Our data do not allow to test whether partner selection depends on expected future transactions with the different potential partners.
Embedded Partner Selection in Relations Between Firms
113
We now translate our arguments into a number of testable hypotheses. The first hypothesis concerns the attributes of the product and the supplier that are considered the most important in the marketing literature (Turban et al., 2001; Verma & Pullman, 1998). Hypothesis 1. A supplier is more attractive, the lower the price, the better the product quality, service, and warranties offered, and the better the supplier’s accessibility. Clearly, these product and supplier attributes are not independent. If differences in quality are small among suppliers, for example, because all suppliers are distributors of the same product, price might become a more important selection criterion compared to the situation that quality varies a lot among suppliers. Because we do not have a well-elaborated theory about possible interaction effects among the product and supplier attributes mentioned in Hypothesis 1, we will restrict our analyses in the following section to tests of non-linearities of effects and possible interaction effects to see whether a model including such effects fits the data considerably better than a model without non-linear and interaction effects. However, we will formulate some hypotheses about interactions between transaction and buyer characteristics with the attributes mentioned in Hypothesis 1 above. We argued that dyadic embeddedness is relevant for the buyer to assess the performance of the supplier in the past. The more positive experiences the buyer has with the supplier, the more attractive a supplier will be. Clearly, the effect of negative experiences is in the opposite direction. Hypothesis 2. A supplier is more (less) attractive, if the buyer has had positive (negative) own experiences with the supplier. Similarly, we expect a positive effect on the attractiveness of the supplier from positive information a buyer receives from other buyers and a negative effect of negative information. Hypothesis 3. A supplier is more (less) attractive, if the buyer has more positive (negative) information about the supplier’s behavior in the past from other buyers of this supplier. One problem with Hypotheses 2 and 3 is that a buyer’s own experiences with and third-party information about a supplier might be incorporated in his estimate for the product quality and the service a supplier offers. We acknowledge that there are common dimensions in these selection criteria. However, there are more aspects and contingencies to be handled by the supplier in a transaction than product quality and service. In addition, there are dimensions related to product
114
VINCENT BUSKENS, RONALD S. BATENBURG AND JEROEN WEESIE
quality and service offered that are not included in own experiences with a supplier and third-party information about a supplier. For example, a certain brand of PCs carries a quality reputation that will be included in product quality and this is not related to the specific supplier of the product. Therefore, we expect own experiences and third-party information to have an effect even after controlling for product quality and service offered, although the effect might be an underestimate of the actual effect. It thus makes sense to test Hypotheses 1, 2, and 3 simultaneously. In addition to these main effects, we expect that the size of the effects of the different selection criteria depends on characteristics of the product to be bought. The problem potential of a product, that is, the size and probability that problems will occur, depends on the importance of the product for the buyer, the potential losses for the buyer if the product turns out to be of inferior quality, the complexity of the product, and the buyer’s expertise in buying such product. In terms of variables for which we have measurements in the data, we distinguish two main dimensions of problem potential, namely, damage potential and monitoring problems. Damage potential includes the importance of the product and the potential losses for the buyer, while monitoring problems combine the complexity of the product and the expertise of the buyer. The larger the damage potential and monitoring problems, the more the buyer wants to pay for indications that the product is satisfactory, for additional safeguards, and the further the buyer wants to travel to obtain these advantages. This implies that if the problem potential is larger, product quality, service, and warranties will be weighted more, while price and accessibility will be weighted less in evaluating the attractiveness of suppliers. In addition, buying standard products involves less trust, which implies that dyadic and network embeddedness are less relevant for standard products compared to complex products. Thus, own information as well as third-party information about past performance of the supplier are more important for more complex problems. More generally, Buskens (2002) derived a range of hypotheses indicating that the importance of embeddedness increases with the problem potential of the transaction. If the problem potential of a transaction is small, a buyer can “solve” this problem at the spot. For example, he could simply ask some suppliers about their options to fulfill his needs, and would choose the one that appears most knowledgeable and trustworthy. The larger the problem potential becomes, the more this buyer needs to lean on his own experiences in the past and his network to overcome the trust problem. This implies that the effects of own information and third-party information are expected to increase with damage potential and with monitoring problems. In sum, these arguments lead to the following hypotheses about interaction effects between product characteristics and the attributes that should determine the choice of a supplier.
Embedded Partner Selection in Relations Between Firms
115
Hypothesis 4. The effects of price and accessibility on the attractiveness of the supplier decrease with damage potential and monitoring problems. Hypothesis 5. The effects of product quality, service, and warranties on the attractiveness of the supplier increase with damage potential and monitoring problems. Hypothesis 6. The effects of own information and third-party information on the attractiveness of the supplier increase with damage potential and monitoring problems. If a buyer finds it more difficult to monitor the quality of the product, he is less likely to select a supplier only on the basis of his own experiences. The buyer might have had some problems with a product he bought from some supplier, but if his expertise is small, he will not be able to discern whether or not this supplier had any possibility to foresee the problem or prevent it from occurring. Similarly, a buyer might have had positive experiences with a supplier in less complex transactions, and would still be unsure whether these experiences are informative about a more complex transaction he is currently planning with that same supplier. Therefore, third-party information is more relevant than his own experiences in transactions in which the buyer has more monitoring problems. This leads to the following hypothesis: Hypothesis 7. The positive interaction effect of third-party information and monitoring problems on the attractiveness of the supplier is larger than the positive interaction effect of own experiences and monitoring problems.
DATA AND METHODS Data We use two data sets on IT purchases by small and medium sized enterprises (SMEs). Respondents were managers or employees of the buyer firms. They were interviewed in the Netherlands in 1998 (Rooks et al., 1998; we use the Dutch acronym “MAT” to refer to the related data set) and in Germany in 1999 (Berger et al., 1999; for this data set we use the German acronym “EDV”). Nearly identical questionnaires were used to survey the management of a particular IT transaction between the buying firm and an IT supplier. The questionnaires focused on a recent significant investment in hardware or software. Being designed as multi-purpose surveys on the management of inter-firm relations, the questionnaires covered the product, the search and selection of a supplier
116
VINCENT BUSKENS, RONALD S. BATENBURG AND JEROEN WEESIE
of IT products, contractual negotiations for the transaction, the execution of the transaction, and post-transaction experiences. The Dutch firms were selected from a previous survey with a similar questionnaire (Batenburg & Raub, 1995). From the 971 firms in the 1995 survey, 664 were reached again in 1998. They were asked to complete the new 1998 questionnaire with more extensive questions on partner selection in relation to a new recent IT purchase. From the 281 firms that agreed to participate again, 136 firms selected their IT supplier from two or more alternatives. No alternative supplier was mentioned for the other 145 Dutch transactions. Some cases had to be excluded from our analyses due to missing values. Complete data are available for 52, 43, and 23 transactions with 2, 3, and 4 potential suppliers, respectively. A new sample of 832 firms participated in the German survey. From these firms, 187 completed the questionnaire for two transactions resulting in a data set of 1019 IT transactions in total. EDV contains complete data about 73 and 405 transactions with 2 and 3 potential suppliers, respectively. For both data sets, the statistics for which complete data are available are described in the remainder of this section, and are used to test our hypotheses in the next section. We refer to the codebooks for a more detailed description and comparison of the two data sources (Batenburg & van de Rijt, 1999; Berger et al., 2000; Buskens & Batenburg, 2000). Information about supplier selection for the IT purchases was obtained with a classical decision matrix. For a number of selection criteria, the buyer had to rate the suppliers using grades commonly used to rate performance in schools; original questions are presented in the appendix. The following selection criteria (attributes) are distinguished:
price of the product as offered by the potential suppliers, quality of the product as offered/presented by the potential suppliers, service offered by potential suppliers, proximity/accessibility of potential suppliers, and warranties provided by potential suppliers.
Embeddedness was measured in somewhat different ways in the Dutch and German questionnaires. Dutch respondents rank-ordered the potential IT suppliers including the selected supplier with regard to: the amount of experience of the buyer with each of the suppliers, positivity of own experience with each of the suppliers, the amount of information about experience of others with each of the suppliers, and positivity of information about experience of others with each of the suppliers.
Embedded Partner Selection in Relations Between Firms
117
Table 1. Descriptive Statistics. Netherlands (MAT) Attributes price quality accessibility service warranty positive own experience negative own experience positive third-party information negative third-party information Number of buyers Number of transactions Number of potential suppliers
Germany (EDV)
Mean
Std. dev.
Mean
Std. dev.
6.90 7.37 6.97 7.11 7.21
1.23 1.20 1.31 1.09 1.05
7.26 7.70 7.37 7.38 7.50 0.37 0.11 0.27 0.10
1.28 0.97 1.30 1.20 1.11 0.48 0.32 0.44 0.30
Mean and standard deviations are not informative 118 118 325
439 478 1,361
German respondents indicated whether or not they possessed information about the potential IT suppliers with regard to:
positive own experience with each of the suppliers, negative own experience with each of the suppliers, information about positive experience of others with each of the suppliers, and information about negative experience of others with each of the suppliers.
Table 1 shows the absolute rating and rankings of suppliers on these criteria. In the Netherlands, grades on a scale from 1 to 10 are commonly used to rate performance in education, 10 being best. Germans use a scale from 1 to 6 in reversed ordering for the same purpose. For comparison with the Dutch data, the German grades are mapped into the Dutch grades: 1 to 9, 2 to 8, 3 to 7, 4 to 6, 5 to 4, and 6 to 2. This mapping satisfies the property that the minimal grade to “pass” in Germany (4) is mapped into the equivalent Dutch grade (6). While the details about such a mapping are always debatable, the results of the analyses reported in the next section do not substantively differ between a number of reasonable mappings – and in fact do not differ much from the reversed German scale. In general, suppliers received rather positive grades (7 is equal to “satisfactory”). The German grades are somewhat higher than the Dutch grades. We do not present summary statistics about the rankings for the embeddedness criteria in MAT, because summary statistics about rankings of different sets of alternatives are not informative. The summary statistics for EDV indicate that buyers have positive own experiences with 37% of the potential suppliers, and
118
VINCENT BUSKENS, RONALD S. BATENBURG AND JEROEN WEESIE
have information about positive experiences of others with 27% of the suppliers. Negative experience and negative information was much less prevalent. Buyers had negative own experiences with 11% of the potential suppliers and had information about negative experiences by others for 10% of the potential suppliers. Next to the data on the alternatives of the buyer to select a supplier, we constructed additional variables to test the hypotheses on interaction effects between decision criteria and characteristics of the buyer and the transaction. The variables “damage potential” and “monitoring problems” were constructed from item sets that specify the importance of the product for the buyer, the potential losses for the buyer if the product had defects, the complexity of the product, and the expertise of the buyer to determine early whether the supplier defaults. For details of these constructs we refer to Batenburg et al. (2003) in this volume. Since we have data from different countries, it is definitely tempting to speculate (theorize) about international differences in IT purchase. Such differences in behavior may be due to institutional differences (e.g. in the costs of litigation or competitiveness of subsectors of the IT section), cultural differences (e.g. intercultural differences in risk aversion as argued by Hofstede, 2001), or differences in the effects of social networks (e.g. V¨olker, 1995). We had to exercise self-constraint here for two reasons. First, since we compare only two countries, it will be impossible to attribute differences in IT purchase to any specific difference of the countries. Second, as discussed above, there were differences in the measurement instruments between the surveys, especially for the parts of the survey used in this chapter. Thus, differences between measurement instruments and from differences between the countries are confounded. As a consequence, we decided to analyze the data from the two countries separately and to make only qualitative comparisons between the Dutch and German results. Methods To analyze the choice data, we use the random utility approach, also known as choice-based conjoint analysis (see Louviere et al., 2000 for a recent survey). A buyer prefers a supplier with attributes z1 to a supplier with attributes z2 if he assigns higher “utility” to the first supplier than to the second one, u(z 1 ) > u(z 2 ). It is assumed that the attractiveness for a buyer of a supplier is linear in attributes of the supplier. A random additive residual term is included representing omitted attributes and decision noise: u(z) =  z + where  denotes the buyer’s “decision weights” for the attributes of the product and the supplier. By making assumptions about the distribution of the
Embedded Partner Selection in Relations Between Firms
119
random residuals, we can evaluate the probability that a buyer prefers supplier 1 to supplier 2. A mathematically convenient assumption is that the residuals s are independent and identically Gumbell distributed, leading to a closed form expression for the choice probabilities
Pr(u(z 1 ) > u(z 2 )) = Pr( z 1 + 1 >  z 2 + 2 ) =
e z 1
, e z 1 + e  z 2 also known as the logistic regression model. Luce (1959) extended the analysis to more than two alternatives and showed that
Pr(u(z i ) > maxj=i u(z j )) =
e z i
. e z 1 + e z 2 + · · · + ez m It is important to realize that choices only depend on the differences between the attributes of the suppliers. Attributes that do not differ between the alternatives cannot affect choice. In particular, attributes of the buyer and attributes of the transaction such as damage potential and monitoring problems cannot directly affect choice – constant variables can never explain differences. Still, it is possible to include and estimate effects of interactions between buyer or product attributes and supplier attributes. An example is the interaction between damage potential and the quality of the product (see Hypothesis 5). Maximum likelihood estimators for this model can be obtained using standard software for the conditional logistic regression model. Because some German buyers were involved in multiple transactions, the statistical analyses should take into account that observations are clustered within buyers. Moreover, some suppliers were involved in multiple transactions. However, our data do not allow identifying suppliers accurately, especially because we have no information about the identity of suppliers that were not selected for a transaction. Therefore, we take into account only clustering of transactions within buyers. We used Stata because it provides a particularly useful feature to account for clustering, namely, the robust but somewhat inefficient parameter estimates from “independent” conditional logit analyses with robust (Huber, sandwich-style) standard errors modified for clustering (Huber, 1967; Rogers, 1993).
RESULTS Product and Supplier Attributes Table 2 presents the estimates of the decision weights in an analysis of buyers’ partner selection based only on attributes of the supplier and of the product. As
120
VINCENT BUSKENS, RONALD S. BATENBURG AND JEROEN WEESIE
Table 2. The Decision Weights of Selection Criteria for Suppliers. Netherlands (MAT) Attributes
Hypothesis
price quality accessibility service warranties
+ + + + +
Model test (Wald) Number of buyers Number of transactions Number of potential suppliers
Germany (EDV)
Coeff.
Std. err.
Coeff.
Std. err.
1.221∗∗∗ 1.548∗∗∗ 0.643∗∗ 0.900∗ 0.957
0.285 0.398 0.218 0.402 0.745
1.336∗∗∗ 1.347∗∗∗ 0.495∗∗∗ 0.601∗∗∗ 0.146
0.156 0.212 0.126 0.152 0.165
2 (5) = 42.19∗∗∗ 118 118 325
2 (5) = 114.97∗∗∗ 439 478 1,361
∗∗∗ , ∗∗ , and ∗ represent two-sided significance at respectively p < 0.001, p < 0.01, and p < 0.05 based on Huber standard errors modified for clustering within buyers.
hypothesized, suppliers are more attractive if they are seen as offering a better price (i.e. a lower price), a better quality, better accessibility, and better service. The partial effect of warranties, while positive as expected, is not even close to being significant in either of the countries. The results of these analyses are in accordance with the earlier findings of Verma and Pullman (1998). This is reassuring given the retrospective nature of our data. It also indicates that the methodological approach via the random utility model for supplier selection seems to apply reasonably well. Since all attributes are measured as grades, a comparison of the coefficients is sensible. Both in MAT and EDV, price and quality are seen as more important attributes than the accessibility and service offered by of the supplier, meaning that a difference of one grade point in price and quality affects the buyers’ choice more than a one grade difference in the other attributes. Assuming that the unexplained variance for the attractiveness of Dutch and German IT suppliers is equal we can compare the coefficients of the two analyses.6 The differences between the Dutch and German models are not significant (Wald 2 = 2.42, df = 5, p = 0.78). In both countries, the effect of warranties becomes significant if we do not control for service.7 Social Embeddedness and Partner Selection We now proceed with the more “sociological” aspects of partner selection and focus on how social embeddedness affects the attractiveness of suppliers to buyers.
Embedded Partner Selection in Relations Between Firms
121
Because the measurements of social embeddedness differ between the Dutch and German questionnaires (see the appendix), we have to estimate somewhat different models. Recall that embeddedness in MAT was measured by the rankings of the suppliers in terms of the positivity and amount of own experiences and third-party information. This was, with hindsight, not the ideal method of measurement. In the analysis of MAT, we proceeded by treating the variables representing the ranking of suppliers as interval level variables ranging from 1 (worst) to 4 (best), because in the random utility model, the attributes are assumed to have been measured as metric interval-level variables. Obviously, it is not fully appropriate to treat ranking information this way, but since in the choice model only differences between the attributes of alternatives (suppliers) are relevant, we feel that the problem may not be quite severe. Our Hypotheses 2 and 3 concern the effects of positive and negative experience and information on the attractiveness of suppliers. MAT concerns 4point items only on the “positivity” and the “amount” of experience and thirdparty information. If the first dimension would have allowed negative experience and information as well, it would have been appropriate to test our hypotheses via the product of these (centered) dimensions. Since this is not the case, these measurements do not allow a full test of our Hypotheses 2 and 3. We chose to test simply the “approximate” hypotheses that the positivity of own experiences with a supplier and the positivity of third-party information with that supplier increase attractiveness to the buyer. The operationalizations of the social embeddedness variables for EDV are less problematic, because 0–1 ratings rather than rankings are used to link own experience and information of third parties to potential suppliers. Moreover, direct measurements of positive and negative information are available, again both based on own experiences and on information from third parties. Thus, EDV allows for a better test of the embeddedness arguments as formulated in Hypotheses 2 and 3. Table 3 shows results of the extended model. The estimates of the decision weights of the product attributes are relatively stable upon the addition of embeddedness attributes. Most of our hypotheses about the effects of embeddedness on partner selection are strongly supported. According to the analysis of MAT, the positivity of own experiences and the positivity of third-party information indeed have positive effects on the attractiveness of suppliers, and hence on the probability that they are selected by the buyer. The directions of the effects of social embeddedness in EDV are clearly as expected, while three of the effects are highly significant.8 An intuitively plausible finding is that positive information from own experience has a larger effect than positive information from third parties. This corresponds with findings in a vignette experiment on buying used cars by Buskens and Weesie (2000). We have to exercise caution with putting
122
VINCENT BUSKENS, RONALD S. BATENBURG AND JEROEN WEESIE
Table 3. The Decision Weights of Selection Criteria for Suppliers, Including Criteria of Social Embeddedness Linking Buyer and Supplier. Netherlands (MAT) Attributes price quality accessibility service warranties positivity own experience positivity third-party information positive own experience negative own experience positive third-party information negative third-party information Model test (Wald) Number of buyers Number of transactions Number of potential suppliers
Germany (EDV)
Hyp.
Coeff.
Std. err.
Coeff.
Std. err.
+ + + + + + + + – + –
1.276∗∗∗ 1.168∗∗∗ 0.728∗∗ 0.706+ 0.836 0.460+ 0.709∗
0.386 0.295 0.251 0.422 0.764 0.280 0.321
1.355∗∗∗ 1.179∗∗∗ 0.400∗∗ 0.373∗ 0.174
0.194 0.219 0.130 0.162 0.194
1.811∗∗∗ −0.769 0.736∗∗ −1.510∗
0.254 0.518 0.260 0.687
2 (7) = 38.73∗∗∗ 93 93 250
2 (9) = 137.37∗∗∗ 439 478 1,361
∗∗∗ , ∗∗ , ∗ , and + represent two-sided significance at respectively p < 0.001, p < 0.01, p < 0.05, and p < 0.10 based on Huber standard errors modified for clustering within buyers.
too much weight on the comparison of effect sizes however, since we are only comparing dummy variables regarding whether or not information was available, and not the “amount of information.” A more counterintuitive finding is that negative information from third parties seems to have a larger effect than negative information from own experiences (although this difference is not statistically significant). The effects for negative information are less significant than the effects for positive information, which may partly be due to the fact that there are relatively few cases in which respondents actually have negative information. Moreover, by removing one outlier,9 both effects for negative information become significant and the effect for negative third-party information becomes significant even at the 1% level.
Interaction Effects The analyses so far represent the assumption that all buyers make supplier selection decisions in the same way, that is, they all assign the same weights to the
Embedded Partner Selection in Relations Between Firms
123
attributes of the product and the supplier. We hypothesized that decision weights might vary with damage potential and monitoring problems. In transactions with a larger damage potential to the buyer, we expect buyers to give higher weights to the quality of the product, service, and warranties offered by the suppliers relative to the price and accessibility of the suppliers (Hypotheses 4 and 5). We also formulated the expectation that with increasing damage potential, buyers base their decisions more on their embeddedness with the suppliers (Hypothesis 6). Finally, we hypothesized that with increasing problems to monitor themselves the performance of an IT product and supplier, we expect that buyers assign more weight to the experiences of third parties relative to the own experiences with that supplier (Hypothesis 7).10 To test these hypotheses, we include interactions between damage potential and monitoring problems with the attributes of suppliers and of social embeddedness. In Table 4, we only present the results for EDV due to the larger number of cases and the better measurement of social embeddedness of the relation between buyer and supplier in comparison to MAT. As shown by the overall tests for the interactions, there are significant differences between buyers and products in how suppliers are selected, both in terms of damage potential and in terms of monitoring problems. Neither the overall test statistic, nor the coefficients for one set of interactions depend strongly on whether or not we controlled for the other set of interactions. We thus focus exclusively on the model with both sets of interactions. Among the interactions with damage potential, only quality of the product has a (two-sided) significant effect. With increasing damage potential, buyers give a higher weight to the quality attribute. The weights of the other four product attributes and the embeddedness attributes do not vary with damage potential, but as a result of the increased weight for quality, they become less important relative to quality with increasing damage potential. With increasing monitoring problems, all significant interactions with product attributes indicate that these attributes are weighted less in relation to the attractiveness of the supplier, although Hypothesis 5 predicted larger weights for quality, service, and warranties. Especially, price and quality are weighted less, also compared to the other attributes. As expected, own experiences become less important with increasing monitoring problems. Finally, information about negative experiences from others becomes, as expected, particularly more important with increasing monitoring problems. The sizes of these interaction effects are hardly affected if the non-significant interaction variables are omitted from the model, though they become somewhat more significant. In conclusion, we find some support for our hypotheses on interaction effects, but many predicted effects are not significant and some are even significant in the opposite direction, especially the interaction between quality and monitoring problems.
124
VINCENT BUSKENS, RONALD S. BATENBURG AND JEROEN WEESIE
Table 4. Analysis of Choice of Supplier with Buyer-Product Interactions (EDV Only). Attributes
Hyp.
Coeff.
Std. err.
Coeff.
Std. err.
Coeff.
Std. err.
price quality accessibility service warranties positive own experience negative own experience pos. third-party info. neg. third-party info.
+ + + + + + + + +
1.566∗∗∗ 1.364∗∗∗ 0.568∗∗∗ 0.447∗ 0.111 1.877∗∗∗ −1.035 0.601∗ −2.470∗∗
0.216 0.238 0.175 0.188 0.217 0.263 0.648 0.271 0.864
1.566∗∗∗ 1.364∗∗∗ 0.568∗∗ 0.448∗∗ 0.112 1.877∗∗∗ −1.035 0.602∗∗ −2.468∗∗
0.216 0.238 0.175 0.188 0.217 0.263 0.648 0.271 0.864
1.567∗∗∗ 1.404∗∗∗ 0.581∗∗ 0.541∗∗ 0.077 2.021∗∗∗ −0.876 0.629∗∗ −2.245∗∗
0.237 0.258 0.195 0.209 0.233 0.289 0.662 0.272 0.790
damage potential × price × quality × accessibility × service × warranties × pos. own experience × neg. own experience × pos. third-party info. × neg. third-party info.
– + – + + + – + –
−0.212∗ 0.222∗ 0.032 0.086 −0.038 0.106 0.146 −0.022 −0.211
0.103 0.112 0.064 0.081 0.092 0.141 0.259 0.151 0.299
−0.100 0.341∗∗ 0.060 0.097 −0.123 0.200 0.198 −0.049 −0.328
0.111 0.131 0.099 0.096 0.087 0.171 0.313 0.153 0.392
monitoring problems × price × quality × accessibility × service × warranties × pos. experience × neg. experience × pos. third-party info. × neg. third-party info.
– + – + + + – + –
−0.403∗∗∗ −0.481∗∗∗ −0.179+ −0.089 0.218 −0.359∗ −0.364 0.158 −1.321∗∗
0.105 0.129 0.106 0.153 0.175 0.186 0.346 0.191 0.443
Test against null model Test against main effects Number of buyers Number of transactions Number of potential suppliers
−0.443∗∗∗ −0.343∗∗ −0.177+ −0.010 0.106 −0.318+ −0.152 0.154 −1.736∗∗∗ 2 (18) = 151.47∗∗∗ 2 (9) = 39.41∗∗∗ 439 478 1,361
0.102 0.139 0.095 0.139 0.165 0.168 0.295 0.185 0.467
2 (18) = 152.17∗∗∗ 2 (9) = 30.21∗∗∗ 439 478 1,361
2 (27) = 169.87∗∗∗ 2 (18) = 80.27∗∗∗ 439 478 1,361
∗∗∗ ∗∗ ∗ , , , and + represent two-sided significance at respectively p < 0.001, p < 0.01, p < 0.05, and p < 0.10 based on Huber standard errors modified for clustering within buyers.
Embedded Partner Selection in Relations Between Firms
125
Specification Analyses In this subsection, we investigate some problematic aspects of the analyses presented above. These problems mostly refer to alleged limitations of our data. First, the data on IT transactions were collected retrospectively. Ten percent of the transactions occurred more than three years before the date of interview. We are suspicious that with increasing lapse of time, recollections are fading and data become less reliable. If this is true, the scale of the random terms in the attractiveness model would increase with the time span, that is, the time between the date of supplier selection and the date of interview. The estimated coefficients in the random utility model are the ratio of the decision weights and the scale of the residual (see Allison, 1999). Thus, the argument about recollections can be studied by including an interaction of the attributes with the time span in the model. These interactions are not significant, and hence this potential problem with retrospective data does not seem to affect our results. Second, a property of the statistical model used is that the decision weights should not vary between the number of potential suppliers a buyer could choose from. This is strongly related to what is known in the language of discrete choice modeling as the assumption of the “Independence of Irrelevant Alternatives.” This property may be violated if, for instance, buyers make increasing errors when describing their evaluations of increasingly remote alternatives. Using a generalized Hausman test (Weesie, 1999) comparing the analysis for all observations with the analyses for observations with two, three, or four alternative suppliers separately, we found neither in MAT nor in EDV, significant differences between the decision weights. Third, we were concerned that the linear specification of the effects of grades in terms of price, quality etc. on the attractiveness of suppliers would fail to capture the potential “discontinuity” in moving from 5 (just fail) to 6 (just pass). We estimated various specifications with different “slopes” below 5 and above 6 and with a “jump” from 5 to 6. We did indeed find evidence for “discontinuities” in all attributes. But, since these more complicated models did not tell a different substantive story about the selection of suppliers – as opposed to a more complicated story about measurement – we have presented the simpler model in the text. Fourth, we addressed how well the model fits the data. Omnibus goodness of fit tests for discrete choice models (e.g. Windmeijer, 1994) are generally not very powerful against meaningful forms of misspecification. An alternative strategy embeds the selected model into a larger model and uses standard tests for nested models to test for misspecification. An example is a quadratic extension in which squares and interactions of all variables in the model are added. This analysis did not yield important new insights in supplier selection. Reflecting the non-linearities
126
VINCENT BUSKENS, RONALD S. BATENBURG AND JEROEN WEESIE
of grades already discussed above, some quadratic terms were significant, but they do not affect our substantive conclusions. The only interaction for which we found a significant effect (p < 0.01) in both data sets is a positive interaction between price and warranties. Since we did not find a large main effect of warranties, this interaction could indicate that buyers care about warranties only if suppliers hardly differ by the price of the product they offer. Finally, we were concerned that the effects of the embeddedness variables are underestimated, because quality and service would already explain part of the effects. The obvious argument is that the buyers’ expectations with respect to quality and service are formed on the basis of social embeddedness, own experiences and information about experiences of known third parties, possibly next to the broader reputation of a brand or a supplier. Several analyses, however, showed that these attributes were not highly interrelated. As we can see from Table 3, the effects of quality and service do not change dramatically if the embeddedness variables are added. Moreover, the effects of the embeddedness variables hardly change if we do not control for quality or service, although the effect of negative own information becomes significant in these analyses. The reasoning behind the interrelation between quality, service, and the embeddedness variables suggests also some interaction effects with product and supplier characteristics for which we did not find any evidence. Related to this issue, the effects of service and warranties are interrelated. If service is removed from the analysis, the effect of warranties becomes significant. This is not very surprising since the correlation between the scores of service and warranties are 0.55 and 0.54 in MAT and EDV, respectively.
CONCLUSION AND DISCUSSION In this chapter, we analyzed how purchase managers of Dutch and German firms select their suppliers of IT products. Empirically, it is important that our data concern real transactions, not imaginary ones. We have hypothesized and shown that next to the product and supplier characteristics studied in earlier research, the buyer’s selection decisions are affected by the embeddedness of suppliers. As expected, price and quality of the product were important decision criteria. In addition, a supplier is more likely to be selected if the buyer has positive own past experience, or has information about positive experiences of third parties with this supplier. Information about negative experiences of third parties likewise makes a supplier less attractive, and hence reduces the probability that he is selected. The effect of negative own experiences turned out to be consistently negative although not significant; the non-significance may well be due to a single outlier. Moreover,
Embedded Partner Selection in Relations Between Firms
127
only relatively few buyers reported negative experiences with suppliers. Possibly, buyers do not consider suppliers with whom they have negative experiences still as “potential supplier.” In sum, our results show that embeddedness criteria have a considerable impact on the selection decision of buyers after controlling for standard decision criteria such as price, quality, and service. Through the analyses of interaction effects between attributes of the buyer and of the supplier or product, we found that with increasing damage potential, buyers give higher weight to the quality dimension relative to the other dimensions, especially in relation to the price dimension. Differences in decision making among buyers with varying monitoring problems are more complicated. Buyers seem to care less about the attributes of suppliers and the product, especially about the standard attributes price and quality. The embeddedness attributes remain important, though the relative weights of the attributes change with monitoring problems. Positive own experience has a less positive impact on the attractiveness of a supplier for buyers with more monitoring problems. However, information about negative experiences by third parties is given more weight by buyers with monitoring problems. The results for damage potential are largely in line with our expectation, while the results for monitoring problems, especially that the weight of quality decreases with monitoring problems, pose a theoretical puzzle. One problematic feature of our survey data is that the attributes on which the potential product and suppliers are rated are likely not unrelated, theoretically as well as empirically. For example, we are concerned about the extent to which a buyer’s assessment of quality of the product was affected by the warranties and service that are offered in addition to this product. Even more important, own experience and third-party information with the product is possibly affected by the assessment of the quality of the product and of the service offered by the supplier. In a sense, this is a consequence of using survey data rather than experimental data. The price we pay for interviewing real purchase managers about actual transaction is that we lose some control over the variations in the independent variables, which can be controlled in experimental settings (see Buskens & Weesie, 2000; Rooks et al., 2000). This shows the value of what is called “empirical pluralism” by Buskens et al. (2003) in the introduction of this volume, because testing theories through surveys as well as experiments provides complementary evidence. The limitations of one method are “compensated” by the advantages of the other method. Another problematic feature of the data is the retrospective nature of questions about consideration related to decisions for which the actual outcomes are already known at the time of the questionnaire. For example, the recollection of the satisfaction with the transaction will probably depend on how the supplier or product actually performed. To overcome this limitation for real transactions, it is necessary to question buyers about decision making processes in ongoing
128
VINCENT BUSKENS, RONALD S. BATENBURG AND JEROEN WEESIE
transactions. Using a longitudinal approach, it would be possible to investigate the success of the transactions after the supplier was selected. An additional advantage of such a longitudinal approach would be that the dyadic embeddedness of a buyer-supplier relation can be followed more closely, and that also these data do not depend solely on the memory of the respondents. Although we should be aware of the limitations mentioned above, we believe that we made a contribution to the theoretical and empirical work on partner selection. We extended theoretical issues including embeddedness variables, which indeed turned out to be important for the choices made in purchases. Finally, we believe that the manner in which we collected and analyzed choice data provides a framework that can be extended in future empirically work.
NOTES 1. We realize that our approach might also be prone to systematic biases. We ask purchase managers to evaluate their potential suppliers on a number of dimensions after the choice for a supplier is made. Then, it is possible that these managers adapt their evaluations (probably unintentionally) to justify their choices. 2. There is some related work in the marketing literature, namely, about the existence of “group” or “consortium” purchasing, which emphasizes the advantage of obtaining lower prices by buying as a group rather than buying individually (Essig, 2000; Patterson et al., 1999). This strategy has become especially important within the new field of electronic commerce. The Internet enables marketing and sales in such a way that comparing suppliers and products becomes easier. Virtual market places are founded that enable electronic bidding, auctions, and retailing. These new options do not only allow business at any time and any place, but also intensify the communication between buyers and suppliers and among buyers in order to join purchase activities to achieve economies of scale (Turban et al., 2001). 3. For similar reasons, positive effects might be expected from hiring employees through employees. In that sense, hiring employees is another example of a partner selection process (see Neckerman & Fernandez, 2003 in this volume). 4. If the transaction occurs only after a careful selection and screening process from both partners, we would need theoretical as well as statistical tools adapted for dyadic decision making (see Weesie, 2000). 5. Depending on the branch and the setting, the supplier’s trust problem might be more prevalent than the buyer’s trust problem, for example, if suppliers invest in specific needs of their buyers, while these buyers might rather easily switch to other suppliers. Sako and Helper (1998) recognize the mutual trust problem among buyers and suppliers and investigate trust in the automobile industry mainly from the viewpoint of the suppliers. 6. In logistic and probit regression, only the ratio of the regression coefficients and the scale of the residual are identified (Allison, 1999). This limits the possibility to compare regression coefficients between groups. A similar limitation applies to conditional logistic regression. 7. EDV contains the reputation of the supplier as an additional decision attribute. We did not include the reputation dimension in the analysis of EDV to preserve comparability
Embedded Partner Selection in Relations Between Firms
129
with the MAT. Moreover, reputation did not have a significant effect when added to the analysis presented above, nor did it substantially influence the magnitude of the other decision weights. The same holds true for the analyses reported below that include social embeddedness variables and interaction effects. 8. In EDV, we cannot distinguish between having no information and a missing value. Therefore, we ran the same analysis with the buyers who at least filled out that they had some experience with a supplier or information from third parties about a supplier and excluded in this way respondents who skipped the question. Substantive results for this analysis are the same as the analysis presented here. 9. For this case, the chosen supplier scored never better and sometimes worse on the five product and supplier characteristics than the two alternatives. In addition, the respondent claimed to have negative own experiences and negative third-party information for the chosen supplier, while he only had positive information for the two alternatives. 10. Unfortunately, our data do not contain adequate measures for the monitoring capabilities of these third parties. In line with our other arguments, we expect buyers to assign larger weights to experiences from more knowledgeable third parties.
ACKNOWLEDGMENTS Useful comments by Werner Raub, Arnout van de Rijt, and Chris Snijders are gratefully acknowledged. The data collection for this research was done in collaboration with Roger Berger, Per Kropp, Werner Raub, Gerrit Rooks, Chris Snijders, Frits Tazelaar and Thomas Voss. This research was supported by a grant from the Netherlands’ Organization for Scientific Research (NWO, pp. 50–370) and the Dutch Association for Purchasing Management (Nederlandse Vereniging voor Inkoopmanagment, NEVI). Financial support for the collection of the German data was provided by the Deutsche Forschungsgemeinschaft (DFG), while data collection in the Netherlands was funded by NWO. We gratefully acknowledge permission for using the German data. Buskens’ contribution is part of the project “Third-Party Effects in Cooperation Problems” of the Royal Netherlands Academy of Arts and Sciences (KNAW).
REFERENCES Allison, P. D. (1999). Comparing logit and probit coefficients across groups. Sociological Methods and Research, 28, 186–208. Arrow, K. J. (1974). The limits of organization. New York: Norton. Banker, R. D., & Khosla, I. S. (1995). Economics of operations management: A research perspective. Journal of Operations Management, 12, 423–435. Batenburg, R. S., & Raub, W. (1995). Het externe management van automatisering 1995. [data set] Utrecht University.
130
VINCENT BUSKENS, RONALD S. BATENBURG AND JEROEN WEESIE
Batenburg, R. S., Raub, W., & Snijders, C. (2003). Contacts and contracts: Dyadic embeddedness and the contractual behavior of firms. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 135–188). Oxford: Elsevier Science. Batenburg, R. S., & van de Rijt, A. (1999). The external management of automation 1998 (MAT98). Codebook of MAT98. ISCORE Paper 155. Department of Sociology, Utrecht University. Berger, R., Kropp, P., & Voss, T. (1999). Das Management des EDV Einkaufs 1999. [data set] University of Leipzig. Berger, R., Kropp. P., & Voss, T. (2000). Das Management des EDV Einkaufs 1999. Codebook. Leipzig: Arbeitsbericht des Instituts f¨ur Soziologie, Nr. 14. Blumberg, B. F. (2001). Efficient partner search: Embedded firms seeking co-operative partners. Journal of Mathematical Sociology, 25, 254–329. de Boer, L., Labro, E., & Morlacchi, P. (2001). A review of methods supporting supplier selection. European Journal of Purchasing & Supply Management, 7, 75–89. Bradach, J. L., & Eccles, R. G. (1989). Price, authority, and trust: From ideal types to plural forms. Annual Review of Sociology, 15, 97–118. Buskens, V. (2002). Social networks and trust. Boston: Kluwer. Buskens, V., & Batenburg, R. S. (2000). The external management of automation: Codebook of the combined data from the Netherlands and Germany. ISCORE Paper 175. Department of Sociology, Utrecht University. Buskens, V., & Raub, W. (2002). Embedded trust: Control and learning. Advances in Group Processes, 19, 167–202. Buskens, V., Raub, W., & Snijders, C. (2003). Theoretical and empirical perspectives on the governance of relations in markets and organizations. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 1–18). Oxford: Elsevier Science. Buskens, V., & Weesie, J. (2000). An experiment on the effects of embeddedness in trust situations: Buying a used car. Rationality and Society, 12, 227–253. Doney, P. M., & Canon, J. P. (1997). An examination of the nature of trust in buyer-seller relationships. Journal of Marketing, 61, 35–51. Essig, M. (2000). Purchasing consortia as symbiotic relationships: Developing the concept of ‘consortium sourcing’. European Journal of Purchasing & Supply Management, 6, 13–22. Grabher, G. (1993). Rediscovering the social in the economics of interfirm relations. In: G. Grabher (Ed.), The Embedded Firm: On the Socioeconomics of Industrial Networks (pp. 1–31). London: Routledge. Granovetter, M. S. (1985). Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91, 481–510. Gulati, R. (1995). Does familiarity breed trust? The implications of repeated ties for contractual choice in alliances. Academy of Management Journal, 38, 85–112. Gulati, R., & Wang, L. O. (2003). Size of the pie and share of the pie: Implications of network embeddedness and business relatedness for value creation and value appropriation in joint ventures. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 209–242). Oxford: Elsevier Science. Hirschman, A. O. (1970). Exit, voice, and loyalty: Responses to decline in firms, organizations, and states. Cambridge, MA: Harvard University Press. Hofstede, G. H. (2001). Culture’s consequences: Comparing values, behaviors, institutions, and organizations across nations (2nd ed.). Beverly Hills, CA: Sage.
Embedded Partner Selection in Relations Between Firms
131
Huber, P. J. (1967). The behavior of maximum likelihood estimates under non-standard conditions. Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, 1, 221–233. Kreps, D. M. (1990). Corporate culture and economic theory. In: K. A. Shepsle (Ed.), Perspectives on Positive Political Economy (pp. 90–143). Cambridge: Cambridge University Press. Levy, M., & Grewal, D. (2000). Supply chain management in a networked economy. Journal of Retailing, 76, 415–429. Louviere, J. J., Hensher, D. A., & Swait J. D. (2000). Stated choice methods. Analysis and application. Cambridge: Cambridge University Press. Luce, R. D. (1959). Individual choice behavior. New York: Wiley. Lyons, B. R. (1994). Contracts and specific investments: An empirical test of transaction cost theory. Journal of Economics and Management Strategy, 3, 257–278. Macaulay, S. (1963). Non-contractual relations in business: A preliminary study. American Sociological Review, 28, 55–67. Mummalaneni, V., Dubas, K. M., & Chao, C.-N. (1996). Chinese purchasing managers’ preferences and trade-offs in supplier selection and performance evaluation. Industrial Marketing Management, 25, 115–124. Neckerman, K. M., & Fernandez, R. M. (2003). Keeping a job: Network hiring and turnover in a retail bank. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 299–318). Oxford: Elsevier Science. Patterson, J. L., Forker, L. B., & Hanna, J. B. (1999). Supply chain consortia: The rise of transcendental buyer-supplier relationships. European Journal of Purchasing & Supply Management, 5, 85–93. Raub, W., & Weesie, J. (1990). Reputation and efficiency in social interactions: An example of network effects. American Journal of Sociology, 96, 626–654. Raub, W., & Weesie, J. (2000). The management of durable relations: Introduction and outline of a research program. In: J. Weesie & W. Raub (Eds), The Management of Durable Relations. Theoretical and Empirical Models for Households and Organizations (pp. 1–32). Amsterdam: Thela Thesis. Rogers, W. M. (1993). Regression standard errors in clustered samples. Stata Technical Bulletin, 13, 19–23. Rooks, G., Batenburg, R. S., Raub, W., & Snijders, C. (1998). Het externe management van automatisering 1998. [data set] Utrecht University. Rooks, G., Raub, W., Selten, R., & Tazelaar, F. (2000). Cooperation between buyer and supplier: Effects of social embeddedness on negotiation effort. Acta Sociologica, 43, 123–137. Sako, M., & Helper, S. (1998). Determinants of trust in supplier relations: Evidence from the automotive industry in Japan and the United States. Journal of Economic Behavior and Organization, 34, 387–417. Stuart, T. E. (2003). Governing strategic alliances. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 189–208). Oxford: Elsevier Science. Turban, E., McLean, E., & Wetherbe, J. (2001). Information technology for management: Making connections for strategic advantage. New York: Wiley. Uzzi, B. (1996). The sources and consequences of embeddedness for the economic performance of organizations: The network effect. American Sociological Review, 61, 674–698. Uzzi, B. (1997). Social structure and competition in interfirm networks: The paradox of embeddedness. Administrative Science Quarterly, 42, 35–67.
132
VINCENT BUSKENS, RONALD S. BATENBURG AND JEROEN WEESIE
Uzzi, B. (1999). Embeddedness in the making of financial capital: How social relations and networks benefit firms seeking financing. American Sociological Review, 64, 481–505. Verma, R., & Pullman, M. E. (1998). An analysis of the supplier selection process. Omega, International Journal of Management Science, 26, 739–750. V¨olker, B. (1995). “Should auld acquaintance be forgot . . .?” Institutions of communism, the transition to capitalism and personal networks: The case of East Germany. Amsterdam: Thesis. Weber, C. A., Current, J. R., & Benton, W. C. (1991). Vendor selection criteria and methods. European Journal of Operational Research, 50, 2–18. Weesie, J. (1999). Seemingly unrelated estimation and the cluster-adjusted sandwich estimator. Stata Technical Bulletin Reprints, 9, 231–248. Weesie, J. (2000). Statistical models for dyadic decision making. In: J. Weesie & W. Raub (Eds), The Management of Durable Relations (pp. 130–133). Amsterdam: Thela Thesis. Williamson, O. E. (1985). The economic institutions of capitalism. New York: Free Press. Windmeijer, F. A. G. (1994). A goodness-of-fit test in the multinomial logit model based on weighted squared residuals. Statistica Neerlandica, 48, 271–284.
APPENDIX INFORMATION ABOUT THE QUESTIONNAIRES This appendix provides English translations of the key questions from the German questionnaire (EDV). Question 1. Please rate the selected supplier on the following criteria (use grades). • Price of the product • Quality of the product • Proximity/accessibility • Reputation • Service offered • Warranties
Grade for selected supplier — — — — — —
Question 2. You probably considered other suppliers during the selection process. The next question is about these other suppliers. How many other suppliers did you consider at that time? • None • Several, namely — suppliers Please rate at most two of these other potential suppliers on the following criteria (use grades).
Embedded Partner Selection in Relations Between Firms
133
Grade for first other potential supplier — — — — — —
Grade for second other potential supplier — — — — — —
• Price of the product • Quality of the product • Proximity/accessibility • Reputation • Service offered • Warranties
Question 3. We are interested in the positive and negative information you had, at that time, about the selected supplier and the two other potential suppliers (see previous question). We ask you to consider your own experiences and the experiences of third parties known to you. Selected supplier
First other potential supplier
Second other potential supplier
• About whom did you have positive information based on your own experience?
Yes/no
Yes/no
Yes/no
• About whom did you have negative information based on your own experience?
Yes/no
Yes/no
Yes/no
• About whom did you have positive information based on third-party experience?
Yes/no
Yes/no
Yes/no
• About whom did you have negative information based on third-party experience?
Yes/no
Yes/no
Yes/no
The Dutch questionnaires differed from the German version in a number of respects. First, the Dutch questionnaire asked information about (at most) three rather than two other potential suppliers. Second, question 1 and question 2 were combined in one question. Third, the attribute “reputation” was not included. Fourth, in question 3, the respondent was asked to rank the suppliers on the dimensions “amount” and “positivity” of own experiences and third-party information, rather than rate the suppliers on positive and negative experience and third-party information.
CONTACTS AND CONTRACTS: DYADIC EMBEDDEDNESS AND THE CONTRACTUAL BEHAVIOR OF FIRMS Ronald S. Batenburg, Werner Raub and Chris Snijders ABSTRACT This chapter addresses social embeddedness effects on ex ante management of economic transactions. We focus on dyadic embeddedness, that is the history of prior transactions between business partners and the anticipation of future transactions. Ex ante management through, for example, contractual arrangements is costly but mitigates risks associated with the transaction, such as risks from strategic and opportunistic behavior. Dyadic embeddedness can reduce such risks and, hence, the need for ex ante management by, for instance, making reciprocity and conditional cooperation feasible. The chapter presents a novel theoretical model generating dyadic embeddedness effects, together with effects of transaction characteristics and management costs. We stress the interaction of the history of prior transactions and expectations of future business. Hypotheses are tested using new and primary data from an extensive survey of more than 900 purchases of information technology (IT) products (hard- and software) by almost 800 small- and medium-sized enterprises (SMEs). Results support, in particular, the hypotheses on effects of dyadic embeddedness.
The Governance of Relations in Markets and Organizations Research in the Sociology of Organizations, Volume 20, 135–188 Copyright © 2003 by Elsevier Science Ltd. All rights of reproduction in any form reserved ISSN: 0733-558X/PII: S0733558X02200063
135
136
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
INTRODUCTION The contractual behavior of firms depends not only on characteristics of transactions but also on prior and expected future business contacts between the contracting parties. This chapter offers theory as well as empirical evidence on how contacts affect contracts. While contracts are a standard way of protecting parties against opportunistic behavior and other risks in business relations, we know empirically at least since Macaulay’s seminal and meanwhile classic study (1963) that the use of contracts as a safeguard for problems in transactions between firms is limited. Macaulay studied business relations by analyzing contracts, jurisprudence, and by conducting interviews with businessmen. His main finding (1963, p. 58) was that non-contractual agreements are more important than had often been assumed: “Business men often prefer to rely on ‘a man’s word’ in a brief letter, a handshake, or ‘common honesty and decency’ – even when the transaction involves exposure to serious risks (. . .) keep it simple and avoid red tape.” Macaulay points out that contracts are but one way to prevent problems with a transaction. They are not always used because less costly alternatives exist. For instance (Macaulay, 1963, pp. 63–64), an ongoing stream of transactions with the partner allows for various “effective non-legal sanctions” and these discourage opportunism. A typical case is that “sellers hope for repeat for orders, and one gets few of these from unhappy customers.” More generally, the prospect of recurrent transactions facilitates reciprocity through conditionally cooperative behavior so that the partners can economize on the writing of otherwise extensive contracts. We address the effects of this kind of social embeddedness on ex ante management of economic transactions. Contracting is a core feature of inter-firm relations. Such relations are addressed in a rapidly growing body of research (see, e.g. Jones et al., 1997; Oliver & Ebers, 1998; Nohria & Eccles, 1992; Sitkin et al., 1998 for surveys of such work as well as examples of specific theoretical and empirical studies). Transaction cost theory (Coase, 1937; Williamson, 1985, 1996), a research program initiated outside sociology, offers a theoretically coherent approach towards explaining characteristics of contractual relations between firms (Shelanski & Klein, 1995 as well as Blumberg, 1998, ch. 2 provide surveys of the empirical literature; a representative selection of empirical applications can be found in Masten, 1996). Transaction cost theory focuses on how “economic” characteristics of a transaction affect contracting. Typical characteristics (e.g. Williamson, 1985, ch. 2) are specific investments associated with a transaction and uncertainty about future contingencies. For example, if transactions are associated with uncertainty, it is unfeasible or at least costly to safeguard them exclusively with explicit and binding contracts that are enforced by third parties like the courts. Explicit
Contacts and Contracts
137
contracting is associated with transaction costs (Williamson, 1985, pp. 20–22). These include: (1) costs of anticipating the conceivable contingencies that might arise in the course of a relation; (2) bargaining and decision costs associated with reaching an agreement on how to deal with these contingencies; (3) costs of writing a sufficiently clear and unambiguous contract that can be externally (e.g. legally) enforced; (4) costs of external enforcement (see Hart, 1987, p. 166). The basic idea of transaction cost theory is that firms choose their arrangements for the governance of transactions by economizing on the anticipated costs for reaching and enforcing agreements, so that all potential gains from trade will be realized. In general, economizing will imply that explicit contracts are incomplete in the sense that many conceivable contingencies are not – at least not clearly and unambiguously – covered. In such a case, transactions rely on implicit contracts (Azariadis, 1987), that is contracts that are partly unwritten, tacit, and not formally binding (see also Macneil, 1980). Williamson (1985, ch. 3) elaborated this idea and developed a typology of arrangements, or governance structures, together with conditions specifying when a particular type of governance structure is appropriate. For instance, if transactions are recurrent, involve “sufficient” uncertainty, and require investments that would be useless in other transactions, firms may choose not to write contracts on a transaction by transaction basis, but instead specify future terms of trade in a long-term contract (Joskow, 1987). Under extreme conditions, such as recurrent transactions that require completely idiosyncratic investments, firms may decide to remove their transaction from the market and produce the good internally, thereby integrating production and exchange (vertical integration). Transaction cost theory tends to neglect the social environment and the interconnectedness of transactions (e.g. Granovetter, 1985; but also Milgrom & Roberts, 1992, pp. 32–33). Of course, Williamson (1985, chs. 2 and 3) highlights “frequency” together with asset specificity and uncertainty as one of three principal dimensions for developing a predictive theory of economic organization. However, his frequency dimension refers strictly to “buyer activity in the market” (1985, p. 72) rather than addressing repeated transactions between the same partners or relations of business partners with third parties. The effects of the social environment and the interconnectedness of transactions have always been a typical focus of sociological approaches to contracting. Durkheim has forcefully argued in his analysis of the division of labor in society (1893, book I, ch. 7) that typical features of many economic transactions deviate from those of transactions that are conventionally assumed in standard models of neo-classical economics. Real-life
138
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
economic transactions are different from what economists usually label as “spot exchange on perfect markets.” More specifically, Durkheim highlighted the limits of what is today often called “contractual governance” of economic transactions. As Durkheim clearly realized, the governance of transactions exclusively via bilateral contracts requires that the present and future rights and obligations of the partners involved in the transaction are specified explicitly for all circumstances and contingencies that might arise during and after the transaction. Anticipating much of the modern economic and game theoretical literature on incomplete and implicit contracts, Durkheim pointed out that such purely contractual governance of economic transactions is problematic: Typically, many unforeseen or unforeseeable contingencies could or actually do arise during or after a transaction. Negotiating a contract explicitly covering all these contingencies would be unfeasible or at least prohibitively costly. Likewise, renegotiations in the case that contingencies arise are also costly. Such renegotiations characteristically offer incentives for opportunistic behavior since an unexpected contingency will often strengthen the bargaining position of one of the partners while weakening the position of the other. Hence, Durkheim argued, mutually beneficial economic exchange presupposes that trading partners follow non-contractual norms and moral obligations, such as norms of trust, reciprocity, and solidarity, that cannot be enforced through the courts and that complement contractual arrangements. Thus, written contracts may be a standard way but presumably not the only way of protecting parties against opportunistic behavior and other risks in business relations. Durkheim’s analysis of economic transactions addressed not only a crucial feature of societies based on division of labor and economic exchange between its members. His analysis was likely meant to show that sociology could offer insights in the analysis of economic exchange that would add in important ways to the “utilitarian” market model of economics. Given that, it seems surprising that sociology has neglected for quite some time to elaborate Durkheim’s arguments and to support them with systematic empirical evidence. The topic Durkheim addressed did not speedily induce the development of a coherent and broad research program on the “sociology of economic life.” Rather, renewed sociological interest on this topic emerged from the sociology of law. Interestingly, arguments on the limits of contractual governance of economic transactions similar to Durkheim’s had already been presented by Weber in his sociology of law (see Weber, [1921] 1976, p. 409). However, it seems fair to say that it was Macaulay (1963) who put the topic back on sociology’s agenda. First, Macaulay presented a broad range of “qualitative” evidence on the limited use of contracts, thus demonstrating the empirical validity of Durkheim’s point. Moreover, he offered a rich set of intuitive explanations why non-contractual relations in business are feasible and how they complement contractual governance. The “law and society” approach in
Contacts and Contracts
139
the sociology of law built on Macaulay’s study (see, e.g. Beale & Dugdale, 1975; Ellickson, 1991 as an important recent contribution), providing rich qualitative case studies as well as more elaborated though typically informal theoretical accounts. The currently most influential sociological approach to the analysis of economic exchange is undoubtedly the new economic sociology (see Smelser & Swedberg, 1994 for a representative overview as well as collections like Swedberg, 1993). This research program has been heavily influenced by Granovetter’s (1985) programmatic article that revitalized Polanyi’s (1944) notion of the “embeddedness” of economic action and argued forcefully for systematically incorporating the effects of social embeddedness in the analysis of economic transactions. The notion of embeddedness covers a variety of dimensions (for a discussion, see Weesie & Raub, 1996, pp. 203–205) so that it is useful to outline how these relate to our analysis here. First, embeddedness refers to institutions, that is, to the constraints for economic (and other) action and exchange that result from human behavior itself. Institutions structure the incentives in social relations. In other words, institutions such as contract law, the availability of standard contracts, warranties and guarantees, or arbitration procedures but also credit rating services constitute the formal or informal “rules of the game” in which economic actors are involved (North, 1990, ch. 1). A characteristic feature of sociological approaches to institutional embeddedness is a focus on the explanation of the emergence and stabilization (or decline) of institutions as a result of “social construction” (see, e.g. Granovetter, 1992) rather than assuming institutional embeddedness as exogenously given. Second, and closer to our concerns in this article, embeddedness of economic exchange refers to ties and relations between the partners as well as their ties and relations with third parties. Such ties and relations include non-economic, personal ones that have repercussions for economic exchange. Macaulay (1963, p. 64) observed that gossip exchange at meetings of purchasing agents’ associations and trade associations as well as at country clubs and social gatherings may deter opportunism and hence reduce the need for contractual arrangements. Granovetter (1985, p. 492) points out that the fascinating practice of diamond exchange sealed by a handshake and without contractual insurance against theft and fraud is supported by the embeddedness of such exchange in close-knit communities of diamond merchants: behavior can be easily policed by quick spread of information and sanctions of malfeasance include the loss of family, religious, and community ties (see also Coleman, 1988, p. S99). However, the embeddedness of a focal transaction likewise includes other economic exchange between the partners as well as their economic exchange relations with third parties. Granovetter (1985, pp. 490–492) ably observed the effects of past dealings with the partner as well as effects of expected future dealings with the partner for a focal transaction. He likewise observed the
140
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
repercussions of economic exchange relations with third parties. In particular, Granovetter points out that the embeddedness of a transaction in a relation of past and future dealings between the partners as well as in a network of economic relations with third parties provides information on the partner as well as sanctioning opportunities via reputation effects (see Raub & Weesie, 1990 for a theoretical analysis of such reputation effects). Research in the spirit of the new economic sociology has meanwhile produced a sizable amount of empirical evidence supporting that effects of social embeddedness exist (see, e.g. many contributions in Nohria & Eccles, 1992; Sitkin et al., 1998; Swedberg, 1993). Much of this work does indeed focus on the effects of embeddedness in the sense of other dealings with a partner as well as economic relations with third parties (see, e.g. Baker et al., 1998; Gulati, 1995a, b; Gulati & Gargiulo, 1999; Larson, 1992; Lyons, 1994; Uzzi, 1996, 1997; in this volume, see the contributions by Stuart, 2003 as well as Gulati & Wang, 2003). In this chapter, we contribute to clarifying the effects of embeddedness on economic exchange. Rather than considering personal, non-economic ties and relations or ties with third parties, we focus on dyadic embeddedness of a transaction in a sequence of economic exchanges between the same partners. Effects of dyadic embeddedness have been investigated in a number of recent studies. Some of these focus on the effects of prior business on the present transaction (e.g. Gulati, 1995b; Lyons, 1994). These studies do address various features of contracting and how transactions are arranged ex ante. For example, Gulati addresses the choice between equity and nonequity alliances. Other studies try to incorporate the effects of expected future transactions (e.g. Heide & Miner, 1992; Noordewier et al., 1990; Parkhe, 1993). A common feature of these latter studies is that they address effects of dyadic embeddedness on ex post performance rather than the effects of dyadic embeddedness on the ways of arranging transactions ex ante. They ask how dyadic embeddedness affects outcomes such as shared problem solving between partners, restraint in the use of power, the avoidance of opportunistic behavior, or the fulfillment of various strategic needs of the partners. We consider the effects of dyadic embeddedness on ex ante contracting and, more generally, ex ante management. Note that this provides a stronger test of hypotheses on embeddedness. Analyzing performance effects of embeddedness focuses on whether economic actors react in predicted ways to the incentives associated with embeddedness. Hypotheses on effects of embeddedness on contracting assume such reactions as given and ask the theoretically deeper question whether contracting characteristics can be understood using the assumption that actors choose such characteristics with performance effects in mind (see Prendergast, 1999 for a similar argument in a different but related context: the design of compensation contracts by employers to align the interests of employees). We add two novel contributions to previous research on
Contacts and Contracts
141
the effects of dyadic embeddedness on contracting in inter-firm relations. First, we distinguish explicitly between the history of prior transactions between business partners (“shadow of the past”) and their anticipations of future transactions (“shadow of the future,” see Axelrod, 1984). Our analysis includes effects of both the shadow of the past and the shadow of the future on contracting and ex ante features of transactions. Second, we show how the shadow of the past and the shadow of the future interact in affecting contractual planning of transactions. We argue that the way in which expected future business affects present contracting depends decisively on previous transactions between the partners. A striking feature of previous empirical research on contracting is that virtually all studies consider the choice between governance structures: some kind of specification is included in the contract or not, production occurs in-house or not, or questions of a similar nature. Essentially, this has been the main thrust of the empirical literature within and outside transaction cost economics: governing a transaction is costly, and properties of the transaction as well as, eventually, embeddedness characteristics determine which governance structure is the least costly and therefore the most appropriate. Instead, we focus on the extent to which a transaction is governed. Given that contracts are used to govern a transaction, the question remains how much time and effort will be invested in ex ante management, and how explicit the contract is going to be. Our focus on the degree of governance not only yields a new explanandum but also allows for more robust explanations by employing more parsimonious assumptions. Deriving hypotheses on the choice between different governance structures requires assumptions on which governance structure is optimal at a certain transaction cost level (e.g. Williamson, 1996, ch. 4; see Milgrom & Roberts, 1996, pp. 466–467 on some problems associated with such assumptions) in addition to the assumption that firms will economize on transaction costs. By focusing on hypotheses on the extent to which a transaction is governed rather than on the choice between different governance structures, we no longer need to employ additional assumptions on comparative-cost relations between alternative governance structures. We try to develop two points theoretically as well as empirically. First, following Durkheim’s theme as well as the research program of the new economic sociology, we wish to show how contractual governance of economic transactions is complemented and supplemented by norms of solidarity and reciprocity that allow for trust. Following a simple and common conceptualization (see, e.g. Barber, 1983; Burt & Knez, 1995; Coleman, 1990, ch. 5; Gambetta, 1988; Snijders, 1996, ch. 1) we consider trust as the willingness of an actor – the trustor – to incur a risk. More precisely, the trustor places him- or herself in a situation where another actor – the trustee – can perform two different actions: either the trustee performs an action
142
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
so that the trustor is better off than had he or she not placed trust or the trustee performs an action such that the trustor is worse off than if trust were not placed. This concept of trust covers two different cases. The case where the risk is that the trustee is willing to perform in the interest of the trustor but lacks the resources, knowledge or skill to do so. This is “trust in competence,” often referred to in the literature as “confidence.” And, the case where the risk is that the trustee has the resources, knowledge or skill to perform in the interest of the trustor, but is not willing to do so. This is the case of “strategic trust” that is analyzed in game theoretical models (see Camerer & Weigelt, 1988; Dasgupta, 1988; Kreps, 1990). We wish to show how trust in both senses is stabilized by one specific dimension of social embeddedness of economic transactions. In addition, we try to endogenize trust as a “lubricant” (Arrow, 1974) of economic exchange. We do this by conceiving of norms of solidarity and reciprocity as rules of conditionally cooperative behavior and by asking when it will be in the best interest of (rational) economic actors to follow such norms. In other words, we consider costly contractual governance and trust based on conditional cooperation as alternative modes of coping with the problem potential involved in economic transactions and analyze how social embeddedness of transactions affects the rational choice between contracts and trust. In Williamson’s (1993) sense, our approach centers on “calculative trust.” By incorporating the assumption of rational behavior as the theoretical core of the analysis, we certainly deviate from the approach Durkheim envisaged as well as from much of the new economic sociology. However, a rational choice approach per se is less at odds with criticism on the narrow neoclassical model than one might think (see Voss, 2003 for a thorough discussion in this volume). Note that Granovetter (1985, pp. 505–506) advocated precisely such a combination of assumptions on the embeddedness of economic behavior and robust assumptions on rational and – in principle – selfish behavior in his often cited programmatic sketch. Granovetter’s criticism of the shortcomings of the neoclassical model of perfect markets of “atomized” actors and transactions has often been enthusiastically endorsed and taken to imply that one had better abandon rational choice models in favor of more “realistic,” socially inspired models of man. It has been widely overlooked that he sharply opposes “psychological revisionism” which he characterizes as “an attempt to reform economic theory by abandoning an absolute assumption of rational decision making” (1985, p. 505). Rather, he suggests to maintain the rationality assumption: “[W]hile the assumption of rational action must always be problematic, it is a good working hypothesis that should not easily be abandoned. What looks to the analyst like nonrational behavior may be quite sensible when situational constraints, especially those of embeddedness are fully appreciated” (1985, p. 506). He argues that investments in tracing the effects of embeddedness
Contacts and Contracts
143
are more promising for sociologists than investments in the modification of the rationality assumption: “My claim is that however naive that psychology [of rational choice] may be, this is not where the main difficulty lies – it is rather in the neglect of social structure” (1985, p. 506). In fact, Granovetter advocates an approach that is surprisingly similar to the position typically associated with Coleman (1987; see Voss, 1985 for an early discussion of a similar perspective). In this article, we exploit the idea that rational choice arguments and an approach focusing on the embeddedness of economic action “have much in common” (Granovetter, 1985, p. 505). We try to contribute to the construction of an interface for both approaches that profits from their strengths while avoiding their weaknesses. Simple principles of action from rational choice theory that are useful for deriving testable hypotheses explicitly and systematically from a common theoretical core are often neglected in the new economic sociology. On the other hand, we introduce a core aspect of the social organization of economic exchange into the analysis that differs from the context typically studied in economic approaches. To test our hypotheses, we present data on the market for IT-products (hardware and software), based on extensive primary data collection on 971 IT-transactions between 788 Dutch small and medium-sized enterprises (SMEs) and their IT-suppliers in the period 1990–1995. We first outline some characteristics of the Dutch market for IT-products and argue that transactions on this market offer strategic opportunities for an empirical study of the management of trust in economic transactions. The following two sections offer intuitive verbal theory on, first, effects of transaction characteristics on ex ante management and, second, effects of dyadic embeddedness. Subsequently, we introduce a novel formal theoretical model generating these hypotheses. We then describe our data and present estimation results. We conclude with a general discussion.
BUYER-SUPPLIER RELATIONS ON THE DUTCH IT-MARKET The Dutch market for hardware and software applications has grown rapidly in the last two decades (Schellekens et al., 2000; Statistics Netherlands, 1998). IT is widely used in every industry and IT-investments constituted about 2.2% of the domestic product in 1997. Though this is still below the U.S.-level of 3.2%, it is above average in comparison with other European countries (Jacobs & de Vos, 1992) and the importance of the IT sector is growing. The impact of IT on economic growth is a debated issue, but according to Dutch calculations 25% of the economic growth in the Netherlands in 2000–2001 was due to the IT sector
144
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
(van der Wiel, 2000; see also Hollanders, 2000 and for thorough recent studies of the situation in the U.S. Oliner & Sichel, 2000; Brynjolfsson & Hitt, 2000; Gordon, 2000). In the seventies and early eighties, potential customers had relatively little knowledge about IT-products, but often had the idea that they should invest in it to keep their market position. As a consequence, the market could be characterized, loosely speaking, as a typical seller’s market. Firms specializing in IT – and willing to reap quick and substantial profits – rapidly emerged. The relative ignorance of the potential buyers, the widespread existence of suppliers not that concerned about their reputation, and the substantial risks associated with IT-purchases (due to, e.g. complexity of the products, monitoring problems, and often high switching costs) made for transactions regularly hampered by problems. These problems frequently emerged because of discrepancies between what the customer thought he would get and what the IT-supplier actually provided (see Auer & Harris, 1981; Riesewijk & Warmerdam, 1986). Over time, the market settled somewhat: some of the less reliable IT-suppliers disappeared and IT-buyers got more acquainted with the kind of benefits investments in IT can generate. Furthermore, several institutional reactions are noteworthy. Since 1992, the separate business associations of Dutch software and hardware suppliers have joined forces in a new organization, the Federation of Dutch Information Technology (FENIT). At the same time, buyers have organized themselves in user associations. For instance, some user associations have specifically been founded to reassure updates and maintenance of software applications when original suppliers do no longer exist or have been taken over by other (larger) IT-companies. Since potential IT-buyers got both better informed and better protected, the market changed from a seller’s market to a buyer’s market. Most firms know reasonably well what they want to have, and that they can get what they want at more than one place. However, rapid improvement of hardware performance and software applications still causes considerable uncertainties with respect to price and quality. In addition, the young and growing market for IT-consultancy and services is still characterized by high rates of firms going bankrupt as well as frequent mergers and acquisitions. Thus, uncertainty with respect to the continuity of transactions still implies considerable risks associated with specific investments and long-term business relations (see, e.g. Schellekens et al., 2000). To summarize, while our theory will obviously apply to transactions on other markets as well as to exchange in inter-firm relations different from buyer-supplier relations, the IT-market appears to be a suitable context and a strategic research site for studying trust and contracting in inter-firm relations. Because of the nature of the IT-business, incentives for opportunistic behavior exist, buyers of IT-products are by now aware of this, and they face the problem of finding an adequate way to manage the specific transaction.
Contacts and Contracts
145
THEORY AND HYPOTHESES ON CONTRACTING: EFFECTS OF TRANSACTION CHARACTERISTICS AND MANAGEMENT COSTS We consider investing in transaction management to be a response to the problem potential associated with a transaction. Potential problems include unforeseen or unforeseeable contingencies like sudden changes in market prices for components. Also, problems can result from strategic risks such as opportunistic behavior. Investing time and effort in ex ante management, the actors may try to reduce such risks. For example, investing time and effort in communication can mitigate coordination problems. Likewise, investing time and effort in negotiating a contract that includes additional or more detailed specifications can mitigate risks from external contingencies. Or, investing time and effort in negotiating warranties, guarantees, and penalty clauses can reduce incentives for opportunistic behavior and it can compensate for the damage inflicted when – in spite of all efforts – opportunistic behavior occurs. Consequently, increasing investments in transaction management increases the probability that the transaction runs smoothly and that a party suffering from a problem gets compensated for losses. However, this decrease in risks comes at a price, namely, the price of typical transaction costs. It could make sense to include some explicit clauses on the time of delivery of the product in a contract – an especially appropriate example given that we are considering the IT-market. On the other hand, negotiating and writing these clauses takes time and effort, so it could be that including them is not efficient. Instead, it may make more sense to invest in trying to make sure that delivery problems will not occur by investing extra time in communicating the specifications of the product. The partners confront the complicated decision to find the optimal investment in governing the transaction. We try to explain such decisions from rational behavior of the transaction partners. Basically, if the problem potential is small, actors have good reasons to trust and have thus fewer incentives to invest in costly ex ante management. Conversely, if risks increase, trust is problematic and costly management makes sense. From the perspective of trust as the willingness of an actor to incur risks, it makes sense to distinguish between two dimensions of the problem potential. One dimension, to which we will refer as the opportunism potential, are the possibilities and incentives of the trustee to behave opportunistically (that is, in a way that impairs the trustor). The other dimension will be refered to as the damage potential and represents the extent to which the trustor would be hurt if the trustee does not perform in the interest of the trustor. Hence, the trustor’s risk increases with increasing opportunism potential as well as increasing damage potential and increasing risk induces more ex ante management.
146
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
We concentrate on the buyer’s risks in buyer-supplier relations and on how the buyer invests in ex ante management. Putting more effort in preventing problems increases the probability that the transaction runs smoothly, but the extra effort comes at a price. Thus, the transaction is conceived as a trust problem in which the buyer is the trustor and the supplier is the trustee. In order to avoid complex modeling of bargaining and negotiation between buyer and supplier, we employ a standard way of including market conditions in game theoretical models on behavior in markets (see, e.g. Rasmusen, 1994, pp. 169–170 for a more technical discussion) and assume for simplicity that, given a match between a buyer and a supplier has formed, the buyer determines the degree of planning for the transaction. The buyer must balance costs and benefits of extra management of the transaction. More specifically, we assume that the buyer takes into account how a rational supplier will react to the buyer’s planning efforts. The buyer chooses a degree of planning such that the supplier’s utility level from the transaction equals the reservation utility, i.e. the minimum for which the supplier would be willing to deliver. Given the market under scrutiny, it is reasonable to assume that the supplier is one of many competitors, so that the buyer can always find another supplier who is willing to sell on the buyer’s terms if the transaction is marginally profitable for the supplier. Whereas this assumption does abstract from many of the dynamics of real life dealings between business partners, the assumption that the buyer determines the degree of planning is less problematic than might appear at first sight and does not imply that we completely neglect the role of the supplier in contracting. First, the supplier has an incentive to accept the buyer’s demand for contractual safeguards or to even provide them voluntarily because otherwise the buyer might not be willing to buy at all. Second, the supplier’s reservation utility could be set such that the safeguards do not exceed those the supplier would have had to provide if the buyer had to compete with other buyers for the same delivery. By explaining contractual planning as a device for mitigating the problem potential associated with a transaction we neglect the internal communication function of contracts within the buyer’s firm as well as within the supplier’s firm (see Macaulay, 1963, p. 65 for a discussion of this feature of contractual practices). Also, by addressing investments in contractual planning we neglect search and selection processes (see, e.g. Blumberg, 1998; Gulati & Gargiulo, 1999). That is, we neglect the cases where a buyer might carefully and extensively search for a trustworthy supplier offering a good product for a reasonable price in order to invest less in subsequent contractual planning. In our statistical analyses we will control for this neglected communication function as well as for search and selection efforts. As we will show, our empirical results do not depend on these assumptions (see the section on stability of results).
Contacts and Contracts
147
We now consider conditions that determine the extent to which buyers invest in ex ante management of a focal transaction through explicit and implicit contracting, starting with determinants of such investments that can be conceived as transaction characteristics. Such transaction characteristics are the typical focus of economic approaches to contracting.
Opportunism Potential It is close to tautological that the problem potential of a transaction increases with the number of possibilities for opportunistic behavior and the incentives for such behavior. One of the determinants of opportunism potential is the size of the transaction (for a detailed overview of the indicators that were used in the analyses, see the section on data and measurement and Appendix B). Transaction size can be defined in terms of its financial volume – which is the proxy we use in our analyses – or, often but not always closely related, the number of products or components of a product involved in the transaction. The more products are included in the transaction, the more possibilities for buyer and supplier to disagree about some aspect of the transaction. For instance, a hospital buying a database program is less likely to experience trouble than a hospital buying a database program, network facilities, and cabling. In daily practice, this problem is sometimes anticipated and dealt with by splitting up transactions involving a larger number of goods into separate deliveries, thereby spreading the opportunism potential across different points in time (see, e.g. Burt, 1992, ch. 7). Here, we try to explain the management of transactions that vary in size but are assumed to be restricted to one certain point in time. A second determinant of the opportunism potential is the degree to which monitoring problems exist. The more difficult it is to judge the quality of a product or service, the larger the opportunism potential surrounding a transaction. Consider for instance a firm selling an automated baking machine to a bakery. For some reason, the baking machine performance does not meet the bakery’s standards. If the bakery does not know what a good baking machine can produce, the supplier could tell the bakery that this is as good as baking machines get, and get away with selling an inferior product. Whether monitoring problems are the result of the objective complexity of the machine, or the result of the ignorance of the bakery, is not so much the issue here. The point is that in anticipation of monitoring problems, the baker would be wise to consider investing more in managing this transaction. This can be achieved, for instance, by investing time in getting acquainted with the technology. Note the difference between this definition of technological uncertainty and uncertainty in the sense of Williamson (1985, pp. 56–59).
148
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
Williamson distinguishes between uncertainty due to external contingencies and strategic, or behavioral, uncertainty that results from risks due to opportunistic behavior of the partners, including “strategic non-disclosure, disguise, or distortion of information” (1985, p. 57). We stress that monitoring problems due to objective complexity of a product or lack of expertise of the buyer make opportunism feasible and individually attractive for the supplier. Hence, monitoring problems increase opportunism potential and are thus expected to induce less trust and more investments in ex ante management. Hypothesis 1. The opportunism potential (as indicated by volume and monitoring problems) of transactions will have a positive effect on the investment of the buying firm in ex ante management.
The Consequences of Problems: Damage Potential Apart from the possibilities and incentives for opportunism, it also matters how severe the consequences would be if problems actually do occur. The extent to which these problems would hurt the business partners determines the transaction’s damage potential. Again, the financial volume of a transaction can serve as an indicator for the damage potential. The more expensive the transaction, the larger the impact on the firm’s profits if problems arise, and thus the larger the damage potential. The importance of the transaction for the buyer in terms of the importance of the durability of the product and the importance of the product for the buyer’s profitability likewise indicate the damage potential. Another indicator are the costs that would be incurred if the product would fail and had to be replaced. This implicitly includes capital that is lost because of investments that are specific to the transaction (see Williamson’s, 1985 arguments on the effects of asset specificity). The larger these costs, the more the firm would suffer losses on the deal, and therefore the larger the damage potential. Hypothesis 2. The damage potential of transactions (as indicated by the financial volume, the importance of the transaction for the buyer and the replacement costs for the buyer if the product fails) will have a positive effect on the investment of the buying firm in ex ante management.
The Costs of Management The efficiency of preventing potential problems of a transaction by ex ante management also depends on how costly it is to prevent these problems. First, firms need
Contacts and Contracts
149
to know or find out which kind of problems are likely to occur. Two firm resources are important in this respect. For firms with legal expertise, it is easier to realize an accurate contract since they know better which kind of problems to prevent. Consequently, other things equal, these firms will invest more in contracting: they write longer contracts (which implies more ex ante management) because for them the marginal transaction costs are lower. Additionally, some firms have tried in the past to minimize their overall costs of management by implementing standardized procedures that are to be followed during negotiation and contracting. Assuming that these procedures indeed decrease the costs of management (as compared to the situation without these procedures), such firms will on the average invest more than others in contracting, since their management is less costly at the margin. Hypothesis 3. The marginal costs of ex ante management (as indicated by the legal expertise and standardized procedures of the firm) will have a negative effect on the investment of the buying firm in ex ante management.
THEORY AND HYPOTHESES ON CONTRACTING: EFFECTS OF DYADIC EMBEDDEDNESS Despite the major impact of the transaction costs approach, criticisms were not slow in coming. A general point of criticism is that too much attention has been given to the “economic” characteristics of transactions like volume and damage potential, and too little attention has been paid to the social environment of transactions. As mentioned before, we focus on one aspect of social embeddedness, namely, the dyadic embeddedness of a focal transaction in the sense of connections with other dealings between trading partners (rather than embeddedness in personal, non-economic ties). We distinguish between two features of dyadic embeddedness: prior transactions between the same two partners (the shadow of the past) and expected future transactions between them (the shadow of the future). Effects of the shadow of the past have been a topic of previous research on contractual management of inter-firm relations (Gulati, 1995b; Lyons, 1994). The shadow of the future is a central factor in game theoretical models for cooperative behavior as has been forcefully argued by Axelrod (1984). Noordewier et al. (1990), Heide and Miner (1992), and Parkhe (1993) have shown empirically that the shadow of the future has performance effects in inter-firm relations. In the following, we examine the interplay between both types of dyadic embeddedness and the ex ante management of firms (see Raub, 1996). Hence, we integrate the conventional sociological focus on trust as emerging from past exchanges and the economic perspective on trust as a result of incentives related to future exchanges
150
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
(see Burt & Knez, 1996). We argue that the effects of the shadow of the future depend crucially on the shadow of the past.
Shadow of the Past The history of transactions with the same partner provides information about the characteristics and the exchange behavior of each of the partners. When entering a new relation, there is a positive probability that the partner is incompetent or excessively inclined to behave opportunistically. This could be the case, for instance, when the partner is on the verge of bankruptcy so that short-term incentives for opportunistic behavior outweigh even severe long run costs from sanctions. In other words, when entering a new business relation, firms do not know whether they face a “normal” partner who is competent and not excessively inclined to opportunism, or a “deviant” partner who is either incompetent or opportunistic. Once firms find out their partner is either not competent or not reliable, this knowledge is probably sufficient to make firms try to find better partners (our data will in fact support this: almost all the firms in the data who have previous experiences with their partner, have positive experiences). We thus assume that buyers with bad experiences will try to find a more suitable supplier, so that buyers having bad experiences with a given supplier and likewise expecting future business with the same supplier do not exist. Hence, we do not model exit of buyers from bad relations with suppliers explicitly and assume that exit costs are not prohibitive because alternative suppliers are available and relation specific investments in the past are sufficiently small. Conversely, positive information on previous transactions will increase trust in the partner’s competence and in the partner’s capability to withstand a short-term temptation for opportunistic behavior. In the latter case, the need for investments in ex ante management of the focal transaction is reduced. A second reason why favorable previous relations with the same business partner may decrease the current investment in management is because partners can make use of their prior investments. Previous investments such as reusable contracts, earlier agreements on certain quality standards, and knowledge about the way to approach the partner will decrease the need for costly new investments in current management. In other words: management investment is probably not completely transaction specific. It seems reasonable to assume that a positive past relationship comes along with relationship-specific investments not only in management but also in other respects. For example, the supplier may have invested in training of employees specifically designed to facilitate service and maintenance of products of the supplier that have been purchased by the buyer. One-sided relationshipspecific investments of the buyer increase unilateral dependency of the buyer on
Contacts and Contracts
151
the supplier and thus enhance the problem potential because the buyer’s damage from malperformance of the product or the supplier increases. Conversely, mutual previous investments reduce the opportunism potential (see Williamson, 1985, pp. 190–195) and, through a reduced opportunism potential, reduce investments in ex ante management of the current transaction. In short, we expect negative previous experience to lead to termination of the partnership and positive previous experience to a decrease in the investments in current ex ante management. Hypothesis 4. Positive previous experiences with the same supplier will have a negative effect on the investment of the buying firm in ex ante management. Note that, in the spirit of Granovetter’s argument, we focus on “anchoring effects” of prior investments in management exclusively as a result of fully rational decision making. Biases due to non-rational adjustments, a typical focus of (social) psychological research on anchoring (see, e.g. Tversky & Kahneman, 1982 for an overview), are not taken into account.
Shadow of the Future In his classic study, Macaulay specifically addressed the use of long-term contracts. Instead of standard short-term agreements, long-term contracts can be used to control business relations. Long-term contracts provide a future, and thereby a way to build up a general business reputation (Macaulay, 1963, p. 63). Analyzing the coal market, Joskow (1987) showed that there is a relation between contract duration and relationship-specific investments: coal suppliers and energy plants more often rely on long-term contracts as relationship-specific investments become more important. At first sight, a large perceived probability of future transactions indeed facilitates less investment in current ex ante management. If transactions are likely to be followed by future transactions, this shadow of the future provides opportunities to preclude opportunistic behavior through tit-for-tat like (i.e. conditionally cooperative) behavior. This is the core argument of game theoretical approaches to repeated interactions (see Kreps, 1990 for an informal account and Taylor, 1976/1987 as well as Axelrod, 1984 for stimulating and influential applications in political science). The threat of sanctioning opportunistic behavior implicit in the mechanism of conditional cooperation makes extensive ex ante management of the focal transaction superfluous. Thus, given a sufficient shadow of the future, it becomes individually rational to indeed follow a norm of reciprocity. Obviously, this would allow substituting trust for costly investments in ex ante management. However, there are reasons to argue exactly the opposite with respect to the relation between the shadow of the future and investments in ex ante management.
152
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
Partners who deal with each other for the first time and have reasonable expectations about the future have incentives to invest more in ex ante management because future transactions will benefit from the set-up investments in the current transaction (see Williamson’s, 1985, pp. 60–61 related discussion of “frequency” of transactions that favors specialized governance structures). In general, some proportion of the investment in management of a given transaction will be useful for the management of future transactions with the same partner as well. For instance, firms who are likely to deal with a business partner more often in similar transactions may choose to be extra careful in the design of the first contract, since it will guide subsequent transactions. Even if transactions are diverse, some parts of contracts written earlier may be useful in future transactions (our data will support the assumption that written contracts are often reused). Moreover, management of a transaction with an unknown partner requires set-up investments of other kinds, like getting to know the partner, knowing whom to call for which kind of information, and the like. We therefore argue that the shadow of the future has two effects on investments in ex ante management. One of these is a reciprocity effect: reciprocity as a basis of trust and, thus, as a substitute for contractual governance is facilitated and this reduces incentives for costly ex ante management. The other is an investment effect: costly ex ante management of the focal transaction has long run effects for future transactions and can be partly reused, thus increasing incentives for ex ante management. These effects yield two implications. First, with regard to business partners without common previous transactions, the relation between a shadow of the future and transaction management is unknown. On the one hand one would expect a negative relation because of future sanction threats deterring opportunistic behavior. But on the other hand one would expect a positive relation because of the need for and the long run benefits from set-up investments. We have no arguments regarding the relative weight of both arguments. Second, however, a shadow of the future leads to a decrease in ex ante management in those cases where set-up investments have already been made (hence: a shadow of the past exists). If a shadow of the past exists, the effects of investments in ex ante management of the focal transaction on the management of future transactions are smaller. Thus, we derive a novel hypothesis regarding the interaction effect between the shadow of the past and the shadow of the future on investments in ex ante management. Business partners, who are past the stage of set-up investments, should indeed benefit from a large shadow of the future by investing less in ex ante management. Hypothesis 5. Given positive previous experiences with the same supplier, the shadow of the future will have a negative effect on the investment of the buying firm in ex ante management.
Contacts and Contracts
153
Summarizing, note that our hypotheses show how and when rational actors will substitute trust at least to some degree for costly investments in ex ante management. Hence, we capture Durkheim’s conjecture that contractual arrangements are complemented by reciprocity. Moreover, we have argued how the choice between contractual and non-contractual management is affected not only by “economic” features of a transaction but also by a core dimension of the social embeddedness of the transaction, namely, dyadic embeddedness.
FORMAL MODEL SPECIFICATION We now provide a theoretical model that captures in a more formal way the arguments we have just put forward. At the heart of the model, we envisage a buyer’s utility function depending on opportunism potential, damage potential, costs of management, embeddedness characteristics, and – finally – effort invested in management, which is the buyer’s choice variable. We first define some variables capturing the dimensions we have put forward. Ipast represents an indicator function equal to 1 if a shared past exists, O represents the opportunism potential, F a function that maps the real numbers to the unit-interval (for instance the standard normal cumulative distribution function), c the marginal costs of management, w the probability of future business, and D the damage potential for the focal transaction. Putting more effort in preventing problems increases the probability that the focal transaction runs smoothly, but the extra effort comes at a price. Costs and benefits of extra management of the transaction must be in balance. The following theorem makes the relation between optimal management and our independent variables explicit. Details can be found in Appendix A. Theorem. Consider a match between two actors in a durable relationship, who have to decide the degree of planning for a focal transaction. Let this focal transaction be characterized by the tuple (Ipast , O, c, w, D) with meanings as described above. Here, we choose these two parties to be a buyer and supplier, but other kinds of actors are, in principle, just as feasible. Assume that the buyer determines the degree of planning. Under these conditions, the optimal amount of management (mopt ) can be characterized by: −1 c(1 − wg 1 ) − g 0 I past , m opt = (1 − g 1 I past ) −b 0 + b 1 O + b 2 F b3D where the bi and gi are parameters to be estimated and supposed to be positive. Proof. See Appendix A.
154
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
Note that the theorem indeed connects optimal management with the theoretical dimensions of the previous sections in the hypothesized ways. This shows that our hypotheses can be derived from a more general model of choice. Optimal investment in management increases with the opportunism potential (O; Hypothesis 1) and the damage potential (D; Hypothesis 2). Conversely, optimal management decreases with the marginal costs of management (c; Hypothesis 3). Furthermore, optimal investment in management decreases if a shared past exists (Ipast ; Hypothesis 4). In particular, using a linear Taylor expansion t 0 + –1 t 1 x (t 1 < 0) for F (x), we can conclude that the coefficient of the interaction effect between past and future (the coefficient of wI past ) equals (g 21 ct 1 )/(b 23 D) < 0. In other words, the model also implies that the interaction effect of past and future is indeed negative (Hypothesis 5). The model adds – besides being more general as well as explicit about assumptions and relevant variables – several distinct advantages to our intuitive arguments and hypotheses on transaction management. First, it allows us to specify additional hypotheses, which would have been difficult to derive on the basis of intuitive reasoning alone. Second, the formal model directly implies the nonlinear statistical model on the basis of which the hypotheses need to be tested (as opposed to just adding all indicators into a linear regression analysis). Third, the formal model allows indicators to enter the statistical analysis more than once. For instance, the volume of the transaction serves as an indicator of the damage potential and as an indicator of the opportunism potential. Using standard linear regression analysis, such specifications are impossible. Finally, as we briefly elaborate in our discussion section, the model is potentially useful for two-party relationships other than the specific buyer-supplier relationships we consider here. As long as assuming a similar underlying structure of the relationship is reasonable, the model provides the building blocks for subsequent analysis.
DATA AND MEASUREMENT Sample “The External Management of Automation 1995” (MAT95) is a large-scale survey on the purchase of IT-products by Dutch SMEs (5–200 employees; Batenburg, 1997a; Batenburg & Raub, 1995). A reason for a survey on IT-purchases of SMEs was that these buyers typically lack expertise and resources for the inhouse production of such products. Hence, we can neglect the make or buydecision and assume the transaction as exogenously given. In fact, according to one of the questions in MAT95, less than 5% of the transactions (46 out of 971)
Contacts and Contracts
155
involve IT-products that could have been produced easily or very easily by the buyer. The sampling frame was a business-to-business database of Dutch SMEs that contained information about the characteristics of these SMEs with respect to automation. The database is known to be far more up to date and reliable than the often used database of the Chamber of Industry and Commerce. It is owned and developed by Directview, a Dutch firm specialized in IT marketing data of Dutch organizations. About 80% of all Dutch firms with more than five employees are included in the database. The database can be considered to be representative for the Dutch population of SMEs (see Batenburg, 1997a). Three criteria were used for stratification. First, the sample was stratified according to the number of IT-specialists employed by the firm. Three groups were distinguished: firms with no specialist, firms that had only part-time specialists, and firms with one or more full-time specialists. Second, the strength of inter-firm relations within certain sectors of industry was determined by judgements of 28 business experts. Their judgements were based on how often firms meet informally and how many activities within the sector were organized to bring firms together. Using these expert judgements, sectors were divided in three groups: sectors with weak, medium, and strong inter-firm relations. The third stratification criterion was the type of IT-products bought by a firm. This criterion distinguished four groups of products: standard hardware, complex hardware, standard software, and complex software. These three stratification criteria were used because they represent three important theoretical dimensions. The expertise of the buyer and the complexity of the transaction are indicators of monitoring problems, while contacts between buyers represent “network embeddedness” (see Buskens, 2002, ch. 5 for an analysis of the effects of network embeddedness). The three stratification criteria resulted in a sampling design with 36 (3 × 3 × 4) cells. Randomization procedures for sampling transactions were used until at least 15 cases were collected for each cell. Key informants of buying firms were first briefly interviewed by a structured Computer Assisted Telephone Interview (CATI). In the CATI-interview, cooperation was asked from an employee responsible for automation in the firm. Most of the key informants were IT-managers of the buying firm. The CATI-interview was then used to randomly select a particular IT-investment the firm had made in the past, in order to define beforehand on which transaction the main questionnaire would focus. More precisely, the transaction was selected randomly from all the IT-investments of the firm in the previous five years that met the third stratification criterion (type of IT-product) and on which the respondent was well informed. Usually, the respondents were involved themselves with and often responsible for the purchase.
156
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
Following this sampling procedure, a main sample of 547 IT-transactions was obtained. Subsequently, the data set was extended with an additional sample. This additional sample was collected in order to obtain more observations on innovative and complex IT-products. Transactions were sampled from SMEs in sectors that typically use such products. Using judgements of IT-market researchers and figures from Statistics Netherlands, five such sectors were identified (food and metal industry, transport equipment, wholesale trade, and road transport). The additional sample was stratified using only the criterion related to the IT-specialists in the buyer’s firm. Complex transactions are assumed to be associated with a higher opportunism potential. Therefore, we include both samples in our analyses. Note that, in contrast with the first stratified sample, the additional sample is not representative for Dutch SMEs. Another 241 questionnaires were collected within this additional sample. About 25% (463 out of 1,798) of the firms contacted turned out not to be suitable for our purposes, either because there were no suitable respondents, no independent IT-investments, or no IT-products used in the firm, or because the firm had ceased to exist, was too large, or too small. Given willingness to cooperate, a member of the fieldwork team visited the respondent with the main questionnaire on a convenient date and time at the site of the firm. Respondents were asked to fill out a questionnaire regarding the purchase of the agreed upon IT-product. From the main sample and the additional sample, data are obtained from 788 (547 + 241) IT-buying firms. About 25% (183 out of 788) of the respondents were willing to fill out a second questionnaire regarding the purchase of a different IT-product, in most cases from a different supplier. In these cases, another questionnaire was left at the site of the firm and returned by mail. In total, the data set thus consists of 971 (547 + 241 + 183) transactions, of which 183 are second transactions from the same buyer. In about 15% (132 out of 788) of the cases, respondents were willing to participate but did not agree with a visit. Questionnaires were then sent to them by mail. The bulk of the questionnaires were filled out between January and June 1995. For 28 transactions, the name of the supplier is unknown. The remaining 943 transactions were furnished by 602 different suppliers. On average, a single supplier is involved with about 1.5 transactions in the data. Four large suppliers occur more frequently: IBM (30 transactions), Baan (18), MAI (15), and Raet (13). The average response rate to the CATI-interview was 67% (902 out of 1,335). Multiplied with the field response rate of 87% (788 out of 902), the total response rate equaled 59% (788 out of 1,335). This is a high response rate in comparison with other surveys among organizations (cf. Kalleberg et al., 1996, chs. 1 and 2). Non-response analysis showed that the response group is not biased on crucial firm characteristics such as size, industry or region. In addition, we know from a question in the CATI-interview that firms in our sample do not differ from firms
Contacts and Contracts
157
refusing to fill out the main questionnaire in their general satisfaction with ITsuppliers. Hence, it is unlikely that we have oversampled firms with either untypically many or untypically few problems with their IT-suppliers (Batenburg, 1997b).
Measurement Next, we describe the questions in the survey that were used to operationalize the theoretical concepts as introduced in the previous sections. In principle, one could try to find indicators for each of the parameters in the theoretical model. For tractability we restrict ourselves to indicators for the opportunism potential (O), the marginal costs of management (c), the shadow of the future (w), the shadow of the past (Ipast ), and the damage potential (D). As indicators for the opportunism potential, we use answers to survey questions about the financial volume of the transactions (volume) and about monitoring problems (monitoring problems). Hence, we neglect that mutual relation specific investments, indicated by the shadow of the past, may reduce the opportunism potential. As indicators for the damage potential (D), we again use answers to survey questions about the financial volume of the transaction (volume), about replacement costs (replacement costs), the importance of the product for the profit of the buyer’s firm (importance for profitability), and the importance of the durability of the product for the buyer (importance of durability). Our indicators for the marginal costs of management are the existence of standardized procedures in contracting (standardized procedures) and the availability of own legal expertise (legal expertise). The expectation w of future business with the same supplier is measured by asking the respondent for an estimate of the volume and frequency of new and additional transactions with the supplier (future). The shadow of the past was measured by an indicator variable, equal to 1 if buyer and supplier had done business before (past). More detailed information about frequency (see Gulati, 1995b who uses such a variable) and volume of past transactions with the supplier and the buyer’s satisfaction with these past transactions are available but do not add much statistically; the important difference lies between having past experiences or not (see below). Half of the firms had a history with the supplier, with an average length of 6.3 years (excluding those without past experience). In some of the analyses, we present two control variables, the size of the supplier’s firm (size supplier) and of the buyer’s firm (size buyer). Note that this is one way of controlling, albeit roughly, for the internal communication function of contracts if one is willing to assume that the need for internal communication increases in the size of the firm. Other controls, such as the type of industry and characteristics of the respondent, were included in the analyses as well. To avoid cluttering up the tables with a lot of
158
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
controls having no substantial effects on the results, we incorporate only the size of the firms explicitly. The subsection on stability of results provides further details. As the dependent variable (management) representing investments in contractual ex ante management and, hence, transaction costs actually associated with purchasing the product, we used a weighted average of various indicators. First, we included the number of person-days of employees of the buyer that were spent on negotiating with the supplier and drafting the contract, the number of departments of the buyer involved in negotiations with the supplier, the use of external legal advisors, and whether the contract was mainly a standard or a tailor made contract. Second, the questionnaire contained a list of 24 financial and legal clauses typically included in contracts for IT purchases as well as a list of 24 technical specifications. For each financial and legal issue, respondents were asked to specify how extensively it was addressed during the negotiations and whether it was arranged verbally or written down in a contract. For each of the technical specifications, respondents were asked how extensively it was addressed in the contract. Note that about 65% (625 out of 971) of the contracts are standard contracts or modified versions of standard contracts (e.g. Berkvens et al., 1991). However, such standard contracts for IT transactions are typically adapted by the users in a flexible way and, in fact, are often provided in a format (e.g. electronically) that facilitates “fine tuning” by the contracting parties. Hence, for the IT transactions considered here, the use of standard contracts does not preclude transaction or relationship specific contractual management. For a more detailed description of the indicators and the construction of variables, we refer to Appendix B. Obviously, one should appreciate the retrospective nature of our data as well as the fact that the data are collected via the buyer. To minimize potential bias, survey questions focused, wherever possible, not on attitudes of the respondent but on the respondent’s actual behavior and knowledge about specific characteristics of the product, the supplier, the buyer-supplier relation, negotiations, and the content of the contract. Table 1 presents an overview of the variables. The scale of most of the variables is meaningless, since most variables are either scores on a five-point scale (like future), or weighted averages of several questions in the survey (like replacement costs). To get a feel for the data, we mention that the average transaction involved a firm of about 80 employees buying a product worth roughly 50,000 U.S.$. Negotiating and contracting took the buyer about five person-days and involved 2 divisions of the buyer’s firm. On average, our respondents have been working at their firm since 1985. Since the average transaction took place around 1992, respondents have an average history at their firm of about seven years prior to the transaction. About two thirds of the respondents stated that the transaction was of “great” or “very great” importance for their IT-situation. The bivariate correlations between the variables are displayed in Appendix C.
Contacts and Contracts
159
Table 1. Overview of Variables and Descriptive Statistics. Variable name
Dependent Variable Management
Description
St. dev.
Min.
Max.
1
−2.29
3.87
1.20
−2.08
1.25
0
1
−2.31
2.90
955
0
1
−1.82
2.35
960
0
1
−3.98
2.03
963
0
1
−2.84
2.52
Standardized contracting procedures of buyerc Legal expertise of buyerd
920
2.49
1.17
1
5
964
0.20
0.40
0
1
Buyer and supplier have done business befored Probability of future business as expected by buyer before transactionc
964
0.50
0.50
0
1
950
2.79
1.38
1
5
Number of employees buyerb Number of employees supplierb
949
3.65
1.04
0
8.70
952
2.95
1.16
0.92
4.32
Total investment in management by buyera
Opportunism Potential/Damage Potential Volume Financial volume of the transaction in 100,000 HFLb Monitoring problems Monitoring problems of buyera Replacement costs Replacement costs for buyera Importance of durability Importance of durability of IT-product for buyera Importance for profitability Importance of IT-product for buyer’s profitabilitya Marginal Costs of Management Standardized procedures
Legal expertise Dyadic Embeddedness Past
Future
Control Variables Size buyer Size supplier
Number of cases
Mean
964
0
956
−0.76
964
Note: See Appendix A for details on indicators (original question formulation and answer categories) and variable construction. a Standardized factor score. b Natural log. c Five point scale. d Dummy, 1 = yes.
160
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
Note that our claim that buyers with bad experiences tend to try to find other suppliers seems to be supported by the data. Only 3% (14 out of 479) of the buyers who had done business with their supplier before was dissatisfied with these former transactions but nevertheless continued to do business with that supplier. These cases were excluded from the analyses. Either including or excluding these cases does not make a substantial difference for the results. Finally, our data also support the assumption that past investments in management are useful in subsequent transactions to a certain extent. In 135 out of 479 transactions where the buyer had already done business with the supplier, the contract used for the focal transaction was a more or less adapted version of a contract for a previous purchase of the buyer from this supplier.
STATISTICAL MODEL AND RESULTS We present our results using two types of regression analysis. As a direct test of our formal model, nonlinear regression is the most appropriate kind of analysis. Additionally, we present several OLS regressions, allowing for a more robust and elaborate way of testing our hypotheses. Our theorem shows that optimal management can be characterized by (see the formal model specification section for notation): −1 c(1 − wg 1 ) m opt = (1 − g 1 I past ) −b 0 + b 1 O + b 2 F − g 0 I past . b3D We estimate this model in two ways. First, we use a linear Taylor expansion for −1 F , which ensures that we are left with a model that can be estimated using nonlinear least squares (see, e.g. Greene, 1993, ch. 10). The 176 transactions in the data set that were second questionnaires filled out by the same respondent, but dealing with a different IT-transaction, are included in the analyses. Missings were deleted listwise. Excluding the second cases or using pairwise deletion has no substantial influence on the estimation results. Table 2 summarizes the estimation results for the model Management = (1 − g 1 Past) 1 + 2 Volume + 3 Monitoring +
(4 Procedures + 5 Expertise)(1 + 6 Future) 1 + 7 Switching + 8 Durability + 9 Profit + 10 Volume
− g 0 Past. Note that both the variable past and volume occur twice in the table. Past because it must allow for an estimation of both g0 and g1 , and volume because it
Contacts and Contracts
161
Table 2. Standardized Coefficients from the Non-Linear Least Squares Regression on Management. Independent Variables
Hypothesis
Opportunism Potential Volume Monitoring problems
2 3
+ +
0.38** 0.12**
6.81 3.70
Damage Potential Replacement costsb Importance of durabilityb Importance for profitabilityb Volumeb
7 8 9 10
– – – –
−0.12** −0.04 −0.12** 0.04
4.81 1.73 4.51 1.17
Marginal Costs of Management Standardized procedures Legal expertise
4 5
+ +
0.05 0.20**
1.84 2.86
Dyadic Embeddedness Pastc (1 = yes) Pastc (1 = yes) Future
(−)g0 (−)g1 6
– – +
−0.17** −0.10 −0.02
2.97 1.25 1.35
1
?
−0.65** 0.38**
10.37
Constant Adjusted R2
Coefficient
|t-value|a
Coefficient
Note: N = 895 transactions. Dummy-variables are not standardized. a t-values are asymptotic approximations. b All indicators for the damage potential are expected to have positive effects on management (e.g. the larger the volume, the more management). Their sign is reversed because they appear in the denominator of the estimated model. c We assumed g to be positive, which implies that we assume the coefficient of Past (−g ) to be 0 0 negative. The same holds for g1 . * p < 0.05; ∗∗ p < 0.01 (two-tailed tests).
was taken as an indicator for the opportunism potential and for the damage potential. Other than in standard regression, the nonlinear structure of the model allows for such a double inclusion of independent variables. The results do not support the assertion that initial management carries over (the g1 -coefficient is in the hypothesized direction but not significant), but they do support that a first transaction creates costs for set-up management (the g0 -coefficient is significant). Note that the effect of the volume of the transaction is significant only in the case where it represented the opportunism potential and not where it represented the damage potential. This suggests that the volume of a transaction is a better indicator for the opportunism potential than it is for the damage potential.
162
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
Most estimates are consistent with the hypothesized relationships between the independent variables and management. There are significant positive effects on management of replacement costs, of the importance for profitability attached to the product, of monitoring problems, and of the volume of the transaction. The existence of a positive past with the same supplier leads to a smaller investment in management. The conclusion with respect to our hypothesis on the interaction of shadow of the past and future cannot be directly derived from Table 2. However, the value of the coefficient of past and future in our estimated model mentioned above equals −g 1 (4 Stand. Procedures + 5 Legal Expertise)6 . 1 + 7 Switching Costs + 8 Importance of Durability + 9 Importance of Profit + 10 Volume
Calculating the value of this expression using our estimated coefficients shows that it is negative for all the cases in the data. Hence, the coefficient of the interaction of past and future is negative for all the cases in the data, which supports our hypotheses. In fact, this also shows that our intuitive arguments regarding the interaction of past and future involve simplifications. The formal model is not just the mathematical equivalent of our more intuitive arguments. An implication of our formal model is that the interaction effect itself turns out to be dependent on the variables mentioned in the above equation. Hence, the model renders conditions under which our arguments regarding the negative interaction of past and future are less likely to be supported, namely, precisely when the values of the variables in the expression above are such that the expression itself has a value close to zero. Of course, the results of the nonlinear regression should be considered with some reservation, since they rely on a strict belief in the functional form of the theoretical relationships. Therefore, we ran several OLS regressions, of which we consider a representative one below. Again, the results are consistent with our hypotheses to a large extent. Note in particular the negative coefficient of the interaction of past and future. As hypothesized, the effect of the shadow of the future is larger if there was a shared (positive) past. Different ways to assess this difference lead to similar conclusions. For instance, separate (OLS) analyses for cases with and without a past show a nonsignificant effect of future if no past exists (0.04, t = 1.17) versus a significant effect of future if a past does exist (−0.12, t = −3.13). Additionally, bootstrapping (1,000 replications) of the coefficient of the interaction effect leads to a (bias corrected) 99% confidence interval [−0.26, −0.04]. Excluding the interaction term of past and future reveals a significant effect of past (−0.09, t = −3.13) and a nonsignificant effect of future (−0.03, t = −1.19).
Contacts and Contracts
163
Table 3. Standardized Coefficients from the Ordinary Least Squares Regression on Management. Independent Variables
Hypothesis
Coefficient
|t-value|
Opportunism and Damage Potential Volume Monitoring problems Replacement costs Importance of durability Importance of profitability
+ + + + +
0.33** 0.10** 0.12** 0.10** 0.13**
9.42 3.16 3.58 3.42 4.19
Marginal Costs of Management Standardized procedures Legal expertise
+ +
0.05 0.06*
1.91 2.25
Dyadic Embeddedness Past (1 = yes) Future Past × Future
– ? –
−0.09** 0.05 −0.12**
3.12 1.28 3.33
Control Variables Size supplier Size buyer
? ?
0.06* 0.03
2.20 1.13
?
−0.04 0.39**
0.25
Constant Adjusted R2
Note: N = 895 transactions. Dummy-variables are not standardized. ∗ p < 0.05; ∗∗ p < 0.01 (two-tailed tests).
As stated above, our dependent variable (management) is a weighted average of several underlying variables. Its scale is therefore meaningless. To get an idea about the relative magnitude of the estimated effects, we compare the size of the coefficients in Tables 3 and 4. Clearly, the volume of the transaction stands out as the variable with the largest effect on management. However, it should be noted that the sizes of the effects of past and future are comparable to the effects of variables representing more standard transaction characteristics, like replacement costs or monitoring problems. This suggests that the dyadic embeddedness of transactions is indeed a factor to be reckoned with in the management of transactions.
STABILITY OF RESULTS We aim to show that, under different reasonable implementations of the data, the coefficients of the independent variables are stable. In particular, we show that
164
Table 4. Standardized Coefficients from Different Regressions on Management. Independent Variables
Hyp.
Model 2 Coeff. (|z-value|)
Model 3 Coeff. (|t-value|)
Opportunism and Damage Potential Volume Monitoring problems Replacement costs Importance of durability Importance of profitability
+ + + + +
0.33** 0.09** 0.12** 0.10** 0.13**
(9.25) (2.87) (3.73) (3.23) (4.36)
0.34** 0.11** 0.12** 0.10** 0.12**
Marginal Costs of Management Standardized procedures Legal expertise
+ +
0.05 0.05
(1.77) (1.89)
0.06* (2.16) 0.13* (2.00)
0.05 (1.31) 0.08* (2.06)
Dyadic Embeddedness Past (1 = yes) Future Past × Future
– ? –
−0.09** (3.08) 0.04 (0.99) −0.12** (3.02)
−0.19** (3.37) 0.04 (1.56) −0.12** (2.93)
−0.06 (1.35) 0.04 (0.66) −0.12* (2.11)
Control Variables Size supplier Size buyer Constant
? ? ?
0.07* (2.46) 0.03 (1.02) −0.02 (0.15)
0.05* (2.42) 0.03 (0.97) −0.04 (0.29)
0.08 0.04 −0.17
Variance of Management Volume
+
N Adjusted (variance weighted) R2
(9.57) (3.60) (3.62) (3.56) (4.03)
0.32** 0.12** 0.10* 0.15** 0.12**
(6.14) (2.60) (2.09) (3.81) (2.72)
(1.85) (1.06) (0.80)
Model 4 Coeff. (|t-value|)
Model 5 Coeff. (|t-value|)
Model 6 Coeff. (|t-value|)
0.34** 0.07 0.12* 0.04 0.16**
(6.90) (1.69) (2.54) (0.93) (3.36)
0.33** 0.17** 0.16** 0.13* 0.06
(5.99) (3.12) (2.72) (2.43) (1.13)
0.26** 0.06 0.10* 0.09* 0.15**
0.06 0.04
(1.51) (1.09)
0.07 0.03
(1.43) (0.62)
0.05 (1.52) 0.08* (2.26)
−0.10* (2.46) 0.04 (0.83) −0.12* (2.36)
−0.02 0.06 −0.08
(0.39) (0.81) (1.16)
−0.12** (3.28) 0.04 (0.91) −0.15** (3.15)
0.03 0.00 −0.05
(0.64) (0.08) (0.25)
0.09* (2.20) 0.05 (1.22) −0.06 (0.29)
0.05 0.02 0.12
(1.33) (0.43) (0.58)
(5.64) (1.48) (2.48) (2.34) (3.82)
0.19** (4.91) 895 0.40
895 0.40
434 0.42
461 0.35
323 0.36
572 0.30
Note: Dummy-variables are not standardized. Model 1 = Regression with standard errors (Huber-)corrected for clustering on supplier (Huber, 1967). Model 2 = Regression with heterogeneous variance determined by volume. Model 3 = OLS Regression on hardware products only. Model 4 = OLS Regression on software products only. Model 5 = OLS Regression on standard hard- and software products only. Model 6 = OLS Regression on complex hard- and software products only. ∗ p < 0.05; ∗∗ p < 0.01 (two-tailed tests).
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
Model 1 Coeff. (|t-value|)
Contacts and Contracts
165
the coefficient of the interaction effect of past and future remains significant and negative across different kinds of analyses. The results of analyses we considered most important are reported in Table 4. To save space, we only report the relevant statistics for other analyses (see Batenburg et al., 2000 for details).
Alternatives: Statistical Models and Additional Control Variables Several extensions to our regressions are displayed in Table 4. Model 1 addresses the potential problem that the regression results might be influenced by the fact that several buyers had bought products from the same supplier. If this gives rise to excessive “clustering” in the data, we run the risk of finding significant relations where in fact there are none (Huber, 1967). The second model is based on the idea that one should also consider the variance in investments in ex ante management (instead of only the effects of several variables on the average amount of investment). In particular, it seems reasonable to assume that the variance in management is necessarily larger for transactions with a larger volume. The third extension we consider is running the analyses separately for hardware and software as well as for standard and complex IT-products (models 3–6). This is one way of testing whether the kind of product being assessed has an influence on the stability of our results. Several features of the results for these additional models are noteworthy. First of all, we can conclude that under these different implementations of the analyses, the results do seem rather stable. The size of the coefficients is similar and, by and large, t-values are of a similar magnitude. Moreover, we indeed find evidence for heterogeneity in the variance: transactions with a larger volume have a larger (log of the) variance (0.19, z = 4.91), but it does not seem to affect the parameter estimates much. In particular, it does not affect the significant effect of the interaction of past and future (−0.12, z = −2.93). The largest differences are found when we discriminate between standard and complex IT-products (models 5 and 6 in Table 4). When we consider exclusively standard IT-products (model 5), we only find significant effects of the variables representing the opportunism and damage potential. Indeed, the interaction effect of past and future is no longer significant here and somewhat smaller in magnitude although it is in the expected direction and has a confidence interval [−0.20, 0.05] containing the previously found value (−0.12). As a further extension of the basic OLS regression analysis in Table 3, we included various additional control variables, none of which revealed significant effects on the amount of management. We considered possible effects of different sectors by categorizing buyers by their (single digit) SIC-code (F-value 1.70,
166
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
df = 7, p = 0.11). This is close to significant. Careful inspection of the data, however, shows that differences – if they are there – are mainly due to five cases in the data that represent governmental firms (water and energy suppliers). Removing those from the data leads to a sector effect that is more clearly not significant (F-value 1.38, df = 6, p = 0.22). Additionally, we controlled for characteristics of the respondent (some evidence for such effects was found in Rooks et al., 2000). We used the number of years respondents had been working for the buyer firm (−0.06, t = −1.42), the number of years of experience with IT of respondents in the buyer firm (0.004, t = 0.11), age (−0.04, t = −1.20), and education (coded in seven categories) of respondents (−0.02, t = −0.58). Post-hoc, we also investigated which of the independent variables in the OLS regression have an effect on the variance in management. Except for the volume of the transaction (as mentioned above and in Table 4), we find effects for two variables. The variance in management increases with increasing importance for profit (0.17, z = 3.13) and there is some evidence for a negative effect of the age of the respondent (−0.01, z = −1.84). Thus, as the importance for the profit of the firm increases, respondents start to differ with respect to the amount of management they apply. Similarly, older respondents seem to be “more alike” in the amount of management they choose.
Alternatives: Search and Selection As a Dependent Variable An objection against our choice of the dependent variable is that it does not include search and selection efforts as part of the ex ante management (see Buskens et al., 2003 in this volume). It might occur that extensive search for a suitable supplier can substitute for management of the transaction in a later stadium. For instance, it could be that a buyer invests considerable effort in searching for an adequate and reliable partner as well as for a product with a good enough price and is therefore willing to invest less in writing an extensive contract. Since we do not consider the search for a supplier in our model, this might affect our results: cases in the data where we conclude that ex ante management is virtually absent or small could actually be cases where large investments in search and selection efforts were made. However, these substitution effects do not occur. Buyers who invest large amounts of time and effort in search and selection, also invest large amounts of time and effort in contractual ex ante management. And, buyers who invest small amounts of time and effort in search, also invest small amounts of time and effort in contractual ex ante management. Our data support this claim in several ways. Search investment was measured as a factor score of the number of suppliers and products considered in the search and selection process, the number of elicited tenders, the number of person-days involved in searching and selecting supplier
Contacts and Contracts
167
and product, the relative number of divisions involved in the search and selection effort, the number of other (potential) buyers that were asked for information, and the number of different ways in which information was collected (through exhibitions, yellow pages, etc.). First, a factor analysis of separate management investments including search (search investment, number of person-days and departments involved in negotiating, whether external advisors were used, whether the contract was tailor made, number of clauses that were orally treated in negotiations, number of clauses that were written down in the contract, and the number of technical specifications) shows a strong single factor with positive weights for all variables. Second, using search investment rather than our variable management as the dependent variable in an OLS regression with the same independent variables as in Table 3, we find a significant negative effect of the interaction of past and future, and coefficients that are similar to those in Table 3 to a large extent. To be precise, if we disregard the coefficient of the shadow of the future, we cannot reject the hypothesis that the coefficients of the two analyses are proportional (p = 0.86). In other words, investments in search and selection would be an alternative indicator to incorporate in what we refer to as ex ante management. Excluding it, as we do, does not affect the results in a substantial manner.
Alternatives: Operationalizations of Independent Variables The emphasis of our contribution is on the effects of having positive past experiences and expectations of future transactions. We mentioned that the effects of having positive past experiences on management are related to the difference between having a past or not, and not so much to differences in the kind of past one had with that same supplier. As noted earlier, “having had past transactions with the same supplier” is almost identical in our data to “having had positive past transactions with the same supplier” since there are only a few cases (3%) in which the buyer was unsatisfied with these past transactions. There also appears to be little variation in the volume of previous transactions. For instance, from the 479 buyers who have had previous transactions with the same supplier, about 75% of these previous transactions are of limited or moderate volume. We ran three separate OLS regressions to find out whether additional effects of the kind of past with the supplier exist. For each of these regressions, we used the same variables as in Table 3, but added an interaction effect of having a past or not with “frequency of past transactions,” “satisfaction with past transactions,” and “volume of past transactions.” None of the coefficients of the interaction effects approached significance (p = 0.83, p = 0.30, p = 0.33). If we run three separate OLS regressions on the cases with positive past transactions with the same
168
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
supplier only, we see a similar result. No effect of the quality of the past exists ( p = 0.85, p = 0.30, p = 0.25). In other words, buyers who do business with suppliers they have dealt with before do invest less in management. But, whether these buyers have had frequent or less frequent transactions before, were moderately or highly satisfied, or have had previous transactions with moderate or high volume, does not have an impact on their transaction management. This is in line with the result from our nonlinear regression analysis that the first transaction creates costs for set-up management, while there is no significant carry over effect of management. A related issue is whether our measurement of the expectations of future transactions is an adequate one. Throughout we consider expectations of future business to represent something like “the probability that buyer and supplier will meet again,” and treat it as if it is exogenous. Surely, this is not an adequate representation of reality. Expectations of future business may also depend on whether buyer and supplier were satisfied with previous transactions and on whether some kind of dependency exists between buyer and supplier. An OLS regression indeed shows that the expectation of future transactions depends on the quality of past transactions ( p < 0.001) and the dependency of the buyer on the supplier ( p < 0.001). However, explained variance is 0.18, which suggests that our measurement of the expectations of future business is certainly not determined only by these two variables. Moreover, estimating the coefficients of the OLS regression of Table 3 with both the quality of past transactions and the dependency of buyer and supplier shows once again that the interaction effect of past and future remains negative and significant ( p < 0.001). As a final robustness check, we recalculated all variables that are factor scores in our original analyses. Instead of using the factor scores, we reanalyzed our OLS regressions using simple addition of the separate indicators. So, for instance, instead of calculating monitoring problems as Monitoring Problems = 0.20[complexity hardware] + 0.21[complexity software] + 0.31[complexity services] + 0.42[quality] + 0.42[tenders] + 0.47[other products] + 0.45[price-quality] − 0.22[experience] − 0.08[expertise] − 0.03[“make” possible], we instead calculated monitoring problems using +1 for indicators with positive weights and −1 for indicators with negative weights. The resulting OLS regression on the basis of these new variables shows similar results to the one we reported in our original analyses. The status of the significance of the effects of independent
Contacts and Contracts
169
variables in Table 3 remains unchanged. For instance, the effect of the interaction between past and future is negative in both cases (p = 0.046 in the analysis based on the simplified factor scores).
CONCLUSION AND DISCUSSION We provided a theoretical and empirical analysis of the extent to which ITtransactions are managed, based on data of 971 IT-transactions between Dutch SMEs and their IT-suppliers. Our analysis considered the extent to which effort invested in writing and negotiating a contract can be explained by the opportunism potential, the damage potential, management costs, and the dyadic embeddedness associated with that transaction. We investigated when and how trust, like trust based on norms of reciprocity and conditionally cooperative behavior, can be used as a substitute for costly contractual governance of economic transactions. We also developed hypotheses on how contractual and non-contractual governance depend on the interplay of economic and social conditions affecting the problem potential associated with transactions. Thus, we tried to integrate two sociological insights with a rational choice approach: the idea of non-contractual complements for contractual governance and the idea that the embeddedness of transactions affects the governance of transactions. We showed how non-contractual governance and reciprocity can be a result of incentive-driven behavior and how embeddedness affects the incentives for relying on contractual or, respectively, non-contractual governance. The data support our hypotheses to a large extent. First, management increases with increasing opportunism potential and damage potential. Clearly, the volume of the transaction seems to be the main determinant of the extent of management of the transaction. We also found that the dyadic embeddedness of a transaction has an effect on management of an order of magnitude similar to the other effects of indicators of the opportunism potential and damage potential. This supports the argument that the social embeddedness of transactions is indeed a factor to be reckoned with in the analysis of trust between firms. Specifically, both the shadow of the past and its interaction with the shadow of the future have a negative effect on management. Firms who have done business with each other before invest less in management and, in particular, invest less in management the larger the likelihood of future interaction. The overall effect of the shadow of the future on management is close to zero. For those cases where no shared past exists, the incentive to invest less in contracting because of the feasibility of conditional cooperation seems to be counterbalanced by the incentive to invest more in contracting because of the need for set-up investments. Note that similar analyses using data sets based on a different design, but containing variables like
170
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
the ones used here, provide considerable support for the validity of our findings (see Blumberg, 2001a; Rooks et al., 2000). Hardly any of the cases in the data consists of business partners with a negative shared past. This suggests that searching for another partner is the most likely response to a problematic transaction. Apparently, firms anticipate more profit by finding another partner than by writing longer and better contracts. Hence, transaction management through search and selection of reliable and trustworthy partners seems to be a fruitful avenue for future research (see Blumberg, 2001b; Buskens et al., 2003 in this volume; Gulati & Gargiulo, 1999). It is interesting that “economic behavior” like the behavior of buyers dealing with their suppliers depends on relational aspects even if we abstract from matters that should facilitate to highlight effects of “social embeddedness.” For instance, it seems plausible that the content of contractual agreements, which may reflect shared conventions or “definitions of the situation,” is affected by “social” as opposed to purely economic forces. Instead we came up with hypotheses on and empirical support for the effects of social embeddedness, abstracting completely from content, and focusing exclusively on the amount of investments in contractual planning. Moreover, one might suspect that the network of relations of buyer and supplier with other business partners affects their contractual behavior (see, e.g. Burt, 1992; Podolny, 1993 for a general perspective; Raub & Weesie, 1990 for a game theoretic model; Buskens, 2002 for an empirical analysis based on our data; and Stuart, 2003 in this volume). Again, we showed that effects of embeddedness are to be expected and empirically confirmed to exist even if we abstract from arguments based on network embeddedness. Finally, we could even abstract from non-economic personal ties and focus exclusively on prior and expected future business contacts for highlighting embeddedness effects. One of the most rigorous assumptions we made in our theoretical model was that we assumed that the buyer decides on the management of the transaction. As we argued before, we feel this is a reasonable approximation for the Dutch IT-market at this point in time. However, it is indeed only an approximation and for other markets it might be less appropriate. A related argument against our model is that the inherent strategic nature of the situation is now somewhat hidden. We do assume effects of the opportunism potential of the supplier through monitoring problems (the more difficult to monitor, the more likely opportunistic behavior of the supplier), but this was operationalized in a parametric rather than an interdependent manner. Extensions of our model will most likely relax both these assumptions. A first step in this direction would be to explicitly model the behavior of buyer and supplier in terms of Trust Games (cf. Buskens, 2002; Snijders, 1996). For a game theoretical model of investments in ex ante management of transactions, see Raub and Snijders (2001). Note that although we chose to consider the investment in negotiating and
Contacts and Contracts
171
writing a contract, the data set also allows analyzing different stages of the governance of transactions. For instance, the data permit the analysis of performance characteristics and ex post management such as type and seriousness of conflicts that emerge ex post and different modes of contractual and non-contractual conflict regulation. In addition, although we considered the extent to which investments were made, as opposed to Williamson’s (1985, 1996) choice of governance structures, it goes without saying that the data allow for both types of analyses. Applying our analysis to other inter-firm relations seems straightforward. Our theory and hypotheses obviously apply to buyer-supplier relations involving the purchase of other types of products or components. They likewise apply to strategic alliances such as R&D-alliances (e.g. Gulati, 1995b; Parkhe, 1993; and the contributions by Gulati & Wang, 2003 as well as Stuart, 2003 in this volume). However, we would like to close with a more speculative remark on our models. Researchers have sometimes argued in favor of a unified analysis of dyadic relations of different kinds (see Becker et al., 1977; Ben-Porath, 1980; Raub & Weesie, 2000 for a more systematic elaboration of this idea). Such an analysis would consider inter-firm relations, households, and also employment relations as empirical realizations of the same underlying principle (in this volume, Neckerman & Fernandez, 2003 focus on the employment relation from a similar perspective). While such an integrated analysis remains a program that still has to be implemented, note that the models presented here offer useful building blocks for this more ambitious project. For example, our variables have relatively straightforward equivalents if one would consider households. One could then likewise give meaning to concepts like damage potential (e.g. how bad would it be if the relationship broke up), opportunism potential (e.g. the attractiveness of the spouse on the “marriage market”), the costs of management (e.g. the investments in getting to know your spouse, friends of your spouse or visiting in-laws), the dyadic embeddedness (e.g. the duration of the relation and the likelihood of continuation of the relation), and, finally, actual investments in management of the relation (e.g. actual investments in getting to know your spouse and friends of your spouse, but also investments in financial and legal arrangements of household partners; see Treas, 1993). Our theoretical and statistical models applied here for an analysis of interorganizational relations would then become directly applicable for the analysis of seemingly completely different types of dyadic relations.
ACKNOWLEDGMENTS The order of authorship is alphabetical. Useful comments by Frits Tazelaar, Jeroen Weesie, Vincent Buskens, Ron Burt, Emmanuel Lazega, and seminar participants
172
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
at Utrecht University and the University of Chicago are gratefully acknowledged. This research was supported by a grant from the Netherlands Organization for Scientific Research (NWO; PGS 50–370) and from the NEVI Research Foundation (NRS) of the Dutch Association for Purchase Management (NEVI). Raub acknowledges support of the Netherlands Institute for Advanced Study in the Humanities and Social Sciences (NIAS), Wassenaar, Netherlands, while Snijders acknowledges support of the Royal Netherlands Academy of Arts and Sciences (KNAW). Direct correspondence to Werner Raub, Department of Sociology/ ICS, Utrecht University, Heidelberglaan 1, 3584 CS Utrecht, Netherlands (
[email protected]).
REFERENCES Arrow, K. J. (1974). The limits of organization. New York: Norton. Auer, J., & Harris, C. E. (1981). Computer contracts negotiation. New York: Van Nostrand. Axelrod, R. (1984). The evolution of cooperation. New York: Basic Books. Azariadis, C. (1987). Implicit contracts. In: J. Eatwell, M. Milgate & P. Newman (Eds), The New Palgrave: Allocation, Information and Markets (pp. 132–140). London: Macmillan. Baker, W. E., Faulkner, R. R., & Fisher, G. A. (1998). Hazards of the market: The continuity and dissolution of interorganizational market relationships. American Sociological Review, 63, 147– 177. Barber, B. (1983). The logic and limits of trust. New Brunswick, NJ: Rutgers University Press. Batenburg, R. S. (1997a). The external management of automation. Codebook of MAT95. ISCORE Paper 58. Department of Sociology, Utrecht University. Batenburg, R. S. (1997b). The external management of automation 1995. Fieldwork, response and non-response. ISCORE Paper 59. Department of Sociology, Utrecht University. Batenburg, R. S., & Raub, W. (1995). The external management of automation 1995 [data set]. Utrecht University. Batenburg, R. S., Raub, W., & Snijders, C. (2000). Contacts and contracts: Dyadic embeddedness and the contractual behavior of firms – Additional analyses. ISCORE Paper 166. Department of Sociology, Utrecht University. Available at http://www.fss.uu.nl/soc/iscore/ Beale, H., & Dugdale, T. (1975). Contracts between businessmen: Planning and the use of contractual remedies. British Journal of Law and Society, 2, 45–60 Becker, G. S., Landes, E. M., & Michael, R. T. (1977). An economic analysis of marital instability. Journal of Political Economy, 85, 1141–1187 Ben-Porath, Y. (1980). The F-connection: Families, friends, and firms and the organization of exchange. Population and Development Review, 6, 1–30 Berkvens, J. M. A., van Duijvenvoorde, G. P., & Grutters, C. A. F. M. (Eds) (1991). Automatiseringscontracten: Modellen voor de praktijk (Automation contracts: Models for practice). Deventer: Kluwer. Blumberg, B. F. (1998). Management von Technologiekooperationen: Partnersuche und vertragliche Planung. Wiesbaden: Gabler. Blumberg, B. F. (2001a). Cooperation contracts between embedded firms. Organization Studies, 22, 825–852.
Contacts and Contracts
173
Blumberg, B. F. (2001b). Efficient partner search: Embedded firms seeking co-operative partners. Journal of Mathematical Sociology, 25, 329–354. Brynjolfsson, E., & Hitt, L. M. (2000). Beyond computation: Information technology, organizational transformation and business performance. Journal of Economic Perspectives, 14, 23–48. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Burt, R. S., & Knez, M. (1995). Kinds of third-party effects on trust. Rationality and Society, 7, 255–292. Burt, R. S., & Knez, M. (1996). Trust and third-party gossip. In: R. M. Kramer & T. R. Tyler (Eds), Trust in Organizations. Frontiers of Theory and Research (pp. 68–89). Thousand Oaks, CA: Sage. Buskens, V. (2002). Social networks and trust. Boston: Kluwer. Buskens, V., Batenburg, R. S., & Weesie, J. (2003). Embedded partner selection in relations between firms. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 107–133). Oxford: Elsevier Science. Camerer, C., & Weigelt, K. (1988). Experimental tests of a sequential equilibrium reputation model. Econometrica, 56, 1–36. Coase, R. (1937). The nature of the firm. Economica, 4, 386–405. Coleman, J. S. (1987). Psychological structure and social structure in economic models. In: R. M. Hogarth & M. W. Reder (Eds), Rational Choice. The Contrast between Economics and Psychology (pp. 181–185). Chicago, IL: University of Chicago Press. Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, S95–S120. Coleman, J. S. (1990). Foundations of social theory. Cambridge, MA: Belknap Press. Dasgupta, P. (1988). Trust as a commodity. In: D. Gambetta (Ed.), Trust. Making and Breaking Cooperative Relations (pp. 49–72). Oxford: Blackwell. Durkheim, E. (1893). De la division du travail social. Paris: PUF. Ellickson, R. C. (1991). Order without Law. How neighbors settle disputes. Cambridge, MA: Harvard University Press. Gambetta, D. (1988). Can we trust trust? In: D. Gambetta (Ed.), Trust. Making and Breaking Cooperative Relations (pp. 213–237). Oxford: Blackwell. Gordon, R. J. (2000). Does the ‘new economy’ measure up to the great inventions of the past? Journal of Economic Perspectives, 14, 49–74. Granovetter, M. (1985). Economic action and social structure. The problem of embeddedness. American Journal of Sociology, 91, 481–510. Granovetter, M. (1992). Economic institutions as social constructions: A framework for analysis. Acta Sociologica, 35, 3–11. Greene, W. H. (1993). Econometric analysis. Upper Saddle River, NJ: Prentice Hall. Gulati, R. (1995a). Social structure and alliance formation patterns: A longitudinal study. Administrative Science Quarterly, 40, 619–652. Gulati, R. (1995b). Does familiarity breed trust? The implications of repeated ties for contractual choice in alliances. Academy of Management Journal, 38, 85–112. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104, 1439–1493. Gulati, R., & Wang, L. O. (2003). Size of the pie and share of the pie: Implications of network embeddedness and business relatedness for value creation and value appropriation in joint ventures. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 209–242). Oxford: Elsevier Science.
174
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
Hart, O. (1987). Incomplete contracts. In: J. Eatwell, M. Milgate & P. Newman (Eds), The New Palgrave: Allocation, Information and Markets (pp. 163–179). London: Macmillan. Heide, J. B., & Miner, A. S. (1992). The shadow of the future: Effects of anticipated interaction and frequency of contact on buyer-seller cooperation. Academy of Management Journal, 35, 265–291. Hollanders, H. (2000). ICT: De nieuwe motor van de economie (ICT: The new engine of the economy). Economisch Statistische Berichten, 85 (ESB-dossier ICT, arbeid en scholing), November 2, 2000, d4–d7. Huber, P. J. (1967). The behavior of maximum likelihood estimates under non-standard conditions. Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, 1, 221–233. Jacobs, D., & de Vos, A. (1992). Nederlands meest concurrerende sectoren (The most competitive sectors in The Netherlands). Economisch Statistische Berichten, 77, 148–151. Jones, C., Hesterly, W. S., & Borgatti, S. P. (1997). A general theory of network governance: Exchange conditions and social mechanisms. Academy of Management Review, 22, 911–945. Joskow, P. L. (1987). Contract duration and relation-specific investments: Empirical evidence from coal markets. American Economic Review, 77, 168–185. Kalleberg, A. L., Knoke, D., Marsden, P. V., & Spaeth, J. (1996). Organizations in America. Analyzing their structures and human resource practices. London: Sage. Kreps, D. M. (1990). Corporate culture and economic theory. In: K. A. Shepsle (Ed.), Perspectives on Positive Political Economy (pp. 90–143). Cambridge: Cambridge University Press. Larson, A. (1992). Network dyads in entrepreneurial settings: A study of the governance of exchange relationships. Administrative Science Quarterly, 37, 76–114. Lyons, B. R. (1994). Contracts and specific investments: An empirical test of transaction cost theory. Journal of Economics and Management Strategy, 3, 257–278. Macaulay, S. (1963). Non-contractual relations in business. American Sociological Review, 28, 55–66. Macneil, I. R. (1980). The new social contract. New Haven, CT: Yale University Press. Masten, S. E. (Ed.) (1996). Case studies in contracting and organization. New York: Oxford University Press. Milgrom, P., & Roberts, J. (1992). Economics, organization and management. Englewood Cliffs, NJ: Prentice-Hall. Milgrom, P., & Roberts, J. (1996). Economic theories of the firm: Past, present, and future. In: P. J. Buckley & J. Michie (Eds), Firms, Organizations, and Contracts (pp. 459–475). Oxford: Oxford University Press. Neckerman, K. M., & Fernandez, R. M. (2003). Keeping a Job: Network hiring and turnover in a retail bank. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 299–318). Oxford: Elsevier Science. Nohria, N., & Eccles, R. G. (Eds) (1992). Networks and organizations: Structure, form, and action. Boston, MA: Harvard Business School Press. Noordewier, Th. G., John, G., & Nevin, J. R. (1990). Performance outcomes of purchasing arrangements in industrial buyer-vendor relationships. Journal of Marketing, 54, 80–93. North, D. C. (1990). Institutions, institutional change and economic performance. Cambridge: Cambridge University Press. Oliner, S. D., & Sichel, D. E. (2000). The resurgence of growth in the late 1990s: Is information technology the story? Journal of Economic Perspectives, 14, 3–22. Oliver, A. L., & Ebers, M. (1998). Networking network studies: An analysis of conceptual configurations in the study of inter-organizational relationships. Organization Studies, 19, 549–583. Parkhe, A. (1993). Strategic alliance structuring. A game theoretic and transaction cost examination of interfirm cooperation. Academy of Management Journal, 36, 794–829.
Contacts and Contracts
175
Podolny, J. M. (1993). A status-based model of market competition. American Journal of Sociology, 98, 829–872. Polanyi, K. (1944). The great transformation. Boston, MA: Beacon Press. Prendergast, C. (1999). The provision of incentives in firms. Journal of Economic Literature, 37, 7–63. Rasmusen, E. (1994). Games and information. An introduction to game theory (2nd ed.). Cambridge, MA: Blackwell. Raub, W. (1996). Effects of temporal embeddedness on ex ante planning under incomplete information. ISCORE Paper 87. Department of Sociology, Utrecht University. Raub, W., & Snijders, C. (2001). A reluctant match: Models for the analysis of trust in durable two party relations. In: C. Castelfranchi & Yao-Hua Tan (Eds), Trust and Deception in Virtual Societies (pp. 195–220). Boston, MA: Kluwer. Raub, W., & Weesie, J. (1990). Reputation and efficiency in social interactions: An example of network effects. American Journal of Sociology, 96, 626–654. Raub, W., & Weesie, J. (2000). The management of durable relations: Introduction and outline of a research program. In: J. Weesie & W. Raub (Eds), The Management of Durable Relations. Theoretical and Empirical Models for Households and Organizations (pp. 1–32). Amsterdam: Thela Thesis. Riesewijk, B., & Warmerdam, J. (1986). Het slagen en falen van automatiseringsprojecten (Success and failure of automation projects). ITS, Nijmegen. Rooks, G., Raub, W., Selten, R., & Tazelaar, F. (2000). How inter-firm cooperation depends on social embeddedness: A vignette study. Acta Sociologica, 43, 123–137. Schellekens, R., Verschuur, J., & van Vuren, B. (2000). Trends in ICT 2000. Rotterdam: Ernst & Young ICT Leadership. Shelanski, H. A., & Klein, P. G. (1995). Empirical research in transaction cost economics: A review and assessment. Journal of Law, Economics, and Organization, 11, 335–361. Sitkin, S. B., Rousseau, D. M., Burt, R. S., & Camerer, C. (Eds) (1998). Special topic forum on trust in and between organizations. Academy of Management Review, 23, 384–620. Smelser, N. J., & Swedberg, R. (Eds) (1994). The handbook of economic sociology. New York: Russell Sage. Snijders, C. (1996). Trust and commitments. Amsterdam: Thesis Publishers. Statistics Netherlands [Centraal Bureau voor de Statistiek] (1998). ICT-markt in Nederland 1993–1996 (ICT-market in The Netherlands 1993–1996). Voorburg: EZ/Statistics Netherlands. Stuart, T. E. (2003). Governing strategic alliances. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 189–208). Oxford: Elsevier Science. Swedberg, R. (Ed.) (1993). Explorations in economic sociology. New York: Russell Sage. Taylor, M. (1976/1987). The possibility of cooperation. Cambridge: Cambridge University Press (rev. ed. of Anarchy and cooperation. London: Wiley). Treas, J. (1993). Money in the bank. Transaction costs and the economic organization of marriage. American Sociological Review, 58, 723–734. Tversky, A., & Kahneman, D. (1982). Judgment under uncertainty: Heuristics and biases. In: D. Kahneman, P. Slovic & A. Tversky (Eds), Judgment under Uncertainty: Heuristics and Biases (pp. 3–20). Cambridge: Cambridge University Press. Uzzi, B. (1996). The sources and consequences of embeddedness for the economic performance of organizations: The network effect. American Sociological Review, 61, 674–698. Uzzi, B. (1997). Social structure and competition in interfirm networks: The paradox of embeddedness. Administrative Science Quarterly, 42, 35–67. Voss, T. (1985). Rationale Akteure und soziale Institutionen. M¨unchen: Oldenbourg.
176
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
Voss, T. (2003). The rational choice approach to an analysis of intra- and interorganizational governance. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 21–46). Oxford: Elsevier Science. Weber, M. ([1921] 1976). Wirtschaft und Gesellschaft (5th ed.). T¨ubingen: Mohr. Weesie, J., & Raub, W. (1996). Private ordering: A comparative institutional analysis of hostage games. Journal of Mathematical Sociology, 21, 201–240. van der Wiel, H. (2000). ICT important for growth. CPB Report, 2, 17–22. Williamson, O. E. (1985). The economic institutions of capitalism. New York: Free Press. Williamson, O. E. (1993). Calculativeness, trust, and economic organization. Journal of Law and Economics, 36, 453–486. Williamson, O. E. (1996). The mechanisms of governance. New York: Oxford University Press.
APPENDIX A PROOF OF THE THEOREM The Base Model Given that a match between a buyer and a supplier has formed, we assume the buyer determines the degree of planning for the transaction. Putting more effort in preventing problems increases the probability that the transaction runs smoothly, but the extra effort comes at a price. The buyer must balance costs and benefits of extra management of the transaction. That is, buyers are assumed to choose a level of management m that optimizes their utility U: U = p(m)R + (1 − p(m))S − C(m) = S + p(m)(R − S) − C(m), where S is the utility of the buyer if problems occur (S for Sucker), R the utility of the buyer if the transaction runs smoothly (R for Reward), p the probability that no problems occur, m the amount of management invested in the transaction, and C the costs of management. The term R − S will be taken to be equivalent to the damage potential of the transaction (it represents the loss the buyer incurs if the transaction turns out to be a problematic one). For simplicity, we assume that management only affects the probability that problems will occur, and not the payoffs connected with a problematic or unproblematic transaction. We likewise assume that the optimal amount of management is consistent with the reservation utility of the supplier. That is, even though the buyer determines the optimal amount of management, he anticipates that the supplier cannot be forced to participate in a transaction that is not at least marginally profitable for the supplier. The optimal amount of management mopt is a (interior) solution of dC/dm = (R − S)dp/dm.
Contacts and Contracts
177
The probability that the transaction runs smoothly is assumed to depend on opportunism potential O and the degree of management m itself. In mathematical terms, we assume p = F (O, m) = F (a0 − a1 O + a2 m)
(a1,2 > 0),
with F a function that maps the real numbers to the unit-interval, for instance the standard normal cumulative distribution function (this ensures that p, which is a probability, is between 0 and 1). The costs of management are assumed to be linear in the amount of management: C = c 0 + cm
(c 0 , c > 0),
where c0 are the fixed costs of management and c the marginal costs of an extra unit of management. The model then reduces to U = S + F(a 0 − a 1 O + a 2 m)(R − S) − c 0 − cm, and the optimal amount of management mopt can be found by solving dU/dm = 0 for m. Through straightforward manipulation, we find that optimal transaction management in this base model is characterized by c a0 a1 1 −1 m opt [base model] = − + O + F . a2 a2 a2 a 2 (R − S) Including Dyadic Embeddedness The formal model still lacks a dynamic component: experiences from past transactions and potential future transactions are not yet incorporated, but are likely to affect the amount of management invested in the present transaction. To introduce the effects of dyadic embeddedness, we distinguish between buyers who have no past experience with the same supplier and those who have positive past experiences with the same supplier. Thus, we assume that buyers with bad experiences will try to find a more suitable supplier in the next period and neglect buyers with bad experiences and a positive shadow of the future. A three-period model will represent a transaction between a buyer who has had a business relation with the supplier. In each of the three periods, the buyer has to decide the extent to which the transaction will be managed. After the completion of the first transaction, a second transaction will follow with (exogenous) probability w. If the second transaction actually occurs, the buyer has to decide the extent to which the second transaction will be managed. After completion of the second
178
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
transaction, the third transaction will happen with the same exogenous probability w. Therefore, buyers who have had (at least) one transaction with the same supplier can be considered to be in the second period of such a three-period model. They have a shared past, are engaged in a second transaction now, and they have a potential future of extended transactions. We assume that past investments in management are useful in subsequent transactions to a certain extent. Hence, we first assume that a fixed percentage of the investment in management in a given period will be useful in the next one. For instance, parts of written contracts are useful in future transactions to some extent. Second, we assume that management of a transaction with an unknown partner requires set-up investments, like getting to know the partner, knowing whom to call for which kind of information, and the like. We denote the part of the investment that carries over to the next period by g1 (0 < g 1 < 1) and the set-up investment by g0 (g 0 > 0). Similarly, we define a two-period model for those buyers who have not completed a transaction with the same supplier before. These buyers can be considered to be in the first period of a two-period model: they are engaged in a transaction now, and they have a potential future of extended transactions. Note that our model has a fixed number of periods. Hence, following the standard “backward induction” argument on repeated games with complete information (see, e.g. Rasmusen, 1994, pp. 121–123), there is no possibility for conditional cooperation. It would be an option to capture the feasibility of conditionally cooperative behavior by assuming that a large perceived probability (w) of future transactions reduces the probability that problems occur. Figure 1 briefly summarizes our approach.
Fig. 1. Schematic Representation of Transactions with and without a Shared Past. Note: The w represents the exogenous probability that another transaction with the same partner will occur. Comparing transactions with and without a shared past implies comparing period 1 of the two-period model with period 2 of the three-period model.
Contacts and Contracts
179
As in the base model, we explicitly outline the buyer’s utility function U (indices denote the period). The two-period model then reads U = p(m 1 )R 1 +(1−p(m 1 ))S 1 − C(m 1 ) + w(p(m 2 )R 2 + (1 − p(m 2 ))S 2 − C(m 2 )) = S 1 + p(m 1 )(R 1 − S 1 ) − C(m 1 ) + w(S 2 + p(m 2 )(R 2 − S 2 ) − C(m 2 )). For simplicity, we assume that the costs of management in period i are linear, C(m i ) = c 0 + cm i , and do not change between periods. We also assume that the damage potential is equal for both transactions: S 1 = S 2 and R 1 = R 2 . By setting the partial derivatives to zero, we can derive the optimal investment in the first period of the two-period model. Of course, we can also derive the optimal investment in the second period, but here we only need optimal management in the first: m opt [first period of two-period model] a0 a1 1 −1 c(1 − wg 1 ) = − + O + F . a2 a2 a2 a 2 (R − S) Optimal management in the two-period model resembles optimal management in the base model. We find that the probability that a future transaction with the same supplier takes place (w) has a positive effect on the optimal amount of management (because F−1 is decreasing). That is, if the buyer has had no previous transaction with the supplier, having a larger shadow of the future will help increase the optimal investment in management. The extension to a three-period model will allow us to say something about the effect of the shadow of the future for those cases in which previous transactions have taken place. Remember that we want to compare the optimal management in the first period of the two-period model (defined above) with the optimal management in the second period of the three-period model. Straightforward manipulation of a similarly defined model with three periods yields that we can express the optimal management in the second period of the three-period model in terms of the optimal management in the first period of the two-period model: m opt [second period of three-period model] = (1 − g 1 )m opt [first period of two-period model] − g 0 . That is, buyers with a shadow of the past with the same supplier should manage less (since g0 is positive and 0 < 1 − g 1 < 1). Intuitively, this makes perfect sense. Buyers with a shared history of investments in management can use some of the investment in management from a previous transaction, which implies that less management is necessary in the current period.
180
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
Summarizing the above in a single equation, we find that optimal management can be characterized by
a0 a1 1 −1 = (1 − g 1 I past ) − + O + F a2 a2 a2
m opt
c(1 − wg 1 ) a 2 (R − S)
− g 0 I past ,
where Ipast represents an indicator function equal to 1 if a shared past exists. Substituting D for R − S and bs for the as completes the proof.
APPENDIX B OVERVIEW OF INDICATORS AND VARIABLE CONSTRUCTION
Variable
Indicator: original question [variable construction label]
Management “How much time did you and your colleagues spend on writing down the agreement and on the negotiations with the supplier of this product?” [person-days] “Which of the following departments of your firm were involved in drawing up the agreement?” (management, IT-department, financial department, production department, purchasing department, sales department, legal department) [departments] “Did your firm make use of external legal advisors to draw up or judge the agreement?” [advisors] “Was the main agreement mainly a standard contract or mainly a tailor made contract?” [tailor]
Answer categories
Open answer category: number of person-days
Not applicable (=0)/no (=0)/yes (=1) (for every department)
No (=0)/yes (=1)
Mainly standard (=0)/mainly tailor made (=1)
Contacts and Contracts
181
APPENDIX B. (Continued ) Variable Indicator: original question [variable construction label] “For each of the following financial and legal clauses, can you indicate the extent to which they were treated during the negotiations?” (price determination, price level, price changes, payment terms, sanctions on late payment, delivery time, liability supplier, force majeure, warranties supplier, quality (norms), intellectual property, piracy protection, restrictions on product use, non-disclosure, insurance supplier, duration service, reservation spare-parts, duration maintenance, updating, arbitration, calculation R&D costs, joint management during transaction, technical specifications, termination) [clauses treated] “For each of the following financial and legal clauses, can you indicate how they were arranged?” (same clauses as in previous question) [clauses arranged]
“For each of the following technical specifications, can you indicate how they were specified in the agreement or whether the specification was not applicable?” (security, user friendliness, definition system boundary, definition system functions, main board, internal and external
Answer categories
Little (=1)/normal (=2)/much (=3) (for every legal clause). [Clauses treated] is the main principal component of these 24 clauses (eigenvalue 6.89; explained variance 28.7%)
Not at all arranged (=0)/only verbally (=1)/in a written document (=2)/(for every legal clause). [Clauses arranged] is a weighted score of the number of clauses that was arranged either verbally or in writing. Very generally (=1)/general (=2)/in some detail (=3)/detailed (=4)/very detailed (=5)/not applicable (=missing) (for every technical specification). [Technical specs] is the average of these 24 clauses (alpha = 0.95).
182
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
APPENDIX B. (Continued ) Variable
Indicator: original question [variable construction label]
Answer categories
memory, speed processors, interfaces with other equipment, environment, additional hardware, installation procedure, monitor quality, type operating system, application software, procedure implementation, required memory, system analysis, system methodology, definition data design, definition programs, definition conversion, definition operation, definition benchmark, program language) [technical specs] Management is the main principal component of the indicators mentioned above (eigenvalue 2.35; 33.5% explained variance). Management = 0.45 [person-days] + 0.24 [departments] + 0.35 [advisors] + 0.17 [tailor] + 0.52 [clauses treated] + 0.50 [clauses arranged] + 0.24 [technical specs]. Volume
“How much was paid to the supplier, not including later supplements?”
Up to 10,000 US$ (=0.125)/10,000–20,000 US$ (=0.375)/20,000–50,000 US$ (=0.75)/50,000–100,000 US$ (=1.5)/more than 100,000 US$ (=3.5)
Contacts and Contracts
183
APPENDIX B. (Continued ) Variable
Indicator: original question [variable construction label]
Answer categories
Monitoring Problems
“Which of the following products were delivered at that time?” (personal computers, workstation, network configuration, mini computer, mainframe, computer-controlled machines, side equipment, cabling) [complexity hardware]
No (=0)/yes (=1) (for every product). [Complexity hardware] is coded as: 0: none of the hardware products is delivered, 1: personal computer/workstation/side equipment/cabling, 2: network configuration, 3: mini computer, 4: mainframe, 5: computer controlled machine. No (=0)/yes (=1) (for every product). [Complexity software] is coded as: 0: none of the software products is delivered, 1: standard software, 2: adjusted software, 3: tailor-made software. No (=0)/yes (=1) (for every service). [Complexity services] is coded as: 0: none of the services is delivered, 1: documentation/support, 2: instruction/consultation, 3: training, 4: design. Very easy (=1)/easy (=2)/somewhat difficult (=3)/difficult (=4)/very difficult (=5)
“Which of the following products were delivered at that time?” (standard software, adjusted software, tailor-made software) [complexity software]
“Which of the following services were delivered at that time?” (design, training, instruction, consultation, documentation, support) [complexity services]
“Was it difficult for you and your employees to judge the quality of the product at the time of delivery?” [quality]
184
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
APPENDIX B. (Continued ) Variable
Indicator: original question [variable construction label]
Answer categories
“Was it difficult for your firm to compare tenders?” [tenders]
Very easy (=1)/easy (=2)/somewhat difficult (=3)/difficult (=4)/very difficult (=5) Very easy (=1)/easy (=2)/somewhat difficult (=3)/difficult (=4)/very difficult (=5) Very easy (=1)/easy (=2)/somewhat difficult (=3)/difficult (=4)/very difficult (=5) None (=1)/little (=2)/some (=3)/much (=4)/very much (=5)
“Was it difficult for your firm to compare the product with other products?” [other products] “Was it difficult for your firm to compare the price-quality relation of potential suppliers?” [price-quality] “Compared to other firms in your sector of industry, how much experience did your firm have with automation?” [experience] “Does your firm have employees with expertise on automation, or an automation department?” [expertise] “Does your firm have the possibility to make or adapt this product?” [“make” possible] Monitoring Problems is the main principal component of the indicators mentioned above (eigenvalue 3.01; 30.1% explained variance). Monitoring Problems = 0.20 [complexity hardware] + 0.21 [complexity software] + 0.31 [complexity services] + 0.42 [quality] + 0.42 [tenders] + 0.47 [other products] + 0.45 [price-quality] − 0.22 [experience] − 0.08 [expertise] − 0.03 [“make” possible].
No (=0)/yes (=1) (‘yes’ means having either or both) No (=0)/yes (=1)
Contacts and Contracts
185
APPENDIX B. (Continued ) Variable
Indicator: original question [variable construction label]
Answer categories
Replacement Costs
“What would have been the damage, in terms of money and time spent on purchasing a new product, if the product had failed to function and had had to be replaced?” [new product] “What would have been the damage, in terms of money and time spent on training personnel, if the product had failed to function and had had to be replaced?” [training] “What would have been the damage, in terms of money and time spent on data entry, if the product had failed to function and had had to be replaced?” [data entry] “What would have been the damage, in terms of money and time wasted by idle production, if the product had failed to function and had had to be replaced?” [idle production] Replacement Costs is the main principal component of the indicators mentioned above (eigenvalue 2.33; 58.2%. explained variance) Replacement Costs = 0.52 [new product] + 0.53 [training] + 0.50 [data entry] + 0.45 [idle production].
Very small (=1)/small (=2)/moderate (=3)/large (=4)/very large (=5)
Very small (=1)/small (=2)/moderate (=3)/large (=4)/very large (=5)
Very small (=1)/small (=2)/moderate (=3)/large (=4)/very large (=5)
Very small (=1)/small (=2)/moderate (=3)/large (=4)/very large (=5)
186
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
APPENDIX B. (Continued ) Variable
Indicator: original question [variable construction label]
Answer categories
Importance of Durability
“How important was a long-term suitability of this product?” [suitability]
Unimportant (=1)/hardly important (=2)/moderately important (=3)/very important (=4)/of major importance (=5) Unimportant (=1)/hardly important (=2)/moderately important (=3)/very important (=4)/of major importance (=5) Unimportant (=1)/hardly important (=2)/moderately important (=3)/very important (=4)/of major importance (=5)
“How important was a long-term support by the supplier?” [support]
“How important was a long-term compatibility of this product with other hardware and software?” [compatibility] Importance of Durability is the main principal component of the indicators mentioned above (eigenvalue 1.69; 56.5% explained variance). Importance of Durability = 0.61 [suitability] + 0.62 [support] + 0.49 [compatibility]. Importance for Profitability
“How important was this product for the profitability of your firm?” [profitability]
“How important was this product for the automation of your firm?” [automation]
Unimportant (=1)/hardly important (=2)/moderately important (=3)/very important (=4)/of major importance (=5) Unimportant (=1)/hardly important (=2)/moderately important (=3)/very important (=4)/of major importance (=5)
Contacts and Contracts
187
APPENDIX B. (Continued ) Variable
Indicator: original question [variable construction label]
Answer categories
“How important was it that the product delivery time was met?” [delivery time]
Unimportant (=1)/hardly important (=2)/moderately important (=3)/very important (=4)/of major importance (=5)
Importance of Profitability is the main principal component of the indicators mentioned above (eigenvalue 1.58; 52.6% explained variance). Importance of Profitability = 0.58 [profitability] + 0.60 [automation] + 0.55 [delivery time]. Standardized Procedures
“Every firm has its standardized procedures. Regarding the negotiations and agreements with this supplier concerning the product as a whole: To what extent could these be considered to be standard procedures?”
Hardly (=1)/to some extent (=2)/to a moderate extent (=3)/largely (=4)/completely (=5)
Legal Expertise
“Firms need legal expertise. Does your firm have (a) employees with specific legal expertise, (b) a separate legal division?”
No (=0)/yes (=1, if any)
Past
“Has your firm had any kind of business relation with this supplier before the purchase of this product?” “To what extent did you expect, before the purchase of this product, that your firm would continue business with this supplier?”
No (=0)/yes (=1)
Future
No business (=1)/incidental business of limited size (=2)/some business of limited size (=3)/regular and/or extensive business
188
RONALD S. BATENBURG, WERNER RAUB AND CHRIS SNIJDERS
APPENDIX B. (Continued ) Variable
Indicator: original question [variable construction label]
Answer categories
Size buyer
“How many full-time employees were working at your firm at the time of the purchase of this product?”
Size supplier
“How many full-time employees were working at the supplier at the time of the purchase of this product?”
(=4)/very regular and/or very extensive business (=5) Open answer category: number of full-time employees Number of full-time employees (<5 (=1)/5–9 (=2)/10–19 (=3)/20–49 (=4)/>49 (=5))
APPENDIX C BIVARIATE CORRELATIONS FOR VARIABLES IN THE ANALYSES Variable
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(1) Management
1
–
–
–
–
–
–
–
–
–
–
(2) Volume
0.54
1
–
–
–
–
–
–
–
–
–
(3) Monitoring Problems
0.34
0.36
1
–
–
–
–
–
–
–
–
(4) Replacement Costs
0.41
0.42
0.41
1
–
–
–
–
–
–
–
(5) Importance of Durability
0.30
0.26
0.24
0.35
1
–
–
–
–
–
– –
(6) Importance for Profitability
0.39
0.45
0.20
0.42
0.33
1
–
–
–
–
−0.08
−0.16
−0.17
−0.08
−0.02
−0.10
1
–
–
–
–
0.10
0.08
−0.07
−0.00
−0.00
0.05
0.04
1
–
–
–
(9) Past
−0.16
−0.07
−0.23
−0.12
−0.07
−0.02
0.16
0.06
1
–
–
(10) Future
−0.03
0.00
−0.04
0.02
0.11
0.08
0.10
0.06
0.36
1
–
(7) Standardized Procedures (8) Legal Expertise
(11) Size Buyer
0.17
0.33
−0.06
0.05
0.07
0.12
−0.02
0.20
0.03
0.01
1
(12) Size Supplier
0.26
0.39
0.12
0.21
0.12
0.20
0.02
0.07
0.08
0.08
0.26
GOVERNING STRATEGIC ALLIANCES Toby E. Stuart ABSTRACT Assuming that information about the participants in the network circulates over the ties comprising it, firms’ structural positions – defined in this chapter by their location in the network of past strategic collaborations – should affect their general reputations as collaborators and the knowledge that structurally proximate organizations possess about their past behavior. In turn, the information benefits associated with different network positions should influence the types of governance structures and contractual features that appear in new alliances. This chapter examines how the positions of biotechnology firms in the established network of strategic alliances influences one of the important contractual terms of new partnerships: whether or not one partner finances the activities of its counterparty as part of the deal.
INTRODUCTION Sociologists posit that the social systems contextualizing economic exchanges offer alternatives to legalistic and hierarchical mechanisms for the governance of interfirm contracts (see Voss, 2003 in this volume for a detailed discussion). Even before Macaulay’s (1963) seminal study showing business executives’ reluctance to rely on formal contractual provisions to manage the terms of business relationships, the discipline had supposed that the broader institutional context in which economic exchanges take place performs a governance function. In fact, the central observation in the sociological perspective is that actors’ identities – assuming The Governance of Relations in Markets and Organizations Research in the Sociology of Organizations, Volume 20, 189–208 © 2003 Published by Elsevier Science Ltd. ISSN: 0733-558X/PII: S0733558X02200075
189
190
TOBY E. STUART
that the relevant actors are embedded in a sufficiently closed social system – serve as an indirect mechanism to control the behavior of exchange partners. One of the implication of this insight is that the problem of transaction management often manifests in the careful attention that economic actors expend in the search for appropriate exchange partners, rather than inhering in the post-deal governance function. A second implication, which is now supported by a growing body of empirical work and is the subject of the data analysis in this chapter, is that the identities of the partners in a transaction affect their perceived need for contractual safeguards and the level of resources they are willing to commit to exchange partners. Expressed in Burt’s (1992) language, sociologists contend that networks comprising prior business relationships among economic actors create “information and control” benefits to certain, advantageously positioned market participants. “Information” specifically refers to the knowledge available to some actors of the past behavior and competencies of other entities (individuals or collectivities) in the exchange system. “Control” typically refers to the ability – actual or simply perceived by an exchange partner – of one member of the social system to sanction another member. In the network literature, actors are often thought to be able to control a counterparty’s behavior when they have the capacity to change the content of the information about the exchange partner available to other entities in the network (Raub & Weesie, 1990). Control thus assumes the form of deterrence: actors may abstain from certain behaviors to prevent counterparties from disseminating negative information about them to other participants in the exchange system (Buskens, 2002; Stuart & Robinson, 2001). In this chapter, I empirically explore the general assertion that actors’ identities are central determinants of the terms of exchange between market participants. Specifically, I examine the population of strategic alliances established in the United States biotechnology sector, looking at how proxies for: (i) the general observability of the reputation of the agent in asymmetric strategic alliances; and (ii) the level of specific knowledge of an agent’s previous conduct that is likely to be available to the particular principal in the deal, affects the level of resources committed by the principal in the transaction.1 The alliances I study are asymmetric in that the prototypical exchange relationship in the dataset is a vertical transaction in which the downstream partner is generally the larger organization, and often purveys financial resources to the upstream partner. Because significant resources are exchanged in some of the alliances in the database, concerns about control and partner malfeasance are endemic to the types of transactions I examine here. In the chapter, I assert – and empirically test – the claims that the downstream partner’s ability to learn about the transactional reliability of the upstream partner affects its willingness to commit resources to the alliance.
Governing Strategic Alliances
191
SUPRA-TRANSACTIONAL GOVERNANCE MECHANISMS Transaction cost theory (Williamson, 1985), along with the literature on incomplete contracts (e.g. Grossman & Hart, 1986), has brought the issue of the governance structures of inter-corporate and inter-individual relationships to center stage in organizational economics. These literatures seek to explain how economic transactions are distributed across different organizational forms, and how control rights are allocated between transaction partners. The canonical view in economics holds that the parties to a transaction will, in Williamson’s language, behave opportunistically if: (i) such behavior maximizes their self interest, regardless of whether or not it is deleterious to the interests of the partner in the transaction; and (ii) the conduct of the actor is not constrained by an ex ante contract that protects the interests of the partner. Much of the literature in economics on the governance of interfirm transactions begins with these assumptions, and then explores the optimal division of control rights between transaction partners to minimize incentive incompatibilities between exchange partners.2 Granovetter (1985), drawing heavily from the insights of Polanyi, offers the programmatic attack on this “asocial” – or purely individualistic – conceptualization of the incentive structures facing market participants in exchanges with transaction partners. Granovetter observes that most economic transactions are embedded in a broader social network – a collection of actors that engage in repeated, often enduring exchange relations. Such networks affect both the information available to members of an economic system about the reliability of particular transaction partners and the willingness of those potential partners to harm the material interests of their partners, irrespective of their own interests. This social context may substantially alter the incentives and, some believe, dispositions of market participants, especially relative to the ideal-type market comprising only episodic transactions among traders. The challenges launched by sociologists at the economic perspective on transaction governance have been aimed both at the level of core assumptions regarding the behavioral motivations of economic actors, and at the narrowness of the scope of the methods presupposed to be available to actors to govern economic exchanges (i.e. that governance forms can be adequately described by a bimodal distribution, with hierarchy and spot market transactions representing the two modes). I will briefly address both of these disputes with the economist’s view. On one side of the debate, sociologists have objected to the assumption that an actor will behave opportunistically every time that he gains material benefit from doing so and that he perceives a high probably that his malfeasant behavior will go unpunished (or, if he incurs sanctions, that the loss suffered as a result of them is outweighed by the benefits derived from behaving opportunistically).
192
TOBY E. STUART
This line of argument highlights transactors’ frequent observance of normatively prescribed conduct, such as the proclivity of economic actors to adhere to the norm of reciprocity (Powell, 1990; Sahlins, 1972; cf. Lincoln et al., 1992 for empirical evidence in an interorganizational context). Scholars of this view also note that friendships and feelings of obligation routinely emerge among longtime transaction partners (e.g. Uzzi, 1996), and that these relationships engender a “spirit of goodwill” vis-`a-vis exchange partners (Dore, 1983). These views assert that economic actors’ conduct in market exchanges cannot be accurately modeled as arising just from narrowly rational calculations that weigh short-term gains and losses, because the behavior of transactors depends in part upon the cultural and normative tenets that characterize a social system, and the strength of the emotional bond that unites the members of a transaction dyad. In other words, many sociologists believe that exchange partners may engage in cooperative behavior even if such behavior is not motivated by self interest in a broad sense – that is, when the assessment of self interest accounts for the threat of future sanctions, loss of potential, future opportunities for exchange, and other opportunity costs associated with malfeasant behavior. Thus, on this point, the polemic between the disciplines occurs at the level of fundamental assumptions. The other side of the sociological critique is more closely aligned with the rational choice perspective in the discipline, and challenges the prevalent assumption that markets and hierarchies are the only viable organizational forms for conducting intercorporate transactions.3 As such, studies in this tradition have no major philosophical opposition to the work in economics on the general risks associated with certain types of economic exchanges (i.e. the assumption that actors behave in strict accordance with cost-benefit calculations), although there is a considerable discrepancy between the two disciplines regarding the structural forms that can be expected to govern such transactions. The distinctive claim in this line of thinking is that the social context of an economic system affects both who is likely to transact with whom, and how transactions are governed, due to actors’ efforts to select exchange partners and govern exchanges to economize on transaction costs (e.g. Batenburg et al., 2003 in this volume). There are a few elements to the argument that the social structure of the market can affect both the pattern of affiliation and the costs of conducting economic exchanges. Because these arguments are by now familiar and because they are reviewed in considerable detail in other chapters of this volume and elsewhere recently (e.g. Podolny & Page, 1998), they will receive just a brief treatment here. First, networks conduct information through the market: the ties among participants demarcate the pathways that information, most importantly that which identifies a potential exchange partner’s capabilities and reliability, traverses. This is most obvious when the same two economic actors are involved in a sequence of
Governing Strategic Alliances
193
exchange relationships, a condition that Buskens, Raub & Snijders (2003) in the introduction of this volume refer to as “dyadic” embeddedness. The fact that two partners have transacted in the past suggests a mutual awareness of character, or at least the perception among transactors of such an understanding. Furthermore, the expectation of future exchanges may introduce opportunity cost to malfeasant behavior, as conduct that one’s exchange partner interprets as non-cooperative often results in the partner’s decision to withdraw from the relationship. Thus, an established relationship between trading partners imparts “information” (in the form of knowledge of past behavior) and “control” (i.e. the ability to sanction opportunistic behavior by withdrawing from the relationship) benefits. When dyadic exchanges take place in a broader network of relationships, information about the qualities of trading partners spills out from one relationship to other market participants, via overlapping ties. This information, which is thought to diffuse locally (i.e. the quality content and depth of the information transmitted likely depends upon the strength and the lengths of the ties that are traversed), affects the knowledge that one market participant has about the trustworthiness of another transactor (Mizruchi & Galaskiewicz, 1993). Thus, when an economic actor has a previous, positive transaction experience with a particular exchange partner, or when that actor has a strong connection to a third party that has had a positive trading experience with the partner, then the likelihood that the focal actor will extend trust is increased (e.g. Coleman, 1990; Granovetter, 1985). The interdependence among transactions, which is created by the spillover of information beyond the dyads that have engaged in past exchanges, also introduces the possibility that an aggrieved exchange partner can sanction an alter by transmitting negative information about that actor to other market participants (Buskens, 2002; Raub & Weesie, 1990; Stuart & Robinson, 2001). The threat of sanctions is understood to be a mechanism for controlling the behavior of another actor. However, an actor’s power to sanction an exchange partner depends upon the extent of his ability to diffuse negative information about the exchange partner to that individual’s potential, future trading partners.4 This ability is thought to be a direct function of the actor’s centrality in the network of market participants. Although sanctioning power varies across social positions and thus is not equally accessible to all potential transactors, the deterrence implicit in it represents a supra-contractual method of controlling the behavior of exchange partners, at least for those actors that possess the power to sanction. Pulling these threads together, the research on embeddedness has argued that the incentive structures that the network ostensibly overlays on any particular transaction reduce the expense associated with negotiating an elaborate contract to govern the exchange. Or, stating this central point in an almost equivalent way, the network reduces the likelihood of malfeasance in many exchanges.5 This central claim is
194
TOBY E. STUART
now supported by a number of empirical studies. These studies either directly examine how the governance of economic exchanges varies with the identities (i.e. network positions) of exchange partners, or indirectly address the issue by demonstrating that network positions significantly influence the process of selecting transaction partners. Examples of the studies directly examining how network positions shape transaction governance include Rooks et al. (2000) factorial survey of purchasing managers, which establishes the substitution of network-based governance for more costly contractual provisions when exchange partners share an embedded relationship. There is also work examining how the social structural positions of participants in strategic alliance networks impact the governance of interfirm alliances (e.g. Gulati, 1995a; Pisano, 1989). Most of this research examines the probability that an equity investment will accompany the formation of a strategic alliance.6 Finally, there is a related literature that empirically addresses the likelihood that a pair of organizations will establish a new relationship. Although the network-based literature examining the selection of transaction partners is often positioned as addressing heterogeneity in actors’ opportunities for collaboration or search strategies for identifying exchange partners, it is market participants’ fundamental concern about transaction governance that provides the rationale for why networks matter in partner selection in these studies. For example, Gulati (1995b) predicts that two firms are more likely to enter an alliance when they have each previously participated in a deal with a third party that is common to both organizations. The explanation invoked for why organizations prefer to transact with structurally proximate exchange partners is twofold: firms are likely to possess more information about the reliability of these potential partners, and the common third party to the exchange might help to enforce cooperative behavior among alliance participants (cf. Coleman, 1990). Hence, the empirical studies demonstrating that social structural proximity influences the formation of exchange dyads also can be understood as evidence about the importance of networks in ameliorating governance problems in interorganizational exchanges.
TWO PREDICTIONS In this chapter, I present an empirical test examining just the “information”-related benefit of network embeddedness, specifically focusing on how the resources committed to an exchange relationship vary according to actors’ abilities to learn about their transaction partners.7 In particular, I examine strategic alliance contracts in the biotechnology sector and treat as the outcome variable an indicator of whether
Governing Strategic Alliances
195
the principal in the transaction supplies significant financial resources to the agent in the partnership as a component of the alliance.8 The rationale for examining the commitment of funds as the outcome variable is that the incentives for agents to cheat, and the losses that the principal incurs in the event of expropriation, are each greatest in funded alliances (see also Buskens et al., 2000). As a result, the requirement for trust is greatest in funded alliances, as principals’ perceived need for trust will increase in the agents’ incentives to behave opportunistically.9 In this chapter, the network from which actors’ positions are distilled is crafted from the collection of all inter-firm strategic alliances in the biotechnology industry. I assess the role of two covariates as determinants of the presence of funding in alliance contracts. The first measure is a characteristic of the individual nodes (firms) in the network and the second is a measure of the structural proximity of the members of a partnership (i.e. one is a nodal property and the second is a dyadic property). The first property is the amount of time that has elapsed since the agent in the alliance – always a biotech firm – initially entered the network.10 The rationale for this measure is that reputations derive from observable performance track records. Little information is available about the reliabilities, competencies, and trustworthiness of new entrants to the network; thus, time since entry is intended to capture the extent to which an actor has a known reputation. My first hypothesis can be formally stated: strategic alliance contracts in partnerships in which the agent in the transaction has a long history in the alliance network are more likely to include funding from the principal in the deal ( prediction 1). As the literature review emphasizes, one of the core claims of the embeddedness perspective is that reputations are not general attributes of actors, but structurally delimited characteristics: all participants in a marketplace rarely are privy to the same information about the qualities of a particular actor. Deep knowledge about a transactor is available through the network from those with previous exchange relations with that actor. The most reliable source of information about an exchange partner is that which is obtained from direct observation. Two actors that have participated in a previous exchange relationship are thus likely to feel confident that they are knowledgeable about each other’s reliability as partners. Conditional on the willingness of both actors to establish a new alliance, my second hypothesis states: alliance contracts between principals and agents that have previously transacted together are more likely to entail funding from the principal in the partnership ( prediction 2). I now turn to a description of the biotechnology industry – the context for the empirical test of the two predictions. This industry serves as an appropriate locale for testing the predictions because the firms in the business are closely linked through an extensive network of strategic alliances.
196
TOBY E. STUART
STRATEGIC ALLIANCES IN BIOTECHNOLOGY Two radical innovations – recombinant DNA (rDNA) and hybridoma (cell fusion) – launched the contemporary biotechnology industry. The recombination of DNA first occurred in 1973. Two years later, cell fusion was used to create monoclonal antibodies, which serve as the body’s defense against diseasecausing bacteria, viruses, and cancer cells. Subsequent developments in areas such as genomics, proteomics, high throughput screening methodologies, and microarrays (biochips), have continued to propel forward the scientific base of biotechnology. Although the pioneering scientific work was conducted in the early 1970s, the first spurt of entrepreneurial activity in the young field did not occur until the early 1980s. A considerable increase in the number of foundings of new biotechnology firms in 1981 can be attributed to a number of pivotal events in 1980. First, in a landmark ruling (in the case, Diamond vs Chakrabarty), the U.S. Supreme Court affirmed the patentability of new life forms. This development insured that biotechnology innovators would be able to appropriate the returns from patentable discoveries. Also, the passage of the Patent and Trademark Amendment Act of 1980 enabled universities to apply for patents on discoveries emerging from federally funded research programs. These events contributed to a surge in company foundings and initial public offerings of biotech equities in the 1981 to 1983 period (to be exact, 22 IPOs took place in this interval). At the scientific advent of contemporary biotechnology, established chemical, pharmaceutical, and agriculture firms were poorly positioned to enter the new field. Biotechnology required skills in molecular biology and biochemistry, which were quite distinct from those demanded by the synthetic chemistry-based technologies for which they were expected to substitute. Hence, biotechnology represented a competence-destroying development, relying upon a novel set of techniques, which existing chemical and pharmaceutical firms found difficult to acquire. As a result, the commercialization of biotechnology was shepherded by start-up, dedicated biotechnology firms. Given high scientific entry barriers as well as the usual gamut of inertial forces that constrain the innovative initiatives of established firms (Sorensen & Stuart, 2000 present evidence in this context), existing pharmaceutical and chemical companies were unable to enter biotechnology immediately. However, diversified corporations did recognize the commercial potential of biotechnology, and have chosen to pursue it primarily through strategic alliances with startup firms. Because innovative capabilities were widely dispersed across the players in biotechnology, firms entered into an extensive set of alliances to gain access to different technologies and capabilities (Powell et al., 1996).
Governing Strategic Alliances
197
Taking the view from the other side of the table, early-stage biotechnology companies have many motivations to enter strategic alliances with established companies. Developing biotechnologically based therapeutics is an extremely costly endeavor (recent estimates of the cost of developing a drug and steering it through the FDA approval process range up to $500 million), and this fact has led to keen interest on the part of early-stage biotechnology firms to form partnerships with resource-rich collaborators. Moreover, significant strategic alliances with well-known companies are now virtually prerequisites for private companies to reach an IPO (Stuart et al., 1999). Biotech firms also lack the sales and marketing organizations required to distribute new drugs, and so seek out strategic partners that possess these resources. With economic interests pushing these organizations toward collaboration, there have now been many thousands of alliances involving dedicated biotechnology companies.
DATA AND MEASURES To construct the covariates for the analysis, I have assembled the strategic alliance network in the biotech industry, dating back to the emergence of the industry in the late 1970s. To allow time for network participants to acquire known identities, the empirical analysis examines all transactions inked between 1985 and 1997. The network is thus traced forward as it evolves over time via the creation of new firms and the formation of new alliances. After cleaning the data, the biotechnology industry alliance network includes just shy of 4,000 alliances, which were established between approximately 1,800 distinct organizations. The primary for source for these data is Recombinant Capital, a data vendor that serves the biotech industry (see RDNA.com). In recent years, the amount of resources exchanged in biotech alliances has risen precipitously (Robinson & Stuart, 2001a, b). For example, an alliance established in May 2000 between the Swiss Life Sciences firm Novartis, and the Boston-based biotechnology company Vertex Pharmaceuticals, entails a funding commitment from Novartis with an upper bound of $815 million. The purpose of this alliance is to discover, develop, and commercialize small molecule drugs directed at targets in the kinase protein family. During the span of the collaboration, Novartis is obligated to make pre-commercial payments to Vertex. These payments, which are triggered by the accomplishment of milestones identified in the alliance contract, are related to the successful discovery and development of eight compounds.11 Although atypical in terms of the level of resources committed, this alliance has the typical structure of the research and development alliances that I examine in this chapter: the pharmaceutical company (Novartis) finances a research program
198
TOBY E. STUART
at the biotech company (Vertex) to discover biological targets (or to perform other types of scientific research) leading to the development of small molecular drugs. The bulk of the research is typically conducted at the biotech company, while the pharmaceutical firm is often responsible for later-stage product development activities (e.g. running clinical trials and manufacturing a drug). In the regressions presented below, the unit of analysis is a strategic alliance and the dependent variable is an indicator variable denoting that the principal in the alliance purveys funding to support research at or acquire technology from the agent. This variable is coded as “1” when the principal supplies funding, and “0” otherwise. As discussed above, I interpret the fact that the principal in an alliance has pledged funds as part of the deal as evidence that the principal assesses its exchange partner in the focal transaction to be worthy of trust.12 Matching the two predictions outlined above, the empirical analysis will focus on the effects of two network-related properties as explanatory variables. The first variable is the amount of time that has elapsed since the agent in the alliance initially entered the network. This variable is simply operationalized as the number of years that have passed since the agent in the present transaction has established its first strategic alliance. If the current alliance is also the first established by the agent, then time since first deal is coded as “0”. The second covariate of interest is a dummy variable indicating whether the members of the current alliance dyad have previously collaborated. Among other factors, this variable is intended to proxy for the amount of direct information that members of an alliance dyad have about one another’s reliability.
ESTIMATION Assuming that the theory is correct, there is a salient complication in estimating the relationships predicted above: the variables posited to predict the characteristics of a transaction also affect the likelihood that the transaction exists in the first place. The literature review in this chapter highlights the intimate connection between the governance of an exchange relationship and the mere fact of its existence in a pair of transaction partners. This work indicates that network-based trust affects more than just how an exchange relationship is structured and governed; it also presupposes that trust is often necessary for two market participants to agree to establish the relationship in the first place. In terms of fitting models that relate network positions to the contractual features of an inter-firm alliance, this fact implies that a standard sample selection problem hampers estimation of the effects of the network-based covariates on the probability of funding in an alliance (see also Buskens et al., 2000, pp. 36–37, who discuss
Governing Strategic Alliances
199
this issue in a similar context). Suppose that one is examining a network with N actors. In each period, this network affords the possibility of N × (N − 1)/2 trading relationships in distinct dyads. In a typical marketplace, such as the market for strategic alliances in the biotechnology sector, only a small fraction of the total number of possible pairings will establish trading relationships. Based upon all social structural theories of tie formation, the realization of these dyads is nonrandom with respect to actors’ and dyads’ positions in social structure: transactions are more likely to occur in social structurally proximate pairings, for the reasons articulated above. Because transactions are selected in part on the basis of network positions, one cannot estimate unbiased parameters in regressions of transaction characteristics on network effects.13 I address this issue by estimating a two-stage probit of the probability that the principal includes funding in an alliance with an agent, in which the proxies for agent reputation and the level of direct knowledge that the principal has about the agent are included in both the selection stage and second stage equations. In the first stage, an estimate of the probability that an alliance is formed in a dyad in a given year is produced. I perform the first stage analysis by drawing a sample of nonrealized dyads (i.e. pairs of firms in which no alliance was created in a given time period), which is chosen to include the same proportions of pharmaceutical-biotech and biotech-biotech alliances as those that occur in the industry in a given year. The reason I draw the sample this way is that there are more agents (biotech firms) than principals (pharmaceutical and life science firms) in the alliance network.14 In effect, this procedure controls for the different alliance formation propensities of these two categories of firms, which occupy systematically different positions in the alliance network. The non-realized sample thus consists of pairs formed by randomly matching principals and agents. In each year, I draw the same number of non-realized dyads as there are actual alliances. The covariates are then computed for the non-realized dyads (time elapsed since the agent’s first partnership and whether or not there had been a previous alliance between dyad members), enabling a two-stage estimation approach. The two equations are estimated jointly and results are presented in Table 1.
RESULTS The first regression in Table 1 reports the findings from a baseline model that includes only control variables. The regression includes a vector of dummy variables representing deal characteristics. These variables indicate: whether or not the alliance is a joint venture; whether the principal acquires an equity stake in the
200
Table 1. Two-Stage Probit Regressions of the Probability of Alliance Funding. Model 1 Variable Alliance is joint venture Alliance includes minority equity Licensing alliance Manufacturing alliance Principal is biotech firm Elapsed time since agent’s first deal Repeat tie Repeat tie × Elapsed time Elapsed time × Elapsed time Constant
∗p
< 0.05.
−1.272*
(0.108) (0.054) (0.045) (0.115) (0.065)
0.579* 0.852* 0.347* −0.103 −0.352* 0.026* 0.152*
(0.092)
−1.037*
0.138* (0.005) 0.896* (0.086) −0.474* (0.022) −0.076 (0.073) 3857 3857
Model 3
Model 4
(0.107) (0.057) (0.045) (0.114) (0.052) (0.007) (0.055)
0.579* 0.848* 0.346* −0.104 −0.350* 0.029* 0.240* −0.011
(0.106) (0.057) (0.044) (0.113) (0.051) (0.007) (0.109) (0.086)
(0.090)
−1.065*
(0.093)
−0.005* (0.002) −1.195* (0.093)
0.138* (0.005) 0.873* (0.086) −0.473* (0.022) 0.270* (0.118) 3857 3857 −7030.1
0.138* (0.005) 0.863* (0.085) −0.473* (0.022) 0.405* (0.135) 3857 3857 −7026.4
0.138* (0.005) 0.873* (0.086) −0.474* (0.022) 0.241* (0.106) 3857 3857 −7031.22
0.573* 0.823* 0.334* −0.117 −0.331* 0.084* 0.186*
(0.105) (0.059) (0.044) (0.111) (0.051) (0.019) (0.055)
TOBY E. STUART
Selection equation Elapsed time since agent’s first deal Repeat tie Constant Arctan (ρ) Number of alliances Number of censored observations Log-likelihood
0.603* 0.864* 0.347* −0.087 −0.717*
Model 2
Governing Strategic Alliances
201
agent as part of the alliance; whether the principal licenses technology from the agent, and whether the alliance designates product manufacturing rights to one of the two partners (in almost every case, to the principal). The two deal characteristics that one would expect to have the greatest impact on the probability of funding are the dummies indicating that the principal acquired an equity stake or that the deal was a joint venture. In minority equity alliances and joint ventures, the principal is likely to have relatively more control over the agent (see note number 5), and may even obtain a seat on the board of directors of the agent as a condition of the equity investment. Because the principal has greater control, it is more likely to be willing to provide funding for research or development initiatives at the agent. In joint ventures, a separate governance structure, staffed by members of both parent firms, is established to oversee the activities of the alliance. This organizational arrangement also grants significant control to the principal. The findings in Table 1 support this view: there is a higher probability of funding from the principal in joint ventures and equity-based alliances. Table 1 also includes a dummy variable indicating that the principal in the alliance is another biotech firm (recall that 25% of the alliances are deals between two biotech firms). The biotech principal dummy has a negative effect on the probability of funding. Because biotech firms tend to be smaller and more resource constrained than pharmaceutical firms and life sciences conglomerates, it is to be expected that there are fewer funded biotech-biotech alliances. Although I do not report coefficients or comment on them, all of the regressions include a vector of 16 calendar time (year) dummy variables. The year dummies are included to remove the effect of equity market conditions or other temporally variable factors that might influence the bargaining power of the organizations in a partnership and agents’ need for funding. Before turning to the tests of the two predictions, I briefly comment on the results of the selection-stage equation. As already noted, the selection-stage equation can be understood as a regression of the probability that firms enter alliances. Although results should be interpreted cautiously because there are no control variables in the model, the regression indicates that two firms are significantly more likely to establish an alliance if they have previously formed a partnership, and as the amount of time elapsed since an agent’s first deal increases, so too does its propensity to form alliances. The regression in Column 2 in Table 1 adds the two variables of substantive interest – the number of years elapsed since the agent’s first alliance and whether or not the alliance is a repeat transaction between the same pair of organizations. The parameter estimates indicate that both variables work in the expected direction. First, when an alliance is a newly established engagement between a pair of former collaborators, it is more likely to involve funding from the principal.
202
TOBY E. STUART
This is consistent with the view that at least a minimal level of trust is a precondition for mutually aware exchange partners to establish new relationships; if either partner had exhibited opportunistic behavior in a previous transaction, a repeated exchange is unlikely. The number of years since the agent’s first alliance also works in the expected direction in Table 1: this variable has a positive and statistically significant effect on the probability of funding. Thus, agents with an observable performance record from at least one past alliance are more likely to attract funding from alliance partners. The final two regressions in Table 1 report on a couple of ad hoc extensions of the core analysis. The third model in the table includes an interaction effect between the two substantive variables: the time since the agent’s first alliance and whether or not the deal is a repeat tie. This interaction term is expected to carry a negative coefficient, because the specific, first-hand knowledge available to former collaborators should supplant the general information available in an observable performance history. Although the coefficient is in the expected direction, it falls a little shy of statistical significance. As in all other types of interfirm contracts, strategic alliances typically last for a contractually specified duration.15 This implies that, even if an organization’s performance in a strategic alliance is perfectly observable, the contribution of an alliance to the total amount of information about an organization’s reputation falls to zero after the deal dissolves. Moreover, the marginal information contributed by additional deals presumably begins to decline as organizations accumulate a performance history. Together, these observations suggest that effect of time elapsed since first alliance should wane over time. To account for this possibility, the final regression in Table 1 includes a quadratic term for elapsed time. As anticipated, the coefficient on quadratic term is negative and statistically significant, indicating that the effect of time elapsed since an agent’s first alliance on the probability of funding in a deal increases at a decreasing rate.16
CONCLUDING REMARKS Many economic sociologists would acknowledge that, to the extent that it is not possible to use producers’ identities to select appropriate transaction partners (i.e. when embeddedness is low due to, for example, continual change in the identities of market participants), the central issues and risks described in the transactions cost literature – that of partner malfeasance – becomes a major concern of market participants. However, when a transaction network exists and promulgates actors’ identities, the network both elevates the likelihood that certain actors will enter
Governing Strategic Alliances
203
into exchange relationships and alleviates many of the risks of transacting. The analysis presented in this chapter is yet another piece of empirical evidence to support this claim. As the body of empirical evidence supporting the core embeddedness claims grows, it remains the case that there are substantial lacunae in the empirical literature. Before concluding, I highlight three areas in which additional theoretical and empirical work promise to contribute to our understanding of network-based governance. First, there is reason to believe that the contractual structure governing an inter-firm transaction affects how the level of trust between exchange partners evolves. In a recent paper, Mayer (2000) examines the evolution of the contractual structure in a multi-year, multi-shot, interfirm relationship. The thesis of this paper is that clear and comprehensive contracts actually augment the level of trust among transaction partners. Because ex ante contracts explicate the roles and responsibilities of exchange partners, they forestall haggling due to differing understandings of the transaction partners. Although the sociological literature typically contrasts extensive contracts and network governance as substitutes, this paper raises the intriguing hypothesis that complete contracts, to the extent they are feasible, may actually contribute to interfirm trust. Second, sociological insight into the question of where networks come from has been lacking. The vast majority of the embeddedness research examines the material benefits of an established network to members of an exchange system. Yet, it is not very satisfying to examine the regulatory and efficiency capacities of a network of transactions without a causal theory of how the particular network structure emerges. There has been some work on this issue, for example, Gulati and Gargiulo’s (1999) study of the evolution of a strategic alliance network (see also the contribution on partner selection by Buskens, Batenburg & Weesie, 2003 in this volume). However, this paper, along with most of the other work that has been done on the subject, actually explores how a network builds upon itself, rather than how a de novo network begins. Thus, while the existing literature offers some endogenous models of tie formation, it lacks a persuasive theory of network emergence. Finally, the literature describing the role of social structure in shaping transaction patterns and the governance of exchange relationship is in desperate need of contributions derived from datasets that contain detailed information on the content of the information ferried across network ties. Similarly, few datasets provide information on the incentives of market participants to circulate reliable information about former transaction partners. Most empirical studies, including this one, propose that the abilities of would-be transaction partners to acquire knowledge about a potential counterparty’s prior conduct is positively associated with the desire of the two entities to enter into an exchange relationship, as well
204
TOBY E. STUART
as a reduced concern of governance-related problems. As a result, the literature has developed a functionalist undertone, which derives not so much from the logic of the argument, but for the equating of more and positive information. A richer and more nuanced understanding of the role of networks hinges on detailed knowledge of the content of transmitted information, and the incentives of the entities represented as nodes in the network to convey this information reliably.
NOTES 1. The use of the terms principal and agent here is consistent with that of Berle and Means (1932): there is a separation of control between principals (the owners of an asset) and agents in many of the strategic alliances examined here. This produces an incentive problem that must be resolved. 2. Williamson (1985) claims that the characteristics of a transaction determine whether it is cost minimizing to conduct the transaction over a market interface, or to expand the boundaries of the corporation to govern the transaction internally. Although I do not specifically address the critique in this chapter, one of the sociological criticisms of Williamson’s theory of the boundaries of the firm is that it employs a stylized conception of hierarchy as a method of governing complex transactions (e.g. Bradach & Eccles, 1989). The theory is very generous in terms of the benefits it ascribes to hierarchical governance. 3. Economists have begun to accommodate this critique. For example, Williamson’s more recent work (e.g. Williamson, 1996) acknowledges networks and reputation as methods of transaction governance. The network metaphor and language has also begun to creep into the writings of other economists. Examples include Kranton and Minehart’s (2000) model of buyer-seller networks and Bala and Goyal’s (2000) model of network formation. 4. Because sanctioning entails disseminating negative information about a transactor to potential, future exchange partners, the distinction between “information” and “control” is really just a conceptual one (see Buskens & Raub, 2002 for an extensive discussion). 5. It is worth adding that focusing exclusively on economizing on the cost of writing extensive contracts probably underestimates the potential savings that are generated when relying on the governance function of the network. To the extent that the social context enables the terms of a transaction to be expressed vaguely, it is possible that exchange partners will have greater flexibility in adapting the terms of the transaction to evolving contingencies. This potential function of the network is directly relevant to a number of the widely influential models in economics describing the efficient allocation of control rights among the parties in economic exchanges (e.g. Grossman & Hart, 1986; Hart & Moore, 1988), which build from the assumption that there can be severe incentive problems in situations where contracts are incomplete. It is possible that the presence of network-based governance options permits a more efficient allocation of control rights between partners (cf. Robinson & Stuart, 2001a). This would be an excellent area for empirical research; I do not know of any studies that examine this possibility in detail. 6. There are numerous reasons why a firm in a strategic alliance might choose to acquire an equity stake in its partner. Because several of these reasons relate to concerns about controlling the behavior of the partner, the presence of an equity investment as part of the
Governing Strategic Alliances
205
contractual terms of a deal has been treated as a proxy for the perception of one firm that its partner might behave opportunistically. For example, large minority equity positions often convey the right to a seat on the board of directors of a company, and thus the ability to monitor the company’s activities. Also, possessing a block of equity obviously entitles the owners of the block to vote those shares in matters of corporate governance. Moreover, when an alliance participant controls a substantial equity stake in one of its partners, it may reduce the likelihood that the partner will enter a long-term, equity-based relationship with a competitor of the equity holder. Finally, when one firm in an alliance owns a percentage of its partner’s equity, the prospect of diminishing the market value of the equity counterbalances the incentive of that organization to behave in a manner that undermines the prospects of its partner. 7. Given the methodology used to estimate the regressions, the analysis will also address the issue of how the network affects the formation of exchange relationships. However, this is an ancillary objective of the chapter. 8. Three companion papers, Stuart and Robinson (2001), Robinson and Stuart (2001a, b) take up the related issue of the “control” benefits of network embeddedness in transaction content, management, governance structure, and other contractual features. 9. Coleman (1990, pp. 97–100) identifies three conditions which, when present, distinguish transactions as requiring the placement of trust: (i) the transaction enables the trustee to undertake an action that otherwise would not have been possible; (ii) the trusting party places resources at the disposal of the trustee without any enforceable guarantee that the resources will not be misused; and (iii) the trusting party is generally made better off if the trustee is reliable, and worse off if the trustee proves to be untrustworthy (i.e. the trusting party exposes himself to the risk that the trustee behaves opportunistically). Based upon this conception of trust, many biotech strategic alliances qualify as contexts in which the placement of trust is necessary. 10. In the database used for the analysis, the party that performs most of the research or licenses technology to its partner (i.e. the agent in the transaction) is always a dedicated biotechnology firm. The principal can be either a pharmaceutical firm or a biotechnology firm; the organizational form of the principal varies across deals. Roughly one fourth of alliances in the dataset have been established between two biotech companies; the remaining 75% of the transactions in the sector pair a biotechnology company with a pharmaceutical or life sciences firm. 11. This deal illustrates the rationale behind the use of the principal-agent terminology. In this and other alliances in the database, one organization provides resources (in this case, Novartis) to its alliance partner. The recipient of the funds often has considerable discretion as to how these resources are deployed; although contracts are clear on guiding resource deployment, the principal’s monitoring ability is often limited. 12. Obviously, when negotiating the terms of a transaction, the principal will demand contractual safeguards aimed at protecting its interests during the implementation of the alliance. Although I will not review the work on the subject here, there are a number of studies that argue that it is not possible to write complete contracts to govern uncertain activities, such as research and development (e.g. Pisano, 1989). This fact also increases the principal’s exposure to opportunistic behavior. 13. One might wonder if there is yet another sample selection problem, which arises because not all firms participate in the strategic alliance network. This would occur when firms are screened before achieving access to the network, giving rise to a two-threshold selection process. First, individual firms are selected in the network; second, exchange
206
TOBY E. STUART
partners are formed through a network-based matching process. I cannot examine this issue because I lack the full roster of industry participants (i.e. knowledge of biotech firms that have never established alliances), but fortunately there are reasons to believe that this problem will not affect the reported results. First, because alliances are the sole source of revenues for most early-stage biotechnology companies, most firms enter the network very quickly (or fail). Second, it is unlikely that characteristics of the interfirm network determine which organizations gain initial access to it. Thus, although there may be an earlier selection process, it does not necessarily bias estimates of transaction characteristics on network effects. 14. Obviously, this implies that the average principal does more deals than the average agent. This could raise an additional problem in the empirical analysis: because the same principals appear in multiple transactions, the residuals may be correlated within principals. I attempt to address this problem by estimating robust standard errors clustered on principals, but the resultant standard errors are virtually identical to those from the regressions that assume transactions are independent. 15. The duration of the relationships are not reported for most of the alliances in the dataset. When contract terms were specified, typical lengths were three, five, or seven years. Of course, many of the contracts specify the option to extend the term of the agreement. 16. The parameter estimates indicate that the effect of time since agent’s first deal peaks just before nine years, which is about the 95th percentile of that variable’s distribution.
ACKNOWLEDGMENTS I extend my thanks to the editors of this volume for extensive comments on a previous draft of the chapter. I would also like to acknowledge the contribution of David Robinson, who was instrumental in preparing the dataset used in this chapter and is a co-author on three related papers.
REFERENCES Bala, V., & Goyal, S. (2000). A noncooperative model of network formation. Econometrica, 68, 1181– 1229. Batenburg, R. S., Raub, W., & Snijders, C. (2003). Contacts and contracts: Dyadic embeddedness and the contractual behavior of firms. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 135–188). Oxford: Elsevier Science. Berle, A. A., & Means, G. C. (1932). The modern corporation and private property. New York: Commerce Clearing House. Bradach, J. L., & Eccles, R. (1989). Price, authority, and trust: From ideal types to plural forms. Annual Review of Sociology, 15, 97–118. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Buskens, V. (2002). Social networks and trust. Boston, MA: Kluwer.
Governing Strategic Alliances
207
Buskens, V., Batenburg, R. S., & Weesie, J. (2003). Embedded partner selection in relations between firms. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 107–133). Oxford: Elsevier Science. Buskens, V., & Raub, W. (2002). Embedded trust: Control and learning. Advances in Group Processes, 19, 167–202. Buskens, V., Raub, W., & Snijders, C. (2003). Theoretical and empirical perspectives on the governance of relations in markets and organizations. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 1–18). Oxford: Elsevier Science. Buskens, V., Raub, W., & Weesie, J. (2000). Networks and contracting in information technology transactions. In: J. Weesie & W. Raub (Eds), The Management of Durable Relations (pp. 77–80). Amsterdam: Thela Thesis. Coleman, J. S. (1990). Foundations of social theory. Cambridge, MA: Belknap Press. Dore, R. (1983). Goodwill and the spirit of market capitalism. British Journal of Sociology, 34, 459– 482. Granovetter, M. (1985). Economic action and social structure: A theory of embeddedness. American Journal of Sociology, 91, 481–510. Grossman, S. J., & Hart, O. D. (1986). The costs and benefits of ownership: A theory of vertical and lateral integration. The Journal of Political Economy, 94, 691–719. Gulati, R. (1995a). Does familiarity breed trust? The implications of repeated ties for contractual choice in alliances. Academy of Management Journal, 38, 85–112. Gulati, R. (1995b). Social structure and alliance formation patterns: A longitudinal analysis. Administrative Science Quarterly, 40, 619–652. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104, 1439–1493. Hart, O., & Moore, J. (1988). Incomplete contract and renegotiation. Econometrica, 56, 755–785. Kranton, R. E., & Minehart, D. F. (2000). Networks versus vertical integration. RAND Journal of Economics, 31, 570–601. Lincoln, J. R., Gerlach, M. L., & Takahashi, P. (1992). Keiretsu networks in the Japanese economy: A dyad analysis of intercorporate ties. American Sociological Review, 57, 561–585. Macaulay, S. (1963). Noncontractual relations in business: A preliminary study. American Sociological Review, 28, 55–67. Mayer, K. J. (2000). Can contracts facilitate trusting relationships? A case study of software contracting. Manuscript, University of Southern California. Mizruchi, M. S., & Galaskiewicz, J. (1993). Networks of interorganizational relations. Sociological Methods and Research, 22, 46–70. Pisano, G. (1989). Using equity participation to support exchange: Evidence from the biotechnology industry. Journal of Law, Economics, and Organization, 5, 109–126. Podolny, J. M., & Page, K. L. (1998). Network forms of organization. Annual Review of Sociology, 34, 57–76. Powell, W. W. (1990). Neither market nor hierarchy: Network forms of organization. Research in Organizational Behavior, 12, 295–336. Powell, W. W., Koput, K. W., & Smith-Doerr, L. (1996). Interorganizational collaboration and the locus of innovation: Networks of learning in biotechnology. Administrative Science Quarterly, 41, 116–145. Raub, W., & Weesie, J. (1990). Reputation and efficiency in social interactions: An example of network effects. American Journal of Sociology, 96, 626–654. Robinson, D. T., & Stuart, T. E. (2001a). Financial contracting in biotech strategic alliances. Mimeo: University of Chicago.
208
TOBY E. STUART
Robinson, D. T., & Stuart, T. E. (2001b). Just how incomplete are incomplete contracts? Evidence from Biotech Strategic Alliances. Mimeo: University of Chicago. Rooks, G., Raub, W., Selten, R., & Tazelaar, F. (2000). How inter-firm cooperation depends on social embeddedness: A vignette study. Acta Sociologica, 43, 123–137. Sahlins, M. (1972). Stone Age Economics. Chicago: Aldine. Sorensen, J. B., & Stuart, T. E. (2000). Aging, obsolescence, and organizational innovation. Administrative Science Quarterly, 45, 81–112. Stuart, T. E., Hoang, H., & Hybels, R. (1999). Interorganizational endorsements and the performance of entrepreneurial ventures. Administrative Science Quarterly, 44, 315–349. Stuart, T. E., & Robinson, D. T. (2001). The origins of interorganizational networks. Mimeo: University of Chicago. Uzzi, B. (1996). The sources and consequences of embeddedness for the economic performance of organizations: The network effect. American Sociological Review, 61, 674–698. Voss, T. (2003). The rational choice approach to an analysis of intra- and interorganizational governance. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 21–46). Oxford: Elsevier Science. Williamson, O. E. (1985). The economic institutions of capitalism. New York: Free Press. Williamson, O. E. (1996). The mechanisms of governance. New York: Oxford University Press.
SIZE OF THE PIE AND SHARE OF THE PIE: IMPLICATIONS OF NETWORK EMBEDDEDNESS AND BUSINESS RELATEDNESS FOR VALUE CREATION AND VALUE APPROPRIATION IN JOINT VENTURES Ranjay Gulati and Lihua Olivia Wang ABSTRACT This chapter examines the factors that may influence the total value created in a joint venture (JV) and also the relative value appropriated by each partner in the venture. We look at the effects of both partners’ embeddedness in prior networks of relationships and the asymmetry of business relatedness of two partners with the JV on these two important outcomes. Results of an event study of stock market reaction to JV announcements by the largest U.S. firms during 1987–1996 suggest that both network embeddedness of partners and the asymmetry of business relatedness of two firms with the JV affect the total value creation of all partners but not the relative value appropriation between the partners.
The Governance of Relations in Markets and Organizations Research in the Sociology of Organizations, Volume 20, 209–242 Copyright © 2003 by Elsevier Science Ltd. All rights of reproduction in any form reserved ISSN: 0733-558X/PII: S0733558X02200087
209
210
RANJAY GULATI AND LIHUA OLIVIA WANG
INTRODUCTION The dramatic growth of new joint ventures (JV) announced over the past two decades indicates that JVs have become a very important strategic tool for firms (Geringer & Hebert, 1991). JVs are perceived to provide a variety of economic benefits for firms including economies of scale and scope, market power, learning, getting access to new markets and new technologies (Koh & Venkatraman, 1991). These perceived economic benefits to partners can vary considerably across JVs. Furthermore, many JVs are asymmetric, with the potential for creating considerably greater value for one partner than the other (Powell, 1987). Despite a growing body of research, there is still a great deal we don’t know about what factors influence the extent of value JVs create for participants and what determines the extent of value each partner derives from a JV in relation to other partners. In this chapter, we consider the relative importance of some of the antecedent factors that influence both the total value created for all partners in a venture as well as the relative value appropriated by individual partners. The current literature has typically assessed the value creation potential of JVs by using an event-study methodology to examine stock market reactions to JV announcements. Since accurate data on value created in JVs are onerous if not impossible to collect, scholars have looked at investors’ expectation of whether and how much the new JV will create value for the parent firms. Investors are a distinctive group of people who hold rich information about firms. Their judgment on whether a JV is expected to create value for parent firms and how the created value will be divided between partners is considered to have incorporated all available information. Prior research on stock market reaction of JV announcements has reached the following conclusions: (1) on average, investors expect that strategic alliances (including JVs) create positive value for participating firms (Chan et al., 1997; Koh & Venkatraman, 1991; McConnell & Nantell, 1985); (2) Investors value those JVs with partners operating in related business areas or with partners operating in business areas related to the new JVs more than those with partners and/or new JVs operating in unrelated areas (Balakrishnan & Koza, 1993; Koh & Venkatraman, 1991); (3) Investors expect that the smaller firm in a JV usually extracts a higher percentage of return from the alliance than the larger firm, but the dollar value of gain is similar (Koh & Venkatraman, 1991; McConnell & Nantell, 1985); and (4) Investors expect that firms that are more dedicated to alliance strategies are more likely to learn how to create value from future alliances and thus are more likely to extract high value from the new alliance formation (Anand & Khanna, 2000). One limitation with prior research on JVs results from their primary focus on the economic and strategic factors yet avoidance of some of the important social factors
Size of the Pie and Share of the Pie
211
that may influence JV outcomes (for an exception, see Anand & Khanna, 2000). Economic sociologists have long suggested that firms are embedded in social structures and that such social structures have an important impact on their ongoing economic actions and outcomes (Granovetter, 1985; see the review by Voss, 2003 in this volume). Researchers have extended these ideas to the study of interorganizational relationships and suggested that social embeddedness among firms resulting from prior interactions engenders trust that in turn can alter their subsequent behavior and outcomes (Buskens & Raub, 2002; Gulati, 1995a, b, 1999). Given this prior work, it is likely that the investors may expect that social embeddedness of partners prior to the new JV announcement affect the total value creation of the new JV and the relative value appropriation between partners. Yet, there is no study assessing whether and how investors value the social embeddedness of partners. Furthermore, there is no discussion of the possible interplay of strategic and social factor for these important firm outcomes. Another limitation with prior research results from the fact that prior studies have typically focused on either total value creation of a JV for participating firms in the eyes of the investors (e.g. Koh & Venkatraman, 1991) or the investors’ evaluation of value creation of JV for one of the partners (Anand & Khanna, 2000; Augereau, 1999). They have ignored factors that may influence the relative value appropriation among the partners when a JV is announced (for an exception, see Koh & Venkatraman, 1991). Theoretical accounts of alliances have suggested that not only total value creation in the JV for all partners but also the relative value appropriation from the JV for each partner is important (Gulati et al., 1994; Hamel, 1991). Firms in a JV do not always extract commensurately equal value from a JV. As research on learning races in alliances suggests, there are many alliances in which one partner may race to learn the other’s skills, while the other has no such intentions. As a result, one partner may acquire greater value from the alliance than its partner (Doz et al., 1989; Hamel, 1991; Khanna et al., 1998). We consider two important drivers of total value creation and relative value appropriation. First, we build on the rich literature on interorganizational embeddedness and consider the role of prior firm and dyadic embeddedness for these important alliance outcomes. Second, we consider the role of an important economic factor – the business relatedness of the parent firms to the JV – as another driver of these outcomes. The asymmetry of business relatedness of two parent firms with the JV refers to the extent to which one partner’s businesses are more or less closely related to the JV activities than the other partner’s businesses. If this asymmetry exists, the two partners may have different opportunities and capabilities to extract value from the joint venture. This asymmetry may affect the total value creation of the JV and the relative value appropriation for each partner.
212
RANJAY GULATI AND LIHUA OLIVIA WANG
This study aims to use both a social perspective and an economic perspective to account for the investors’ evaluation of JV announcement. We examine how the extent to which a JV tie is embedded in prior relationships between the two firms and the asymmetry of business relatedness of two partners with the JV affects both the total value creation for all firms in a JV and also the relative value appropriation of each partner in the JV.
THEORY AND HYPOTHESES Network Embeddedness and Total Value Creation in JVs Total value creation refers to the stock market’s expectation on how much total value a JV is going to create for all partners collectively. As a separate entity created by partners, a JV is usually an enduring commitment for a certain period of time in which at least two firms come together to combine their resources in order to achieve some common objectives (Lewis, 1990). Just as in many negotiation situations, it is not only the share of the pie that each party gets that is important but also the total size of the pie (Thompson, 1998). The network embeddedness between parent firms in a JV. Prior studies of the total value creation by two firms from the announcement of a new JV have typically assumed that investors are only concerned with the economic antecedents of the new JV and used firm attributes (such as size) or dyad attributes (such as business relatedness of two firms) as proxies for these factors. This assumption represents an asocial account of interfirm relationships where firms’ outcomes are expected to be solely influenced by economic factors. It is now widely recognized that dyadic economic relationships between firms are often embedded in broader systems of social relations and this embeddedness can have an influence on the economic behavior and outcome of firms (Granovetter, 1985). For example, researchers have found that ties in which firms are embedded resulting from the prior economic relationships affect the pattern of new alliances formed (Gulati, 1995b; see also Buskens et al., 2003 in this volume for the case of buyer-supplier relations), the choice of governance structure for interfirm alliances (Gulati, 1995a; see also Batenburg et al., 2003 and Stuart, 2003 in this volume), and the performance of firms (Stuart, 2000). Gulati (1995b) found that the social relationships resulting from the prior direct and indirect ties between firms may in fact promote trust between two partners. The trusting relationship reduces the hazard of opportunism that is common in alliances and provides the basis for a new alliance to form (Zaheer et al., 1998). Given these differences between interfirm transactions that are and are not embedded in prior relationships between firms, it is likely that investors will consider
Size of the Pie and Share of the Pie
213
the extent to which a new JV is embedded in prior relationships between firms when they anticipate the extent of value likely to be created by a new JV. The social context likely to influence such expectations examined here is the accumulation of prior JV ties between firms. We believe that the more prior direct or indirect ties between two firms resulting from the prior JVs, the more likely two firms are embedded in social relationships (Granovetter, 1985; Gulati, 1995a, b). At two extremes, a new JV can be either formed between two firms with no prior direct or indirect connections or between firms that may have dense connections prior to the new JV announcement. For example, on April 2, 1990, Allied-Signal and Exxon announced a JV with a 50:50 equity distribution. At that time, the two companies had never done a JV together before, and they had not had any JVs with common third partners. In this case, the new JV is not embedded in the prior direct or indirect social relationships between two firms. In contrast, on July 8, 1993, AT&T announced a JV with Harris. By that time, the two companies had already operated two JVs with each other, and they had both previously had JVs with three common third-party firms (IBM, Sprint Corp, Intel). In this case, the two companies are embedded in the prior direct and indirect ties before they formed the new JV. How can network embeddedness resulting from prior direct and indirect JV relationships between two firms be beneficial to partner firms in a JV? Joint ventures involve at least two parent firms who have distinct interests that are not entirely compatible. Each parent firm is vulnerable to its partner’s opportunistic behavior (Doz et al., 1989; Hamel, 1991; Kogut, 1988). Given the different private interests of parent firms, firms may be concerned about their own interests and limit their contribution to the JV at the cost of their partners’ interests (Khanna et al., 1998). In addition, by participating in a JV, firms are not only vulnerable to their own environmental uncertainty but also to changes in their partners’ environment. The risks involved in a JV are significantly reduced by the information provided from the social network of prior JV ties (Gulati, 1995b). Firms learn about the reliability, specific capabilities and trustworthiness of current and potential partners from the network of prior JV ties. The information serves as important basis for trust between two firms in a new JV. The knowledge obtained from this network of prior JV ties enables the JV partners to have a more accurate expectation about their partners and also makes the behavior of the partners more predictable. This knowledge-based trust reduces the concern for uncertainty involved in the JV. Buskens and Raub (2002) refer to this mechanism as the learning mechanism of network embeddedness. The circulation of information within the network also raises reputational concerns for the firms and prevents them from acting opportunistically (Gulati, 1995b). If both firms in a new JV have a previous JV with a common third party, either partner’s behavior can be reported to the third party and
214
RANJAY GULATI AND LIHUA OLIVIA WANG
in turn may further spread through the wider network. The anticipation of future interaction and the desire for developing long-term relationships between partners also prevent the firms from acting opportunistically. Such reputation concerns and the anticipation of future interaction are also referred to as the control mechanism (Buskens & Raub, 2002). There are two distinct components of network embeddedness between two firms – the relational embeddedness resulting from prior direct ties between two firms and the structural embeddedness resulting from the prior indirect ties (Gulati & Gargiulo, 1999). Relational embeddedness between two firms in a new JV provides the experience-based information about each other while structural embeddedness between two firms in a new JV provides indirect information about each other through prior common partners of the two firms. We examine how these two components of network embeddedness between two firms affect the investors’ evaluation of total value creation of the new JV formation in turn. Relational embeddedness and total value creation. Relational embeddedness captures the effect of a direct prior relationship between two partners on the investor’s expectation of a new JV they may enter. Relationally embedded partners in a new JV have had the opportunities to directly interact with each other through prior JV(s). The intense interaction through the operation of prior JV(s) can result in relation-specific assets accumulated by the partners that can be exploited in a new relationship (Dyer & Singh, 1998; Zaheer & Venkatraman, 1995). These relationspecific assets include the valuable information about each other that is not possible to get outside the close relationships, a trusting relationship developed through intense interaction, and the development of dyadic absorptive capability based on their intimate knowledge of each other’s operating routines. These assets in turn can make future interactions in new JVs less uncertain and less costly by limiting transaction costs (Gulati, 1995a). Given that partners in a JV do not necessarily always have perfectly compatible interests, JVs provide plenty of opportunities for participants to act opportunistically in order to pursue their own private interests. Transaction costs are likely to be reduced in several ways if the partners of a new JV are embedded in prior direct ties (Dyer, 1997). First, monitoring costs may be reduced because prior direct interactions promote interfirm trust that reduces the incentive for opportunistic behavior (Gulati, 1995a). Second, negotiation costs are reduced because the partner firms develop deeper knowledge of each other in terms of their interests, needs, and capabilities through prior connections, and information exchange can be more efficient. Moreover, the trusting relationship of the partners facilitates open communication and private information exchange. As prior research suggests, communication and information exchange are considered to be keys in reducing the negotiation costs and increasing the chance of integrative solutions (Thompson, 1998).
Size of the Pie and Share of the Pie
215
In addition to the above information and learning benefit, relational embeddedness resulting from the prior direct ties between two firms also prevents firms from acting opportunistically and promotes cooperative behavior if they expect that they will have future interactions (Buskens & Raub, 2002). Raub and Weesie (1990) refer to this mechanism as the reputation effect in the narrow sense. Game theorists studying the repeated Prisoner’s Dilemma or other social dilemma problems have argued that partners who anticipate future interaction with each other are more likely to trust each other and less likely to behave opportunistically (Axelrod, 1984; Camerer & Weigelt, 1988; Kreps et al., 1982; Taylor, 1976). Second, prior direct ties also play a role in making a new JV potentially more fruitful by reducing coordination costs that encompass the anticipated complexity of decomposing tasks among partners along with ongoing coordination of activities to be completed jointly or individually across organizational boundaries and the related extent of communication and decisions that would be necessary (Gulati & Singh, 1998, p. 4).
As partners build up experience with each other, they not only have more information about each other’s strategies, culture, and capabilities, they also develop shared norms of behavior and routines of joint decision making and coordination (Walker et al., 1997). As a result, coordination costs are reduced. Finally, prior direct ties may enhance the prospects for joint learning. Mutual learning is perhaps one of the most important sources of value creation of a JV (Dyer & Singh, 1998; Khanna et al., 1998). The ability of a firm to recognize and assimilate valuable knowledge from a particular partner is called partner-specific absorptive capability and should be heightened as firms develop routines to work together (Cohen & Levinthal, 1990; Dyer & Singh, 1998). Given the various benefits JV partners may enjoy from the relational embeddedness resulting from their prior direct ties, we hypothesize Hypothesis 1a. The degree of relational embeddedness between the new JV partners has a positive relationship with the total value creation of the JV. While prior direct ties do indeed have many beneficial consequences, it is possible that the benefits from prior direct ties may have a diminishing return such that when the number of direct ties between two firms increases beyond a certain point, the collaboration opportunities may reach its limit and do not provide additional benefits for firms if they establish another JV (Gulati, 1995b). Indeed, two firms that have many prior direct ties may be at risk of overly depending on each other through their significant investments in relation-specific assets. The increasing interdependence between the two firms places them in a vulnerable situation in case one or both of their priorities change. There is evidence that when a firm is
216
RANJAY GULATI AND LIHUA OLIVIA WANG
overly dependent on its co-opetitors, be they customers, suppliers or collaborators, the very source of competitive advantage from the collaborative relationship may adversely affect the performance of both firms if the technology of one of the firms becomes obsolete or the environment of one or both firms abruptly changes (Afuah, 2000; Lorenz, 1988). The relation-specific assets of firms developed through their prior JV ties not only create a situation where two firms overly depend on each other, but can also encourage the two firms to keep investing in those relation-specific assets to reap immediate benefits. This may come at the expense of their under-emphasizing the exploration of new opportunities by forming ties with new partners with whom they have not had prior experience. Unlike a JV between firms with prior ties, a new JV between firms that have no prior direct ties may also be important for firms’ competitive advantage as it may allow them to extend their reach and explore new and unique rewarding opportunities. Although risky at the outset, the exploration of new relationships can enhance each firm’s opportunity set in the long run by providing them with access to the other party’s information sources on new deals with additional partners. This access gives each firm new opportunity for referrals to rewarding opportunities within the network of the new partner (Gulati, 1995b). New JVs between firms with no prior direct ties may also benefit participants by providing them with opportunities for sharing complementary resources or pooling common resources to enhance economies of scale and scope (Ahuja, 2000), while also allowing them to acquire new and unique skills that could potentially be used in their own activities outside the scope of the JV (Khanna et al., 1998). Finally, new JVs between firms with no direct ties can create potential future opportunities for the two firms to further cooperate with each other and thus further exploit the benefit of cooperation. The above discussion indicates that as the number of direct ties between two firms increases, the benefits accrued to partners from a new JV might increase and then diminish beyond a certain point. As two partners become more relationally embedded, they become overly dependent on each other and under-explore new opportunities. This argument is consistent with the trade-off between exploration and exploitation suggested in the organizational learning literature (March, 1991). Exploration refers to the behavior of a firm trying to discover new information about alternatives and improve future returns. Exploitation refers to the behavior of a firm using the information currently available to it to improve present returns. Koza and Lewin (1998), who have applied these notions to the study of JVs, suggested that the formation of an alliance is an indication of a firm’s adaptive choice between exploration and exploitation. On the one hand, exploration of new opportunities by the participants with no prior direct interactions may be more risky, but it might also be more beneficial. On the other hand, exploitation of the
Size of the Pie and Share of the Pie
217
benefits of existing cooperative relationships by participants with prior direct ties may be safer but less beneficial in the future (March, 1991). As a result, investors are likely to expect that new JVs between firms with intermediate number of direct ties balance exploration of new opportunities with exploitation of existing ones. Based on the above rationale, we also provide an alternative hypothesized relationship between relational embeddedness and total value creation of a JV. Hypothesis 1b. The degree of relational embeddedness between the new JV partners has an inverted U-shape relationship with the total expected value creation of their new joint ventures, with the highest value occurring at an intermediate degree of relational embeddedness. Structural embeddedness and total value creation. Structural embeddedness captures the effect of network embeddedness resulting from the prior indirect ties between two firms. In many instances, two firms may not be relationally embedded, or have prior direct ties, but they may be linked by indirect ties through common partners. Here we consider the simplest case where two firms forming a JV have a common third firm with whom both firms have a direct tie. When two structurally embedded firms announce a JV, investors likely consider two benefits of such indirect ties in enhancing the value creation potential of the new JV: information and reputation. The common third party shared by two firms in a JV serves as an important conduit for information about each other. The information reduces uncertainty about the behavior of the partner and serves as the basis for trust (Burt & Knez, 1995; Gulati, 1995b). The fact that the two firms announce a new JV may also provide a signal to investors that both are considered by the third party as suitable and trustworthy. Investors expect that such JVs have a high probability to create value for both firms. The dual connection with the common third firm also raises reputation concerns for both firms because the opportunistic behavior can be reported to the third common firm by either partner and this information can in turn spread to wider set of firms in the alliance network (Ahuja, 2000; Gulati, 1995b; Raub & Weesie, 1990). Such reputational concerns resulting from structural embeddedness play a significant role in problematic social situations (Raub & Weesie, 1990). This has been widely studied in interpersonal relationships (Burt, 1992; Burt & Knez, 1995; Coleman, 1988), Prisoner’s Dilemma games (Kreps, 1990; Raub & Weesie, 1990), trust situations (Buskens, 1998), and interfirm relationships (Gulati, 1995b). This body of research demonstrates that people or firms are motivated to promote their good reputations to allow for future beneficial transactions (Kreps, 1990). Reputation is a valuable economic asset that individuals and firms build, maintain and exploit for efficient transactions. As a concomitant, reputation loss is very costly,
218
RANJAY GULATI AND LIHUA OLIVIA WANG
especially in the contemporary context where firms are frequently partnering. We expect that the fear of reputation loss promotes good behavior of both partners and this in turn increases the value creation potential of a new JV. The more structurally embedded two firms are, or the more indirect ties two firms have, the richer the information two firms can get access to about each other (Gulati, 1995b; Gulati & Gargiulo, 1999). The richer information is more reliable and makes the behavior of partners more predictable in the new JV operation. This in turn enhances the effectiveness of the new JV and thus its expected value creation potential. In addition, increasing numbers of indirect ties also provide more behavioral constraints for firms to behave in a trustworthy way that benefits the new JV operation. A large number of indirect ties between two partners in a newly announced JV indicates a high level of cohesiveness of the local network. Such cohesive networks prevent opportunistic behavior because such behavior can be diffused quickly to a large number of other directly or indirectly connected partners (Coleman, 1988). The threat of reputation loss to multiple partners prevents the firm from behaving opportunistically with any single partner (Ahuja, 2000). Investors may expect that new JV partners with more indirect ties are more likely to be subject to such information benefits and social constraints and thus more likely to behave in a way that benefits the new JV. Therefore, we hypothesize a positive relationship between the degree of structural embeddedness of JV partners and total value creation of the JV. Nevertheless, it is possible that as two partners become too structurally embedded, or they have more indirect ties, the benefits provided by the information may diminish as the information becomes more redundant and less valuable (Burt, 1992; Coleman, 1988). Multiple common third parties may have similar information about partners and the value of the information may not increase proportionally with the increase of the indirect ties between partners. In terms of reputation concerns, as two firms have more common partners and the local network becomes extremely cohesive, behavioral constraints may develop within the whole network. Two firms may have to follow the norms of the local network (Coleman, 1988). These norms, however, may become a constraint for the two firms for their strategies (Baum & Oliver, 1991). Based on this argument, we suggest an alternative hypothesis that the degree of structural embeddedness of two partners may have a diminishing positive relationship with the investors’ evaluation of a new JV. Two firms may be both relationally embedded and structurally embedded and the effects of them cannot be separated easily. To separate out the effect of structural embeddedness, we only examine those cases where two partners are structurally embedded, but not relationally embedded in the following two hypotheses.
Size of the Pie and Share of the Pie
219
Hypothesis 2a. In the absence of relational embeddedness, the degree of structural embeddedness between two JV partners has a positive relationship with the total value creation of the JV. Hypothesis 2b. In the absence of relational embeddedness between two firms, the degree of structural embeddedness between two firms has a positive diminishing effect on the investors’ expectation of total value creation of the new JV.
Network Embeddedness and Relative Value Appropriation in JVs The relative value appropriation of each partner in a JV refers to the relative proportion of total value creation that each partner extracts from the JV. It is important because the ultimate goal of each partner in a JV is to extract its own value from the JV. Hamel (1991) was perhaps the first to highlight the importance of not only the total value created in a JV but also the relative value appropriated by each partner. Several recent studies have built on these ideas and examined how interfirm partnerships may turn into arenas for learning races in which each partner hopes to gain the greatest benefits from the alliance (Khanna, 1998; Khanna et al., 1998, 2000). In spite of its importance, few empirical studies have examined the factors that affect the relative value appropriation between partners in a JV. This study is one of the few to examine this issue with a large sample of JVs. Relational and structural embeddedness between two firms and relative value creation. The relational embeddedness resulting from prior direct partnering experiences between two firms serves as important basis for trusting relationship to develop (Granovetter, 1985). The trust between two partners encourages reciprocity and considerations of mutual benefits. In the process of dividing the total pie, partners are likely to consider not only the individual costs and benefits but also the mutual costs and benefits. They are less likely to emphasize claiming the individual economic returns than maintaining a more harmonious and mutually fruitful cooperative relationship. When two firms are relationally embedded, they are more likely to view the JV transaction as part of a long-term relationship, and this orientation reduces the desire for the firms to seek asymmetric returns (Dyer, 1997). As a result, in this case, investors are likely to expect symmetric returns to both partners entering a JV. While prior direct ties between two partners provide first-hand information about each other and the opportunity to interact with each other directly, prior indirect ties provide similar types of information through common third parties. Third parties are important go-betweens in new relationships enabling individuals to transfer their expectations from well-established relationships to others in which adequate
220
RANJAY GULATI AND LIHUA OLIVIA WANG
knowledge or history was not yet available. Since both firms have direct interaction with the third common party, the common party passes the information about the trustworthiness of the two firms to each other, this information provides a basis for the trusting relationship to develop. Investors witnessing the announcement of a JV are likely to expect that prior indirect ties promote trust between the parent firms and are likely to encourage cooperative behavior leading to parity in the benefits each party extracts from the JV. In contrast, firms that are not relationally or structurally embedded with each other lack accurate information about each other and are not likely to have a trusting relationship. They are more likely to consider the relationship with private costs and benefits in mind and view the relationship in terms of power and dependence. Private benefits are likely to be emphasized over mutual gains. Furthermore, since firm managers may feel uncertain about the strategic intent of their JV partners, they are more likely to engage in learning races to extract as many private benefits as possible in the quickest possible way (Khanna et al., 1998). With this orientation, investors are likely to expect that the relative value extraction of the two firms from the JV is more likely to depart from an equal sharing of the expected total value created. The above discussion suggests the following: Hypothesis 3. As two firms are more relationally embedded, the relative value appropriation to each partner in the JV becomes more symmetric. Hypothesis 4. As two firms are more structurally embedded, the relative value appropriation to each partner in the JV becomes more symmetric.
Asymmetry of Business Relatedness of Partners with the JV and Its Impact on Total Value Creation and Relative Value Appropriation Business relatedness has been widely studied in the literature on diversification strategy and mergers and acquisitions by firms. In these two literatures, the term refers to the similarities in products, markets, and technologies among business units within a diversified firm (Rumelt, 1974) and to the similarities in products, markets, and technologies between acquiring and acquired firms respectively (Datta & Puia, 1995). Current research suggests that a high level of business relatedness is beneficial to firms both for diversification and for mergers and acquisitions because it creates value for firms by enhancing economies of scale or scope, increasing market power and providing access to necessary technology in familiar industries (Datta & Puia, 1995; Seth, 1990). The rationale for the benefits of business relatedness has also been extended to the study of JVs and is variously described as market overlap or technology overlap between partnering firms in a
Size of the Pie and Share of the Pie
221
JV. Recent research suggests that greater overlap between JV partners positively affects the firm’s capacity to learn from its partner, and, in turn, positively affects the economic rent the partners can extract from the partnerships (Dyer & Singh, 1998; Mowery et al., 1996). Stuart (2000) suggested that technology overlap facilitates effective collaboration among firms in an alliance because (1) the firms are better at evaluating and internalizing each other’s technologies because they share the knowledge of the technologies and the market segments and (2) the firms provide high value for each other in terms of information exchange and cost sharing, especially in a crowded technological area. Because JVs involve the creation of a third neutral entity, not only the overlap between two partners but also the overlap between each of the parent firms’ core activities and the activities undertaken by the JV are important. In a JV partnership, each partner’s business operation can be either related or unrelated to that of the JV. When the business of a focal firm is related to the JV, it is likely that the firm can use its core competencies in the operation of the JV and further exploit any potential economies of scale with its current resources. Relatedness is also likely to increase the market power of the focal firm (Pfeffer & Nowak, 1976). In addition, it provides potential opportunities for spillover effects that allow the focal firm to apply the skills learned from the new JV to its other activities and thus fully exploit new learning opportunities (Doz et al., 1989). It is also likely that firms that are closely related to their JV operations may have greater absorptive capacity for knowledge coming from the areas of development in the JV, reflected in a greater ability to recognize the value of the new knowledge, assimilate it, and apply it for their economic benefits (Cohen & Levinthal, 1990). Building on the research in individual learning, the research on absorptive capacity suggests that a firm’s absorptive capacity is closely associated with the extent to which the firm possesses prior related knowledge about the technology being developed outside its boundaries. Applying this logic to our context, a firm operating in areas related to the JV has prior related knowledge, is familiar with the JV operation, and thus is likely to have the ability to recognize and absorb the new knowledge and information from the JV and exploit it in the rest of its operations. The business relatedness of a firm with the JV thus not only provides the firm with potential opportunities for exploiting economies of scale and scope as well as learning opportunities but also provides conditions that enhance the ability of the firm to realize such opportunities. When assessing the potential value creation of a JV, the business relatedness of each of the parent firms to the JV activities becomes significant. The dynamics of the relatedness of their business operations to those of the JV may affect investors’ expectations of the total value created from the JV. Specifically, asymmetry of business relatedness for each of the partners with the JV may negatively
222
RANJAY GULATI AND LIHUA OLIVIA WANG
affect the investors’ expectations about the future cooperative relationship between the partners and thus the total value that may be created in a JV (Borys & Jemison, 1989; Harrigan, 1988). When one partner’s business is related to the JV, while the other partner’s is not, the partners themselves are likely to value the JV differently because the benefits of economies of scale, scope and learning opportunities from the new JV are not symmetric. Investors may expect that the firm that is related to the JV may extract private benefits that it can use in its own operations before both firms are able to obtain any common benefits (Khanna et al., 1998). In addition, in such asymmetric JVs, the ability of the two partners to learn and absorb the new knowledge is also likely to be asymmetric. This may lead the advantaged firm to decide to race with its partner to learn the partner’s skills as fast as possible from the JV operation before the other party realizes its learning objective (Khanna et al., 1998). The asymmetry of knowledge transfer benefits between the two partners may lead to shifts in the bargaining power in a JV, which may discourage knowledge-sharing between partners and thus reduce the joint value creation of the two partners from the JV (Chi, 2000; Inkpen, 1998). Moreover, it also encourages the partners to expand resources to learn as much as possible from the other partner for private benefit while making fewer efforts to take advantage of the potential synergy between the partners for common benefits (Chi, 2000). The firm that receives fewer benefits may not be able to achieve its strategic goals and may become dissatisfied with the JV in the future. According to the literature on fairness and relative deprivation, individuals tend to view the benefits each participant extracts from the relationship as more fair and reciprocal if they are equal, which then are more likely to lead to a positive and stable relationship. The likely unequal and (possibly unfair) relationships in asymmetric JVs may cause investors to expect that the future cooperative relationship between the partners will not be fruitful and the JV may not perform well. Therefore, investors’ expectation about the total value creation of the JV may be lower on the announcement of the JV given the asymmetry of business relatedness of two firms with the JV: Hypothesis 5. Asymmetry of business relatedness of the parent firms with the JV is negatively related to the total value creation of the JV. As mentioned above, the asymmetry of business relatedness of two firms with the JV is likely to contribute to the partners extracting different levels of value from the JV. Firms whose activities are more closely related to the JV are more likely to extract greater value from the JV than their more unrelated partners. As a result, investors will expect that the asymmetry of business relatedness of the two partners with the JV may promote greater differences in the likely value extracted by the two partners. When the two partners both have similar relatedness to the JV
Size of the Pie and Share of the Pie
223
operation, however, their potential opportunities to extract value are more likely to be similar. Based on the above discussion, we hypothesize, Hypothesis 6. The asymmetry of business relatedness of two partners with the JV is associated with greater departure from an equal share of the created value of the venture.
METHODS Sample and Data Our sample includes all bilateral JVs formed among the 300 largest Fortune/Forbes firms (in 1988) from 1987 to 1996. We excluded from the analysis 20 firms that did not have any financial data available from the COMPUSTAT industrial firms database, leaving us with 280 firms. Among these 280 firms, 186 firms had at least one JV with another firm in the sample, generating a total number of 658 JVs in the final sample. Figure 1 gives the frequency distribution of JVs during 1987–1996. We collected our data from a variety of sources. The longitudinal data on the announcement of bilateral JVs came from two major sources: Lexis-Nexis and Security Data Corporation (SDC). Rich descriptions of JVs in Lexis-Nexis were
Fig. 1. Number of Joint Ventures by Year (1987–1996).
224
RANJAY GULATI AND LIHUA OLIVIA WANG
complemented and cross-validated by SDC data. We resolved duplications and contradictions between these sources by consulting additional sources, such as the Wall Street Journal Index. We also collected data on prior JVs for the sample firms from 1970 to 1986, to control for firm history. These data are all from Lexis-Nexis because SDC data only go back to 1986. We collected data on stock market reactions to JV announcements from the Center for Research in Security Prices (CRSP). For each firm, we obtained data on stock market movements for that firm over 241 days (250 days before the announcement until 10 days before the announcement) and over a two-day period (1 day before the announcement and the announcement day) as well as over a 21day period (10 days before the announcement until 10 days after the announcement. Event periods (windows) are used in event studies to capture the possible stock market reaction to the information leakage prior to the announcement of the JV and additional information available to the market after the JV announcement (Anand & Khanna, 1995; Datta & Puia, 1995). We used both the most widely used two-day event period (Anand & Khanna, 1995; Chan et al., 1997; Datta & Puia, 1995) and a longer event period of 21 days (Black & Grundfest, 1988; Bradley et al., 1983; Datta & Puia, 1995; Sicherman & Pettway, 1987).1 The methodology is described below in more detail. For all sampled firms, we also collected financial data from 1986 to 1995 (inclusive) from COMPUSTAT. These financial data are lagged by one year, i.e. we use the financial data of 1986 to predict the announcement of JV in 1987, etc. Variables Abnormal returns. A firm’s abnormal return reflects the expected return that the market believes the firm will capture by participating in a JV (Anand & Khanna, 2000). We use the most widely used market model, based on residual analysis, to calculate the firm’s abnormal return (Fama et al., 1969). We designated the event (JV announcement) date as t = 0. Accordingly, t is equal to −10 if the date is 10 trading days before the announcement date and t equals 10 if the date is 10 trading days after the announcement and so forth. We used daily data on the stock market returns of each firm over a period 241 days prior to the event day (250 days before the announcement of the JV until 10 days before the announcement of the JV) to estimate the market model (Binder, 1998; Fama, 1970): r it = ␣i + i r mt + it , where rit is the common stock return of firm i on day t, rmt is the corresponding daily market return on the equal-weighted S&P 500, ␣i and i are firm-specific parameters, and it is the error term.
Size of the Pie and Share of the Pie
225
We used the estimates from the above model to predict the daily returns for each firm i over a two-day period and a 21-day period surrounding the event date using the following equation: rˆit = estimated ␣i + estimated i × r mt , where rˆit is the predicted daily return and estimated ␣i and estimated i are the model estimates. Next, the daily firm-specific abnormal returns are calculated as: Estimated it = r it − rˆit . We calculated the cumulative abnormal return of firm i during the event period by adding up the daily abnormal returns of the firm over the event period, i.e. (Estimated it ), CARi = t
where t is either from −1 to 0 or from −10 to 10. Total value creation and relative value appropriation. We used total return and relative return to measure total value creation and relative value appropriation, respectively. Total return refers to the aggregated abnormal return of the two firms involved in a JV. It reflects the total anticipated stock market gain of the two firms from the JV. Relative return compares the abnormal returns of the two participating firms entering the JV. It reflects the asymmetry of the two firms’ abnormal returns from the JV participation. Both variables were calculated from the firms’ abnormal returns. The literature describes two ways to calculate the value creation of a new JV to the partners – equal-weighted single security and value-weighted single security (Anand & Khanna, 1995). The value-weighted single security approach aggregates the abnormal return of the participating firms, using the market value of the firms as weights, while the equal-weighted single security approach gives each participating firm equal weights. We follow McConnell and Nantell (1985) in using the equal-weighted approach. In those studies in which the value-weighted approach was used, one of the major interests was to find out the absolute total dollar value creation from the announcement of JVs. Here we were only interested in percentage gains and their correlation with the respective independent variables. In addition, we view the two partners as equally important. Therefore, we chose the equal-weighted approach. Total return was calculated as the simple addition of the two firms’ cumulative abnormal return over the event period, i.e. Total return = (CAR1 + CAR2 ),
226
RANJAY GULATI AND LIHUA OLIVIA WANG
where CAR1 and CAR2 are Firm 1’s and Firm 2’s respective cumulative abnormal returns over certain event period (we use two-day and 21-day event periods in this study). We calculated relative return by first taking the exponential of the two firms’ abnormal returns and then using the ratio of the absolute difference of the two transformed returns and the addition of the two transformed returns to represent the relative return: Relative return =
exp(max(CAR1 , CAR2 )) − exp(min(CAR1 , CAR2 )) . exp(min(CAR1 , CAR2 )) + exp(max(CAR1 , CAR2 ))
The transformation corrects for the problem of a possible mixture of positive versus negative cumulative returns for the two firms entering the same JV. For example, if Firm 1’s cumulative abnormal return is −0.02 while Firm 2’s cumulative abnormal return is 0.001, then the absolute difference 0.021 is not very intuitive. Thus, we used the exponential of their cumulative abnormal returns to transform them into positive numbers. For example, if Firm 1’s transformed return is exp (−.02) = .98, and Firm 2’s transformed return is exp (0.01) = 1.01, then the relative return is the absolute difference of the two transformed returns, i.e. .03. This absolute difference, though, treats the difference between a return of 0.01 and 0.02, for example, the same as the difference between 0.49 and 0.50. To correct this problem, the absolute difference of the two above transformed returns is divided by the addition of the two transformed returns. To continue the above example, the final calculation of the relative return is: .03/(.98 + 1.01) = .015. When the value of relative value appropriation is 0, the investors expect equal share of the pie between two firms. The larger this value is, the value creation of two firms are further away from the equal share of the pie. Relational embeddedness and structural embeddedness. Relational embeddedness is indicated by the number of prior direct JV ties between two firms while structural embeddedness is indicated by the number of common third party ties (indirect ties with the distance of two) between two firms. The number of prior direct ties is measured by counting the number of JV ties between two firms from 1970 to the prior year of JV announcement. The number of indirect ties is calculated by counting the number of all the common third firms that both parent firms have had a JV with from 1970 until the prior year the JV was announced. This includes those third common firms outside our sample. Asymmetry of business relatedness of two firms with the JV. We used the first two digits of the primary Standard Industrial Classification (SIC) codes of the two firms and the primary SIC code of the JV to calculate their respective business relatedness with the JV. The business relatedness of firm A with the JV equals 1 if the first two digits of the primary SIC code of firm A are the same as the
Size of the Pie and Share of the Pie
227
first two digits of the primary SIC code of the JV. The asymmetry of business relatedness of two firms with the JV equals 1 if only one of the partner firms has the same SIC code as the new JV and equals 0 if both or neither firms have the same SIC code as the new JV. Our reliance on the first two digits of the SIC code to indicate business relatedness is consistent with Palepu (1985). Although subject to criticism, there is evidence that the congruence between this approach and finer-grained approaches is very high if the sample size is large (Montgomery, 1982). Control variables. We included some economic variables and firm characteristics as control variables. These variables are considered to be very important in prior studies. First, we controlled for the business relatedness of two parent firms. It is an important factor for value creation of a new joint venture (Koh & Venkatraman, 1991). It is measured as a binary variable with a value of 1 if the two parent firms have same two-digit primary SIC codes and 0 otherwise. We also included firm size and firm performance indicators (return on assets and solvency) as control variables, lagged by one year. These variables have been viewed as important in previous studies of alliances (Anand & Khanna, 1995, 2000; Chan et al., 1997; Gulati & Westphal, 1999). Firm size was indicated by total sales. Solvency was calculated as long-term debt divided by current assets. Because the unit of analysis was a dyad, we created control variables for total size by summing across the two partners’ total sales and taking a log. We also created control variables of total return on assets (ROA) and total solvency as the simple additions of the respective return on assets and solvency of the two firms. We used the above three control variables when the dependent variable was total return. We used relative size, relative return on assets, and relative solvency, respectively, when the dependent variable was relative return. Relative size was calculated as the ratio of the smaller size and the larger size of the two firms. Relative performance was calculated as the ratio of the exponential of the smaller value of return on assets (or solvency) between two firms and the exponential of the bigger value of return on assets (or solvency) (Gulati & Gargiulo, 1999). Resource dependence theory suggests that JVs may allow firms to effectively manage their resource dependence on other firms (Pfeffer & Nowak, 1976). We used data from input-output tables to capture the possible effect of such a resource dependence on the value creation of the JVs. Input-output tables are constructed to reflect the resource dependence of any two industries in the American economy (U.S. Department of Commerce, 1994). We used the 1987 Benchmark commodityby-industry direct requirement coefficients as the indicator of the interdependence of the two firms. We calculated the total dependence of the two firms on each other as the sum of the coefficients for Firm 1 and Firm 2 and the relative dependence of the two firms as the ratio of the smaller coefficient and the larger coefficient.
228
RANJAY GULATI AND LIHUA OLIVIA WANG
Finally, to capture any unobservable time effect, we created dummy variables for each year from 1988–1996 and control for the time effect in all regression analyses. We also used the first digit of the SIC codes of the two firms in a JV as additional control variables because the value creation and appropriation across different economic sectors may vary (Madhavan & Prescott, 1995). Some firms involved in the JV activities more than others. Firms involved in more than one JV create interdependence among these cases. Although the eventstudy methodology controls for the firm-specific characteristics in the calculation of firm abnormal returns, and thus partially takes care of the problem, we included the dummy variables for the top fifteen firms with most JV activities as control variables.2
RESULTS Table 1 presents the distribution of positive and negative cumulative stock market reactions for all participating firms (658 pairs and 1,316 firm cases). In 45.97% of the cases, the abnormal returns are positive, in 46.28% of the cases, they are negative (the data for the remaining 7.75% of the cases are missing). Tables 2 and 3 provide the descriptive statistics and the correlation matrix among variables. The correlation matrix indicates that total value creation is highly correlated with all major independent variables (the degree of relational embeddedness, the degree of structural embeddedness and asymmetry of business relatedness of two firms with the JV). Relative value appropriation, however, is only highly correlated with the degree of relational embeddedness. In addition, the degree of relational embeddedness and the degree of structural embeddedness are highly correlated (the correlation is 0.582). We do not include the two variables simultaneously in any model, so this correlation does not introduce any problem. Tables 4 and 5 present the results of the regression models using total value creation and relative value appropriation as the dependent variable, respectively. Model 1 to 8 use total value creation as the dependent variable. Model 1 is the Table 1. Frequency of Positive and Negative Cumulative Stock Market Reaction for All Partners (658 Pairs and 1,316 Firm Cases). Reaction Positive Negative Missing
Frequency 605 609 102
Percentage (%) 45.97 46.28 7.75
Size of the Pie and Share of the Pie
229
Table 2. Descriptive Statistics of the Major Variables. Variable Total value creation Relative value appropriation The degree of relational embeddedness (Number of direct ties) The degree of structural embeddedness (Number of indirect ties) Asymmetry of business relatedness of two firms with the JV Business relatedness of two firms Log(total size) Total return on assets Total solvency Total dependence of two firms Relative size Relative return on assets Relative solvency Relative dependence of two firms
N
Mean
St. dev.
Min
Max
570 570 658
0.001 0.012 0.716
0.043 0.013 1.696
−0.240 0 0
0.282 0.126 15
658
1.125
2.295
0
13
658
0.351
0.478
0
1
658 630 630 504 658 630 630 504 658
0.347 10.450 0.084 1.147 0.108 0.410 0.942 0.388 0.374
0.476 0.805 0.098 1.019 0.164 0.275 0.058 0.302 0.430
0 8.395 −0.432 0.006 0 0.010 0.552 0 0
1 11.982 0.411 5.836 0.748 1 1 1 1
baseline model, including only control variables. Model 2 introduces the degree of relational embeddedness between two firms into the regression equation to test Hypothesis 1a. The result indicates that there is a positive relationship between the degree of relational embeddedness between two firms and total value creation. Hypothesis 1a is supported. The total value creation from announcing a new JV is 0.342% more with an additional prior direct tie between two firms. For example, for two firms with total current value of $4 billion dollars, the benefits from having an additional direct tie is $13.68 million. Model 3 tests Hypothesis 1b, which predicts that the degree of relational embeddedness (indicated by the number of prior direct ties) between two firms has an inverted-U-shaped relationship with total value creation. The coefficient of the square term of the number of direct ties is not significant. Hypothesis 1b is not supported. Model 4 and 5 examine the relationship between the degree of structural embeddedness (indicated by the number of indirect ties between two firms) and total value creation. Since the effect of structural embeddedness could be confounded with the effect of relational embeddedness, we evaluate the effect of structural embeddedness only when there is no relational embeddedness between two firms. Model 4 examines the linear relationship between the degree of structural embeddedness and total value creation (Hypothesis 2a). The result shows that there is no linear relationship between them. Model 5 tests Hypothesis 2b, which predicts the curvilinear relationship between the degree of structural embeddedness between
230
Table 3. Correlation Matrix of the Major Variables (N = 436). (1)
∗p
< 0.05.
(3)
1.00 0.190*
1.00
0.116*
0.104*
1.00
0.107*
0.065
0.582*
−0.144*
0.012
0.057
−0.018
0.070 −0.088* −0.034 0.061 −0.038 −0.011 −0.047 0.035
0.026 −0.086* −0.020 0.052 −0.120* −0.093* −0.093 −0.009
−0.053
0.120* 0.274* −0.101* −0.221* 0.189* −0.161* −0.159* −0.130* 0.108*
(4)
(5)
(6)
(7)
(8)
1.00 −0.139* −0.104* 0.026 −0.423* 0.096 −0.002 −0.028
1.00 −0.188* −0.170* 0.077 0.302* −0.051 −0.138*
(9)
(10)
(11)
(12)
(13)
(14)
1.00 −0.112*
1.00
0.111* −0.536* 0.330* −0.158* −0.332* 0.232* −0.142* −0.160* −0.125* 0.135*
−0.073 0.101 −0.006 −0.351* −0.015 −0.004 −0.083 −0.362*
1.00 −0.066 −0.097* −0.030 0.727* −0.005 0.005 0.038 0.647*
1.00 0.009 1.00 0.126* −0.069 1.00 * 0.131 −0.054 0.033 1.00 0.032 −0.043 0.110* 0.251* 1.00 0.120* 0.731 −0.004 −0.009 0.037 1.00
RANJAY GULATI AND LIHUA OLIVIA WANG
(1) Total value creation (2) Relative value appropriation (3) The degree of relational embeddedness (4) The degree of structural embeddedness (5) Asymmetry of business relatedness of two firms with the JV (6) Business relatedness of two firms (7) Log(total size) (8) Total ROA (9) Total solvency (10) Total interdependence (11) Relative size (12) Relative ROA (13) Relative solvency (14) Relative interdependence
(2)
Model 1
The degree of relational embeddedness (No. of direct ties) The degree of relational embeddedness (No. of direct ties) squared The degree of structural embeddedness (No. of indirect ties) The degree of structural embeddedness (No. of indirect ties) squared Asymmetry of Business relatedness of two firms with the JV
Model 2 0.342
**
(0.151)
Model 3 0.515
*
−0.019
Model 4
Model 5
Model 6
(0.282)
Model 7 0.376
**
Model 8
(0.151)
(0.026) 0.183
(0.228)
0.220
(0.480)
−0.005
(0.063)
0.169
(0.227)
−1.128** (0.556)
−1.267** (0.555)
−1.411** (0.618)
Business relatedness of two firms Log(total size) Total return on assets Total solvency Total dependence
0.573 −0.075 −2.732 −0.393 1.933
(0.666) (0.470) (2.700) (0.268) (1.944)
0.415 −0.059 −1.890 −0.325 1.935
(0.666) (0.467) (2.712) (0.268) (1.935)
0.388 −0.058 −1.743 −0.316 2.064
(0.668) (0.467) (2.721) (0.268) (1.944)
0.409 0.054 1.199 −0.484* 3.232
(0.728) (0.502) (3.252) (0.288) (2.106)
0.411 0.050 1.164 −0.482* 3.227
(0.730) (0.505) (3.283) (0.289) (2.110)
−0.102 −0.070 −2.674 −0.387 2.300
(0.742) (0.468) (2.689) (0.267) (1.945)
−0.357 −0.052 −1.739 −0.311 2.348
(0.744) (0.465) (2.698) (0.267) (1.933)
−0.431 0.065 1.272 −0.471 3.553*
(0.811) (0.498) (3.227) (0.286) (2.094)
Period 88 Period 89 Period 90 Period 91 Period 92 Period 93 Period 94 Period 95 Period 96
1.294 1.627 2.693** −0.615 1.280 1.368 1.350 1.313 1.986
(1.282) (1.336) (1.201) (1.114) (1.096) (1.154) (1.092) (1.176) (1.357)
1.169 1.525 2.571** −0.684 1.190 1.166 1.024 0.934 1.385
(1.277) (1.330) (1.196) (1.109) (1.091) (1.152) (1.096) (1.182) (1.376)
1.133 1.505 2.548** −0.718 1.144 1.083 0.974 0.878 1.376
(1.279) (1.331) (1.197) (1.110) (1.094) (1.158) (1.099) (1.185) (1.377)
2.190 1.739 2.895** −1.002 0.962 1.199 1.391 0.900 2.582*
(1.365) (1.413) (1.230) (1.143) (1.176) (1.308) (1.153) (1.293) (1.527)
2.194 1.737 2.895** −1.007 0.956 1.192 1.384 0.890 2.574*
(1.368) (1.415) (1.232) (1.147) (1.180) (1.313) (1.158) (1.301) (1.533)
1.274 1.486 2.607** −0.722 1.102 1.191 1.082 1.179 1.723
(1.277) (1.333) (1.197) (1.111) (1.095) (1.153) (1.096) (1.173) (1.358)
1.135 1.357 2.463** −0.811 0.981 0.946 0.691 0.745 1.029
(1.271) (1.326) (1.191) (1.104) (1.089) (1.150) (1.100) (1.178) (1.377)
2.135 1.708 2.763** −1.154 0.754 0.902 0.957 0.848 2.215
(1.355) (1.402) (1.221) (1.136) (1.170) (1.304) (1.160) (1.284) (1.523)
Industry sector of Firm 1 Industry sector of Firm 2
−0.034 0.081
(0.283) (0.288)
−0.030 0.097
(0.282) (0.287)
−0.027 0.092
(0.282) (0.287)
−0.224 0.275
(0.299) (0.294)
−0.226 0.276
(0.300) (0.295)
0.019 0.052
(0.283) (0.287)
0.029 0.066
(0.281) (0.285)
−0.155 0.225
(0.298) (0.293)
Constant Number of cases
−0.169 (4.901) 436
−0.296 (4.876) 436
0.101 0.032
0.112 0.042
R2 Adjusted R2
1.322 (4.879) 436 0.113 0.041
−1.454 (5.274) 300
−1.410 (5.309) 300
0.194 0.098
0.195 0.095
0.454 (4.891) 436 0.110 0.039
0.392 (4.860) 436 0.124 0.052
Size of the Pie and Share of the Pie
Table 4. Regression Analysis Using Total Value Creation (%) As the Dependent Variable. Variables
−0.670 (5.245) 300 0.210 0.112
231
Note: Fifteen dummy variables representing 15 firms that participated in the most JVs are also entered as control variables for all models in Tables 4 and 5. For simplicity, the results are not shown in the tables. These firms are: Apple Computer, AT&T, Bell Atlantic, B. F. Goodrich, Digital Equipment, Dow Chemical, Du Pont, Eastman Kodak, General Electric, General Motors, HP, IBM, Intel, Motorola, Sun Microsystems. ∗ p < 0.10; ∗∗ p < 0.05.
232
Table 5. Regression Analyses Using Relative Value Appropriation (%) As the Dependent Variable. Variables
Model 9
The degree of relational embeddedness (No. of direct ties) The degree of structural embeddedness (No. of indirect ties) Asymmetry of business relatedness of two firms with the JV
Model 10 0.062
Model 11
Model 12
(0.043)
Model 13 0.063
Model 14
(0.044)
−0.0004 (0.073)
−0.0004 (0.073) 0.015
(0.160)
−0.007
(0.161)
0.006 (0.193)
−0.043 −0.366 −3.012** −0.033 −0.072
(0.174) (0.256) (1.261) (0.227) (0.192)
0.006 −0.373 −2.831** −0.035 −0.064
(0.176) (0.255) (1.265) (0.226) (0.192)
0.051 −0.313 −2.248 −0.115 −0.105
(0.206) (0.320) (1.523) (0.282) (0.229)
0.051 −0.366 −3.011** −0.032 −0.073
(0.194) (0.256) (1.262) (0.228) (0.192)
0.002 −0.373 −2.831** −0.036 −0.064
(0.196) (0.256) (1.267) (0.227) (0.192)
0.055 −0.312 −2.248 −0.114 −0.105
(0.233) (0.321) (1.526) (0.284) (0.230)
Period 88 Period 89 Period 90 Period 91 Period 92 Period 93 Period 94 Period 95 Period 96
−0.260 −0.711* 0.549 −0.021 −0.078 −0.317 −0.406 −0.214 −0.598
(0.362) (0.373) (0.339) (0.312) (0.314) (0.333) (0.312) (0.331) (0.372)
−0.268 −0.711* 0.539 −0.022 −0.088 −0.348 −0.459 −0.272 −0.692*
(0.362) (0.372) (0.338) (0.311) (0.314) (0.334) (0.314) (0.333) (0.377)
−0.260 −0.551* 0.701 −0.027 −0.138 −0.310 −0.388 −0.482 −0.464
(0.419) (0.431) (0.376) (0.349) (0.364) (0.407) (0.357) (0.395) (0.455)
−0.259 −0.709* 0.550 −0.020 −0.076 −0.315 −0.402 −0.212 −0.595
(0.363) (0.374) (0.339) (0.312) (0.316) (0.335) (0.315) (0.332) (0.374)
−0.268 −0.712* 0.538 −0.022 −0.089 −0.349 −0.461 −0.272 −0.674*
(0.362) (0.373) (0.339) (0.312) (0.315) (0.335) (0.317) (0.334) (0.380)
−0.260 −0.550 0.702* −0.027 −0.137 −0.309 −0.386 −0.482 −0.462
(0.420) (0.432) (0.378) (0.350) (0.365) (0.410) (0.362) (0.396) (0.459)
Industry sector of Firm 1 Industry sector of Firm 2
0.075 0.037
(0.078) (0.079)
0.076 0.043
(0.077) (0.079)
0.064 0.074
(0.088) (0.088)
0.075 0.037
(0.078) (0.079)
0.076 0.043
(0.078) (0.079)
0.064 0.075
(0.089) (0.088)
Constants Number of cases
4.014*** (1.260) 436
R2 Adjusted R2 ∗p
< 0.10; ∗∗ p < 0.05; ∗∗∗ p < 0.01.
0.141 0.078
3.867*** (1.263) 436 0.146 0.080
3.236** (1.516) 300 0.167 0.070
4.003*** (1.267) 436 0.141 0.075
3.872 (1.268) 436 0.146 0.078
3.231** (1.524) 300 0.167 0.067
RANJAY GULATI AND LIHUA OLIVIA WANG
Business Relatedness of two firms Relative size Relative return on assets Relative solvency Relative dependence
Size of the Pie and Share of the Pie
233
two firms and the total value creation. The coefficient for this variable is also not significant. Hypothesis 2b is not supported. Model 6 to model 8 test Hypothesis 5, which predicts a negative relationship between asymmetry of business relatedness of two firms and total value creation. Model 6 introduces only asymmetry of business relatedness of two firms with the JV in addition to the control variables. The result suggests that the asymmetry of business relatedness of two firms with the JV significantly discounts the investors’ expectation of the total value creation of the new JV, which supports Hypothesis 5. If one firm is related to the JV operation while the other firm is not, the total value creation will be 1.128% lower than if neither or both firms are related to the JV operation. The result holds even when we add embeddedness variables in the equations (model 7 and model 8). To further test the hypothesis, we divided all cases into three groups: (1) Both partners’ primary businesses are related to the JV activities; (2) Both partners’ primary businesses are not related to the JV activities and (3) one partner’s primary business is related to the JV activities while the other partner’s primary business is not related to the JV activities (in this case, the asymmetry exists). We found that the total value creation between the first two groups (symmetric case) are not significantly different, but they are significantly different from the third group, where the asymmetry of business relatedness between two firms and the JV exists (the results are not reported here). Models 9 to 14 test the hypotheses using relative value appropriation as the dependent variable. Model 9 is the baseline model, with only control variables in the equation. The result indicates that relative return on assets of the two firms has a negative relationship with relative value appropriation, indicating that JVs formed by partners with similar returns on assets are more likely to have similar abnormal returns. Model 10 introduces the degree of relational embeddedness (the number of direct ties) between two firms into the regression equation to test Hypothesis 3, which predicts that as the degree of relational embeddedness of two firms increases, the value to each partner becomes more symmetric. The result is not significant and Hypothesis 3 is not supported. Model 11 tests the effect of structural embeddedness in the absence of relational embeddedness (Hypothesis 4). This hypothesis predicts that the more structurally embedded the two firms are, the more likely they will have a symmetric value appropriation. The non-significant result does not support this hypothesis. Model 12 to model 14 tests Hypothesis 6, which predicts a positive relationship between asymmetry of business relatedness of two firms with the JV and relative value appropriation, i.e. asymmetry of business relatedness of two firms with the JV promotes the unequal share of the pie between two firms. Model 12 adds asymmetry of business relatedness of two firms with the JV into the equation in addition to control variables. The result shows that asymmetry of business relatedness of two
234
RANJAY GULATI AND LIHUA OLIVIA WANG
firms with the JV has no relationship with relative value appropriation. The results do not change as embeddedness variables are added into the equation. Therefore, Hypothesis 6 is not supported.
DISCUSSION This chapter used an event-study methodology to examine the conditions under which a new JV is expected to create value for the shareholders of all partners of a JV and how the value created is appropriated by each partner. We proposed that the study of the dynamics underlying dyadic-level value creation and appropriation in JVs is important because any JV involves at least two firms (Zajac & Olsen, 1993). Much of prior firm-level research of these issues has provided at best an incomplete picture. In addition, this study also extends prior research by considering the importance not only of strategic and economic factors but also of social factors that may influence the dynamics of value creation and appropriation in JVs. Our results show that the magnitude of total value created in a JV is influenced by the extent to which the two partners are embedded in a network of prior JV ties. Our results suggest that the degree of relational embeddedness between two parent firms has a positive effect on total value creation in the JV. We also find that the asymmetry of business relatedness of two firms with the JV negatively affects the total value creation of the JV. We find consistently across all the models that relative value appropriation is not influenced by either network embeddedness or the asymmetry of business relatedness of two firms with the JV. An important finding of this study is that the degree of relational embeddedness between two firms has a positive relationship with the total value creation. The results indicate that there is no limit to the benefits a firm can extract from a relationship with another firm with whom it has a rich history of direct JV relationships. The investors expect that the more relationally embedded two firms are, the more likely that the new JV is going to create more value. This finding is at odds with the arguments and the empirical evidence that the benefits of relational embeddedness may reach a limit beyond which an increase of relational embeddedness may not provide additional benefits for JV partners or even becomes a liability to prevent firms from pursuing new opportunities, as our Hypothesis 1b suggested. It is possible that our sample is not large enough and the time horizon is not long enough so that the range for the relational embeddedness has not reached to a point beyond which the positive value of relational embeddedness starts to diminish. Future research with a larger sample size and a longer time horizon could further test this hypothesis. We failed to find the support for Hypotheses 2a and 2b, two competing hypotheses on the relationship between the degree of structural embeddedness between
Size of the Pie and Share of the Pie
235
two firms and total value creation. Unlike many prior studies showing the significant information and control roles provided by structural embeddedness in the formation of an alliance between firms or in the innovative capabilities of firms (Ahuja, 2000), it seems that investors do not look beyond relational embeddedness to search for cues about whether a JV will create more value for the parent firms. Underlying both relational embeddedness and structural embeddedness are the mechanisms of information (learning) effects and reputation (control) effects on the total value creation. The strengths of these effects, however, are likely to be different between relational embeddedness versus structural embeddedness. In terms of information (learning) effects, although both relational embeddedness and structural embeddedness offer information about partners’ characteristics, the partners may value the information resulting from the relational embeddedness more than the information resulting from the structural embeddedness. This is because the former is directly from the past interaction while the latter is from third party interactions. Therefore, the former is likely to be considered to be more reliable, richer and more fine-grained than the latter. In terms of reputation and control effect, the effect from structural embeddedness may be more profound than the effect from relational embeddedness. The reputation concern from the structural embeddedness between partners affects the wider sets of partners and thus affects the reputation of partners in a wider network. Conversely, the effect of relational embeddedness is only between partners. Thus if reputation is lost, it only hurts the relationship between the current two partners. Our findings indicate that investors do not look beyond the effect of relational embeddedness on total value creation. It is quite possible that indirect ties are not salient enough for investors to attend to signals resulting from them. It is also possible that investors consider that information carried through prior indirect ties may not be reliable and is hence discounted by them. In a point of departure from prior research, we empirically assessed the effect of network embeddedness on relative value appropriation of two partners in a JV. Our rationale for this set of hypotheses was that relationally or structurally embedded firms are more likely to view the relationship as long-term and are more likely to consider the mutual benefits and costs instead of merely individual costs and benefits. Thus, the desire for asymmetric returns is reduced and investors are more likely to expect symmetric returns between the partners. Contrary to our expectations, we did not found any evidence that embeddedness plays any role in influencing the extent of relative value appropriation between two partners. One possible explanation is that our data does not allow us to distinguish whether the differences in value extraction by firms in a JV result from the anticipated relative contributions by the partners. Future research should
236
RANJAY GULATI AND LIHUA OLIVIA WANG
further investigate this issue with richer data on projected contributions by each partner. We found very strong evidence that the asymmetry of business relatedness of the two firms is detrimental to total value creation in the JV. Prior research at the firm level has found that those firms that are more closely related to their new JVs obtain higher returns. This is because such firms have more opportunities to use their current capabilities to the JV operations, and also more likely to use the skills learned from the JV operations to their other related businesses outside the JVs. In addition, such firms typically have higher capacity to absorb the knowledge learned from the JV operations (Cohen & Levinthal, 1990). At the dyadic level, Harrigan (1988) found that the asymmetry of business relatedness of two firms with the JV decreases the survival rate of the JV. In her definition, if both firms are related to the JV, the symmetry exists. Asymmetry exists otherwise. This definition does not distinguish those JVs where one partner is related to the JV but the other partner is not (the asymmetry case in our chapter) from those JVs where both partners are not closely related to the JV (one of the symmetric cases in our chapter). Our findings indicate that the extent of total value created does not differ between those JVs in which both partners are related to the new JV and those in which neither partner is related to the new JV. It is the asymmetry of business relatedness (one partner is related to the JV but the other partner is not) that reduces the total value creation. It is possible that the asymmetry of knowledge transfer benefits between the two partners may lead to shifts in the bargaining power in a JV, which may discourage knowledge-sharing between partners and thus reduce the joint value creation of the two partners from the JV (Chi, 2000; Inkpen, 1998). While the asymmetry of business relatedness of two firms plays an important role in total value creation of a JV, it does not affect how much of the total value is appropriated by each partner. This is contrary to our expectations that differences in potential learning opportunities and the absorptive capacity of the partners may lead investors to expect that the returns to each partner may vary. It is possible that investors expect that the independent legal status of the JV may provide a buffer by which neither partner can gain a differential advantage over the other. Our research has important implications for broader research on network embeddedness and strategic alliances. Network researchers have long elaborated on how cohesion and bridging mechanisms play different roles for a variety of individual and organizational outcomes, such as job search, managers’ promotions, and alliance formation (Burt, 1992; Granovetter, 1974; Gulati, 1995b). We extend these ideas to the arena of value creation in JVs and show that in the context of market evaluation of a new JV, a cohesive relationship between partners resulting from a large number of prior direct JV ties ensures investors with confidence that the new JV will create more value for partners. The investors seem to value the
Size of the Pie and Share of the Pie
237
new opportunities brought by two firms that have not had connections before the new JV less than those firms that have previous ties. Research on exploration and exploitation indicates that firms try to balance the exploitation of current opportunities and the exploration of new ones (March, 1991; Koza & Lewin, 1998). This study suggests that firms may value the exploitation of safety benefits resulting from the embeddedness between partners more than the exploration of new opportunities in joint venture formation. Nevertheless, we acknowledge that firms may value the role of exploration of new opportunities over exploitation of existing ones in some other contexts (Rowley et al., 2000). Our study provides important insights about the social and economic conditions under which a new JV creates value for partners and how the value created is divided between partners. But it also has some limitations and points to several important areas for future research. The first limitation has to do with the nature of event-study methodology. The typical assumption of an event study is that the investors have rich information about the firms and this information has been incorporated efficiently in the stock price of the firm (Fama, 1970). Although this market efficiency assumption has been criticized for not considering the actual processes of decision making of shareholders and their roles in determining stock prices (Baker, 1984), the evidence shows that the investors do react to the significant firm actions such as investment decisions, financing decisions and changes in corporate controls (Fama, 2000). In our study, like other event studies, we assume that investors have the information about the relational and structural embeddedness between two partners of a newly announced joint venture and they react to the new JV announcement based on this information. We think that this assumption is plausible because our sample is the biggest firms in the United States. The press and analysts follow the actions of these firms quite closely. Many joint ventures are announced and analyzed in public and the information is often well circulated. Nevertheless, we do not claim that all investors are fully informed around the new JV announcement. Our result that relational embeddedness (but not structural embeddedness) between two firms is important in affecting total value creation in JVs indicates that investors do pay attention to the relational property of two firms but not structural embeddedness. We acknowledge the limits of event studies but would also like to highlight that in the absence of rich detailed data on a large sample of alliances, event studies are one of the best available methods to tease out the likely antecedents of value creation and appropriation in JVs. The actual value created in JVs remains an elusive construct in alliance research due to the difficulty of measuring it. As a result, event studies have become perhaps the only measure of value creation in JVs. We continued in this tradition and extended the rich body of research by examining both the total value creation and the relative value appropriation, while
238
RANJAY GULATI AND LIHUA OLIVIA WANG
also assessing the social bases for these outcomes. We hope that future researchers will collect rich survey data to validate and extend our findings on the value created and appropriated in JVs. An additional arena for future research would be to consider not only the antecedent factors but also the dynamic interaction processes between partners over time, which may have a profound influence on the ongoing value creation and appropriation between firms in a JV (Doz, 1996; Doz & Hamel, 1988; Ring & van de Ven, 1992, 1994). As a result, over the life course of a JV, the dynamics of interaction between partners in a JV may lead to varying outcomes at different points in time, which is consistent with broader strategy research on strategy implementation (Mintzberg, 1978; Mintzberg & Waters, 1985). Stock market reactions at the announcement of a JV may thus reflect the value creation and division at the outset, but these may change later during the operation of the JV. The challenge for future research is to integrate market-based assessments of organizational effectiveness at the outset with ongoing developments over time. Important extensions of this study will benefit from the more direct measures of value creation in the context of a longitudinal design.
NOTES 1. All the results for 21-day window are not significant so they are not reported. A 21-day event window might be too long, as the effect of the announcement may have run its course earlier. 2. These control variables are included but not reported in the results.
ACKNOWLEDGMENTS We would like to thank Bharat N. Anand, Michael Faulkender, Martin Gargiulo, Michael Jensen, Tarun Khanna, Peter Murmann, William Ocasio, Brian Silverman, Jim Westphal, Ed Zajac, participants at seminars at Northwestern University and especially the editors for the special issue Vincent Buskens, Werner Raub and Chris Snijders for their helpful comments on previous drafts of this chapter.
REFERENCES Afuah, A. (2000). How much do your co-opetitors’ capabilities matter in the face of technological change? Strategic Management Journal, 21, 387–404.
Size of the Pie and Share of the Pie
239
Ahuja, G. (2000). Collaboration networks, structural holes, and innovation: A longitudinal study. Administrative Science Quarterly, 45, 425–455. Anand, B. N., & Khanna, T. (1995). On the market valuation of inter-firm agreements: The computer and telecommunications industries, 1990–1993. Working paper of NBER Summer Industrial Organization Workshop, Cambridge. Anand, B. N., & Khanna, T. (2000). Do firms learn to create value? The case of alliances. Strategic Management Journal, 21, 295–315. Augereau, A. (1999). Three empirical essays on networks. Unpublished doctoral dissertation, Northwestern University, Evanston, IL. Axelrod, R. (1984). The evolution of cooperation. New York: Basic Books. Baker, W. E. (1984). The social structure of a national securities market. American Journal of Sociology, 89, 775–811. Balakrishnan, S., & Koza, M. (1993). Information asymmetry, adverse selection and joint ventures. Journal of Economic Behavior and Organization, 20, 99–117. Batenburg, R. S., Raub, W., & Snijders, C. (2003). Contacts and contracts: Dyadic embeddedness and the contractual behavior of firms. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 135–188). Oxford: Elsevier Science. Baum, J., & Oliver, C. (1991). Institutional linkages and organizational mortality. Administrative Science Quarterly, 36, 187–218. Binder, J. J. (1998). The event study methodology since 1969. Review of Quantitative Finance and Accounting, 11, 111–137. Black, B. S., & Grundfest, J. A. (1988). Shareholder gains from takeovers and restructurings between 1981 and 1986. Journal of Applied Corporate Finance, 11, 71–78. Borys, B., & Jemison, D. B. (1989). Hybrid arrangements as strategic alliances: Theoretical issues and organizational combinations. Academy of Management Review, 14, 234–249. Bradley, M., Desai, A., & Kim, E. H. (1983). Synergistic gains from corporate acquisitions and their division between the stockholders of target and acquiring firms. Journal of Financial Economics, 21, 3–40. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Burt, R., & Knez, M. (1995). Kinds of third-party effects on trust. Rationality and Society, 7, 255–292. Buskens, V. (1998). The social structure of trust. Social Networks, 20, 265–289. Buskens, V., Batenburg, R. S., & Weesie, J. (2003). Embedded partner selection in relations between firms. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 107–133). Oxford: Elsevier Science. Buskens, V., & Raub, W. (2002). Embedded trust: Control and learning. Advances in Group Processes, 19, 167–202. Camerer, C., & Weigelt, K. (1988). Experimental tests of a sequential equilibrium reputation model. Econometrica, 56, 1–36. Chan, S. H., Kensinger, H. W., Keown, A. J., & Martin, J. D. (1997). Do strategic alliances create value? Journal of Financial Economics, 16, 199–221. Chi, T. (2000). Options to acquire or divest a joint venture. Strategic Management Journal, 21, 665–687. Cohen, W. M., & Levinthal, D. A. (1990). Absorptive capability: A new perspective on learning and innovation. Administrative Science Quarterly, 35, 128–152. Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94(supplement), S95–S120.
240
RANJAY GULATI AND LIHUA OLIVIA WANG
Datta, D. K., & Puia, G. (1995). Cross-boarder acquisitions: An examination of the influence of relatedness and cultural fit on shareholder value creation in US acquiring firms. Management International Review, 35, 337–359. Doz, Y. L. (1996). The evolution of cooperation in strategic alliances: Initial conditions or learning processes? Strategic Management Journal, 17, 55–83. Doz, Y. L., & Hamel, G. (1988). Alliance advantage: The art of creating value through partnering. Boston, MA: Harvard Business School Press. Doz, Y. L., Hamel, G., & Prahalad, C. K. (1989). Collaborate with your competitors and win. Harvard Business Review, 67, 133–139. Dyer, J. H. (1997). Effective interfirm collaboration: How firms minimize transaction costs and maximize transaction value. Strategic Management Journal, 18, 535–556. Dyer, J. H., & Singh, H. (1998). The relational view: Cooperative strategy and sources of interorganizational competitive advantage. Academy of Management Review, 23, 660–679. Fama, E. F. (1970). Efficient capital markets: A review of theory and empirical work. Journal of Finance, 25, 383–417. Fama, E. F. (2000). Efficient capital markets: II. Journal of Finance, 46, 1575–1617. Fama, E. F., Fisher, L., Jensen, M. C., & Roll, R. (1969). The adjustment of stock prices to new information. International Economic Review, 10, 1–21. Geringer, J. M., & Hebert, L. (1991). Measuring performance of international joint ventures. Journal of International Business Studies, 22, 249–263. Granovetter, M. (1974). Getting a job. Cambridge, MA: Harvard University Press. Granovetter, M. (1985). Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91, 481–510. Gulati, R. (1995a). Does familiarity breed trust? The implications of repeated ties for contractual choice in alliances. Academy of Management Journal, 38, 85–112. Gulati, R. (1995b). Social structure and alliance formation patterns: A longitudinal analysis. Administrative Science Quarterly, 40, 619–652. Gulati, R. (1999). Network location and learning: The influence of network resources and firm capabilities on alliance formation. Strategic Management Journal, 20, 397–420. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104, 1439–1493. Gulati, R., Khanna, T., & Nohria, N. (1994). Unilateral commitment and the importance of process in alliances. Sloan Management Review, 35, 61–69. Gulati, R., & Singh, H. (1998). The architecture of cooperation: Managing coordination costs and appropriation concerns in strategic alliances. Administrative Science Quarterly, 43, 781–814. Gulati, R., & Westphal, J. (1999). Cooperative or controlling? The effects of CEO-board relations and the content of interlocks on the formation of JVs. Administrative Science Quarterly, 44, 473–506. Hamel, G. (1991). Competition for competence and inter-partner learning within international strategic alliances. Strategic Management Journal, 12, 83–103. Harrigan, K. R. (1988). Strategic alliances and partner asymmetries in international business. In: F. J. Contractor & P. Lorange (Eds), Cooperative Strategies in International Business (pp. 205–226). Lexington, MA: Lexington Books. Inkpen, A. C. (1998). Learning and knowledge acquisition through international strategic alliances. Academy of Management Executive, 12, 69–80. Khanna, T. (1998). The scope of alliances. Organization Science, 9, 340–355.
Size of the Pie and Share of the Pie
241
Khanna, T., Gulati, R., & Nohria, N. (1998). The dynamics of learning alliances: Competition, cooperation, and relative scope. Strategic Management Journal, 19, 193–210. Khanna, T., Gulati, R., & Nohria, N. (2000). The economic modeling of strategic process: “clean models” and “dirty hands”. Strategic Management Journal, 21, 781–790. Kogut, B. (1988). Joint ventures: Theoretical and empirical perspectives. Strategic Management Journal, 9, 319–332. Koh, J., & Venkatraman, N. (1991). Joint venture formations and stock market reactions: An assessment in the information technology sector. Academy of Management Journal, 34, 869– 892. Koza, M., & Lewin, A. (1998). The co-evolution of strategic alliances. Organization Science, 9, 255– 264. Kreps, D. M., Milgrom, P., Roberts, J., & Wilson, R. (1982). Rational cooperation in the finitely repeated prisoners’ dilemma. Journal of Economic Theory, 27, 245–252. Kreps, D. M. (1990). Corporate culture and economic theory. In: J. Alt & K. Shepsle (Eds), Perspectives on Positive Political Economy (pp. 90–143). Cambridge: Cambridge University Press. Lewis, J. D. (1990). Partnerships for Profit. New York: Free Press. Lorenz, E. H. (1988). Neither friends nor strangers: Informal networks of subcontracting in French industry. In: D. Gambetta (Ed.), Trust: Making and Breaking Cooperative Relationships (pp. 194–210). Oxford: Blackwell. Madhavan, R., & Prescott, J. E. (1995). Market value impact of joint ventures: The effect of industry information-processing load. Academy of Management Journal, 38, 900–915. March, J. G. (1991). Exploration and exploitation in organizational learning. Organization Science, 2, 71–87. McConnell, J., & Nantell, T. (1985). Corporate combinations and common stock returns: The case of JVs. Journal of Finance, 40, 519–536. Mintzberg, H. (1978). Patterns in strategy formulation. Management Science, 24, 934–948. Mintzberg, H., & Waters, J. A. (1985). Of strategies: Deliberate and emergent. Strategic Management Journal, 6(July-September), 257–272. Montgomery, C. (1982). The measurement of firm diversification: Some new empirical evidence. Academy of Management Journal, 25, 299–307. Mowery, D. C., Oxley, J. E., & Silverman, B. S. (1996). Strategic alliances and interfirm knowledge transfer. Strategic Management Journal, 17, 77–91. Palepu, K. (1985). Diversification strategy, profit performance and the entropy measure. Strategic Management Journal, 6, 239–255. Pfeffer, J., & Nowak, P. (1976). Joint ventures and interorganizational interdependence. Administrative Science Quarterly, 21, 398–418. Powell, W. W. (1987). Hybrid organizational arrangements: New form or transitional development? California Management Review, 30, 67–87. Raub, W., & Weesie, J. (1990). Reputation and efficiency in social interactions: An example of network effects. American Journal of Sociology, 96, 626–654. Ring, P. S., & van de Ven, A. H. (1992). Structuring cooperative relationships between organizations. Strategic Management Journal, 13, 483–498. Ring, P. S., & van de Ven, A. H. (1994). Developmental processes of cooperative interorganizational relationships. Academy of Management Review, 19, 90–118. Rowley, T., Behrens, D., & Krackhardt, D. (2000). Redundant governance structures: An analysis of structural and relational embeddedness in the steel and semiconductor industries. Strategic Management Journal, 21, 369–386.
242
RANJAY GULATI AND LIHUA OLIVIA WANG
Rumelt, R. P. (1974). Strategy, structure, and economic performance. Cambridge, MA: Harvard University Press. Seth, A. (1990). Value creation in acquisitions: A reexamination of performance issues. Strategic Management Journal, 11, 99–115. Sicherman, N. W., & Pettway, R. C. (1987). Acquisition of divested assets and shareholders wealth. Journal of Finance, 42, 1261–1273. Stuart, T. E. (2000). Interorganizational alliances and the performance of firms: A study of growth and innovation rates in a high-technology industry. Strategic Management Journal, 21, 791–811. Stuart, T. E. (2003). Governing strategic alliances. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 189–208). Oxford: Elsevier Science. Taylor, M. (1976). Anarchy and cooperation. London: Wiley. Thompson, L. (1998). The mind and heart of the negotiator. Upper Saddle River, NJ: Prentice Hall. United States Department of Commerce. (1994). Survey of current business, the input-output structure of the US economy, 1977. Voss, T. (2003). The rational choice approach to an analysis of intra- and interorganizational governance. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 21–46). Oxford: Elsevier Science. Walker, G., Kogut, B., & Shan, W. (1997). Social capital, structural holes and the formation of an industry network. Organization Science, 8, 109–125. Zaheer, A., McEvily, B., & Perrone, V. (1998). Does trust matter? Exploring the effects of interorganizational and interpersonal trust on performance. Organization Science, 9, 141–159. Zaheer, A., & Venkatraman, N. (1995). Relational governance as an interorganizational strategy: An empirical test of the role of trust in economic exchange. Strategic Management Journal, 16, 373–392. Zajac, E., & Olsen, C. (1993). From transaction costs to transactional value analysis: Implications for the study of interorganizational strategies. Journal of Management Studies, 30, 131–145.
NETWORK ENDORSEMENT AND SOCIAL STRATIFICATION IN THE LEGAL PROFESSION Harris H. Kim and Edward O. Laumann ABSTRACT In this chapter, we show how embedded social relations affect economic transactions and intraprofessional stratification in the market for legal services. We argue that networks are critical in fostering individual upward mobility not only because they provide valuable information to ego (lawyer) but also because they function as a crucial mechanism by which the focal actor’s (lawyer’s) status is transmitted to outside alters (clients). Consistent with Podolny’s (1993) general theoretical statement, we claim that ties to prestigious network partners send positive signals concerning the ability and trustworthiness of legal professionals. That is, networks help reduce the information asymmetries faced by potential buyers concerning the actual (unknowable) quality of legal services. Our argument is that, in doing so, the network-embedded status of lawyers serves to contribute to their earnings. We demonstrate this point empirically by examining how networks relate to the process of income attainment among a random sample of lawyers in Chicago (1995).
The Governance of Relations in Markets and Organizations Research in the Sociology of Organizations, Volume 20, 243–266 © 2003 Published by Elsevier Science Ltd. ISSN: 0733-558X/PII: S0733558X02200099
243
244
HARRIS H. KIM AND EDWARD O. LAUMANN
INTRODUCTION The market for legal services is well recognized to be replete with uncertainty on the part of consumers (Abel, 1989; DiMaggio & Louch, 1998; Galanter & Palay, 1991; Nelson, 1988). Given the highly esoteric nature of legal issues and related expertise, consumers are often at the mercy of the decisions and advice put forth by their service providers. Even corporate clients with relatively sophisticated internal legal staff face a great deal of information problem in not only finding competent and dependable outside lawyers but also evaluating the quality of the services rendered (Nelson, 1988). This chapter addresses the role of social ties in mediating this market uncertainty surrounding the buyers and sellers of legal services, and how that ultimately affects the stratification of legal professionals. More specifically, it shows how embedded relations and market transactions relate to the income inequality among individual lawyers. As Podolny (1993; see also Podolny & Hsu, 2003 in this volume) points out, networks can serve as a site whereby the status of focal actors (lawyers) can be inferred by external alters (clients). Similarly, networks also provide important referral benefits to ego (Burt, 1992). Our main claim in this chapter is that ties to prestigious network partners send positive signals concerning the ability and trustworthiness of legal professionals (cf. learning effects in Buskens & Raub, 2002). That is, lawyers’ networks help reduce the information asymmetries faced by potential buyers concerning the actual (unknowable) quality of legal services. Moreover, network partners also play the brokerage role of linking potential buyers and sellers of legal services. We contend that, other things being equal, greater access to more prestigious alters, what we call “notables” in this chapter, leads to greater network embedded status benefits that in turn translates into higher earnings. In his well-known work, Burt (1992) introduces the concept of structural holes, i.e. the relational gap between nonredundant network contacts. He argues that access to structural holes holds the key to entrepreneurial success, for both individual and organizational actors, by providing important information and control benefits. His theoretical project extends Granovetter’s (1973, 1974) previous thesis concerning the “strength of weak ties,” which focused more or less exclusively on the information story. Drawing on the Simmelian concept of tertius gaudens (i.e. “the third who benefits”), Burt discusses the additional advantages of structural autonomy and strategic bargaining associated with networks rich in structural holes. In specifying the variety of information benefits, he also mentions, along with access and timing, an additional aspect which did not figure prominently in Granovetter’s earlier formulations: that is, referrals. As Burt explains,
Network Endorsement and Social Stratification in the Legal Profession
245
[the] network that filters information coming to you also directs, concentrates, and legitimates information about you going to others. . . . [That is, referrals] are the motor expanding the third category of people in your network, the players you don’t know who are aware of you (Burt, 1992, p. 14).
Curiously, however, this concept does not receive any systematic empirical treatment in Burt’s book that bears the title of his novel theoretical construct. Though he fully recognizes network referrals as a “positive force for future opportunities” (Burt, 1992, p. 14), this recognition is not fully incorporated into his analytical repertoire in understanding his examples of how organizations turn a profit or individuals get ahead in organizational context. In fact, his analyses of business firms and managers focus strictly on, aside from the topic of control or autonomy, the issue of timely and reliable information, that is, the centripetal information transmissions from alters to ego. In this chapter, we investigate the functionality of networks that reverses this information flow. More specifically, we analyze how networks convey information about the focal actor to those outside the egocentric network, and, more importantly, the economic ramifications this reversed, centrifugal flow has for ego.
STATUS POSITION AND MARKET INEQUALITY Podolny’s (1993) “status-based model of competition” provides a convincing theoretical rationale for the argument that networks serve the additional role of sending information about ego to, not only from, alters. The fundamental assumption underlying Podolny’s status-based explanation is that there is a decoupling or loose linkage between the actual quality of the producer’s goods or services and the quality that is perceived during the economic transactions. Market status, defined as the producer’s perceived quality, serves as a signal of that which is, by definition, unobservable. And the loose linkage between perceived and actual quality is, according to him, “mediated by a producer’s ties to others in the market” (1993, p. 832). This relational emphasis is what differentiates the network conception of status from the asocial economic conception of reputation, as reflected in, for instance, Williamson’s (1991) writing. For Podolny and other network-oriented scholars, “an actor’s status derives not from past demonstrations of quality; it derives from the status of the actor’s exchange partners” (Podolny, 1993, p. 460). As he further explains: If an actor is uncertain of the actual quality of the goods that confront her in the market, or if she is unwilling or unable to bear the search costs of investigating all the different products in the market, then the regard that other market participants have for a given producer is a fairly strong indicator of the quality of that producer’s output (Podolny, 1993, p. 831).
246
HARRIS H. KIM AND EDWARD O. LAUMANN
Podolny specifies three key types of “other market participants”: namely, consumers, third parties, and co-producers. In other words, the focal actor’s preexisting network of relations with these types of alters powerfully determines its own status as well as how future potential transactional partners may view the quality of ego’s product or performance. And, more importantly, market rewards are differentially allocated by the particular position an actor occupies on the status hierarchy: [This is because] status lowers the transaction costs associated with the exchange between buyer and seller. Implicit and explicit promises of a higher-status producer regarding product quality are more likely to be accepted; therefore, the higher-status producer need not devote as much time or expense to convincing the buyer or relevant third parties of the validity of its claims (Podolny, 1993, p. 838).
Podolny also cites lower advertising and financial costs as a source of economic advantage enjoyed by high-status actors. Based on this logic, he illustrates that higher-status investment banks can underbid their lower-status competitors in underwriting securities. In this way, he says, Merton’s notion of “Matthew Effect” contributes to market concentration and inequality. In another article, Podolny and Phillips (1996) make a similar observation that market exchange often involves not only the manifest transfer of goods and resources, but the latent transfer of status. To the extent that actors in a market can be arrayed in terms of status and to the extent exchanges can be witnessed or verified by third parties, actors transfer some of their status when they enter into exchanges with one another (Podolny & Phillips, 1996, p. 453).
Status thus functions as an important signal of quality because market transactions, and, by extension, much of the social interaction in general, entail a process of status conferral that serves to alleviate the problem of information asymmetry on the demand side of the market.
SOCIAL NETWORKS: RESOURCE VERSUS STATUS BENEFITS In his review of extant research by economic sociologists and organizational scholars, Podolny (2001) introduces two useful metaphors to describe how social networks have been conceptualized in the literature: “pipes” and “prisms.” On the one hand, scholars have stressed the network-as-pipes metaphor by concentrating on the resource benefits of structural embeddedness for focal actors. On the other, the network-as-prisms metaphor suggests how network ties function as a market signal to alleviate the uncertainty about quality facing the focal actors’ audience. In doing so, Podolny makes an important analytic distinction between what he
Network Endorsement and Social Stratification in the Legal Profession
247
calls “egocentric” and “altercentric” uncertainty. The former, to large extent, has to do with the pipes metaphor and the uncertainty ego (i.e. producer) confronts in the competitive market. The latter largely deals with the prisms metaphor and alter’s (i.e. consumer’s) uncertainty concerning the ego’s product or service quality. His main point is that studies highlighting the notion of “social capital” explore how network ties serve to alleviate the problem of information asymmetry surrounding the focal actor and how these ties operate to enhance that actor’s economic welfare (e.g. job placement, promotion, income attainment). Those studies emphasizing the status-based functionality of social networks, by contrast, illuminate the mechanism through which affiliation or interaction with prominent network figures serves to emit status-enhancing or status-confirming signals on behalf of ego. In sum, in a market where at least a minimum degree of quality uncertainty exists, the focal actor’s network structure not only functions as a mechanism of resource transfer that benefits ego but also as a powerful source of status transfer that decreases altercentric uncertainty. As Podolny correctly points out, much of the previous scholarship on networks and social stratification has indeed dealt specifically with the information and, to lesser extent, control benefits of network embeddedness (i.e. egocentric uncertainty). Past studies focus on the economic value of relational linkages in, for example, aiding the process of job search (e.g. Bian, 1997; de Graaf & Flap, 1988; Fernandez & Weinberg, 1997; Granovetter, 1973, 1974; Lin, 1999, 2001; Wegener, 1991), job hiring (Neckerman & Fernandez, 2003 in this volume), intraorganizational promotion (e.g. Burt, 1992, 1997; Gabbay & Zuckerman, 1998; Podolny & Baron, 1997), or governance and value creation in alliances (see Gulati & Wang, 2003; Stuart, 2003 in this volume). Following Podolny’s theoretical lead, organizational scholars, on the other hand, have paid increasing attention to the other, relatively unexplored, aspect of social networks (i.e. altercentric uncertainty), which has to do with the conferral of status via network affiliations with known status superiors (e.g. Baker et al., 1998; Gulati & Gargiulo, 1999; Podolny, 1993, 1994; Podolny et al., 1997; Podolny & Stuart, 1995; Stuart et al., 1999). The gist of the argument in organizations literature is that because of the inherent quality uncertainty, potential transactional partners rely on the status of the focal producer’s (provider’s) past or existing network members as a proxy for assessing ego’s capacity to deliver high-quality goods (services). That is, the prestige of one’s network alters serves as a market signal to help reduce the information asymmetry confronting prospective buyers. Moreover, the argument goes, this status effect interacts with the level of market uncertainty, i.e. the effects of network-embedded status vary according to the degree of client’s information problem.
248
HARRIS H. KIM AND EDWARD O. LAUMANN
In this chapter, we draw on this theoretical insight in analyzing and making sense of a representative survey of Chicago lawyers. We argue that a lawyer benefits not only from having access to social resources, as network-oriented scholars have typically argued, but also from the endorsement by high-status network partners, given that a lawyer’s exact performance quality is highly uncertain, even unknowable, from the viewpoint of prospective clients (see, e.g. Stuart et al., 1999). The process of status transfer by affiliation with prominent others, in other words, increases the lawyers’ reputation and in turn their earnings.
UNCERTAINTY, STATUS, AND STRATIFICATION While we believe the network-as-prisms thesis offers significant analytic mileage in studying the dynamics of income attainment in the legal profession, there is another possible mechanism underlying the relationship between networks and stratification in the legal services market. Aside from the information benefits, as we have emphasized above, there may be some control or governance benefits related to network embeddedness. More specifically, networks can serve to mitigate opportunism in market exchange through reputation effects (see, e.g. Burt & Knez, 1995; Buskens, 2002; Greif, 1994; Raub & Weesie, 1990). Consider the following simple scenario. Suppose a client contacts a mutual source in searching for and recruiting a legal service provider. What this network triangulation does is it minimizes the provider’s likelihood of engaging in malfeasance. This is because the buyer can sanction or impose penalty on the seller’s opportunistic behavior, by informing the third-party contact of the seller’s damaged reputation. This idea is nicely captured by Granovetter (1992). As he writes: My mortification at cheating a friend of long standing may be substantial even when undiscovered. It may increase when the friend becomes aware of it. But it may become even more unbearable when our mutual friends uncover the deceit and tell one another (Granovetter, 1992, p. 44; italics added).
In the economic realm, whenever there is a problem of trust between exchange partners, network embeddedness via third-party connections helps to contain malfeasance and thereby facilitates dyadic interaction. This is certainly a plausible causal story in many instances, but it fails to shed adequate light on our own empirical case. In the business of buying and selling legal services, the reputation concerns of the lawyer can indeed discourage him or her from cheating the client or not delivering high-quality services, and this network functionality can in turn serve to strengthen the existing client-provider relations. Yet such control benefits merely result from the information benefits we have stressed earlier. For instance,
Network Endorsement and Social Stratification in the Legal Profession
249
when a firm is involved in, say, corporate litigation, the general counsel typically goes through network channels to locate and hire the best outside lawyer for the job. Here, the primary function of network ties is to reduce the client’s information asymmetry concerning the reputation and competence of the lawyer. To the extent that the lawyer is tied to many high-status members of the legal community, he receives greater endorsement and sponsorship in the eyes of the client. And in this way, network ties allow better-connected lawyers to acquire more and better clients. The decision to hire a lawyer thus has much less to do with whether or not that lawyer can be trusted, but rather with whether he is fit to do the job right. The informational aspect of network embeddedness, therefore, is causally prior. For this reason, while fully recognizing the associated governance benefits, we focus on the notion that networks primarily function as market signals of quality. Indeed, the fundamental assumption behind Podolny’s (1993) original theoretical statement firmly holds in the market for lawyers: there is undoubtedly a tremendous degree of quality uncertainty regarding legal services, and this constitutes a serious problem for many would-be clients. Nelson’s (1988) description aptly captures the problem of uncertainty in the (corporate) legal services market: Even in litigation, where there are seemingly straightforward outcomes, how do clients assess the performance of counsel? If lawyers win a case there is the question of whether the expenditure on legal costs was worth the results or whether a less costly settlement could have been reached. If they lose, the question is whether different tactics could have changed the result. Given the uncertainties, clients often must live with their choices, trusting that counsel will perform competently. Inside legal officers may provide a more sophisticated evaluation of the performance of outside lawyers, but ultimately they too must make a leap of faith and rely on the judgement of the chosen external counsel (Nelson, 1988, p. 67).
Consistent with Nelson’s apt portrayal of the legal market as “a highly textured set of networks,” we argue that consumers rely on preexisting ties to gain crucial information about lawyers. In the legal services market, buyers actively seek and consider the opinions of others regarding a given producer as “a fairly strong indicator of the quality of that producer’s output” (Podolny, 1993, p. 831). Or, as Nelson explains, clients attempt to minimize [market] uncertainty by selecting lawyers they know personally or to whom they have been referred by a trusted source (Nelson, 1988, p. 67).
The following quote from the interview conducted with a general counsel of a major Chicago-based telecommunications company further highlights this theoretical point of view. When asked how he would go about hiring outside legal service providers, he offers a poignant statement on the role of social ties in the search process:
250
HARRIS H. KIM AND EDWARD O. LAUMANN
I ask people. I think there is a [referral] system at work . . . I know the reputation of various firms, and what they’re good at. So I have that base of knowledge, and you supplement it with talking to general counsels of similar companies. There’s a lot of telecom companies in Chicago, and I’ve talked to them. You know, who do you use, and there’s some networking that goes on.
It is our specific contention that the prestige of lawyers’ social network alters functions as one critical source of market signal, that is, the “perceived” quality of lawyers’ work. This implies that lawyers tied to elite actors in the professional community are at a competitive advantage vis-`a-vis those who lack such ties, since network connections confer status benefits on lawyers whose performance ability is difficult to evaluate. And this competitive advantage via network affiliation, we maintain, comes primarily from the mechanism of peer endorsement or sponsorship. As Sandefur et al. (1999) remark in an earlier study, a lawyer’s “reputation benefits both from one’s seal of approval of a respected judge and from the ramifying nature of elite lawyer’s social network” (Sandefur et al., 1999, p. 221). Knowing high status network partners is crucial because it is a primary mechanism by which a lawyer’s own status is legitimated and endorsed in the eyes of others, both coproducers and buyers. Network connections thus provide not only direct benefits of embedded resources like case-specific legal advice or other general information about contingent opportunities or strategies, but also more indirect symbolic benefits through the network partners’ sponsorship. The endorsement process by virtue of network affiliation is thus a key element in reducing the informational difficulty the audience confronts. And this reduced demand-side uncertainty leads to more propitious business opportunities for lawyers in terms of recruiting more lucrative and durable clients and ultimately translating their embedded lawyer-client ties (via third-party endorsement) into higher earnings. In their comprehensive analysis of the General Social Survey (1996) data, DiMaggio and Louch (1998) present ample evidence documenting the socially embedded nature of consumer transactions. According to them, the issues of buyers’ product uncertainty and seller’s propensity toward opportunism lead consumers toward “search embeddedness” (the use of known brokers or third parties) and/or “within-network exchange” (the transaction with sellers who are directly connected by preexisting social ties). Buskens et al. (2003) test a similar argument for partner selection in buyer-supplier relations in this volume. The purchasers of legal services are in no way exempt from this dilemma of information asymmetry. In fact, as their study shows, under conditions of higher uncertainty, such as purchasing a car or seeking legal advice, consumers are in fact more likely to pursue a socially embedded, as opposed to an asocial arm’s length, transactional strategy. We contend that because of the high uncertainty that characterize the market for lawyers, economic actors, or buyers, do not engage in the search process in a
Network Endorsement and Social Stratification in the Legal Profession
251
socially atomistic fashion. Instead, as Granovetter’s (1985) well-known theoretical argument states, their actions are embedded in their networks of interpersonal relations. Using social ties allows actors to gain otherwise unavailable knowledge about the dependability, trustworthiness, and competency of potential service providers. Because of the high costs involved in locating and evaluating the quality of lawyers, clients embed their search process by tapping into information sources available within their networks, that is, they engage in what DiMaggio and Louch (1998) have called “search embeddedness.” Invoking existing network relations becomes an effective way of economizing on search costs. The strategy of employing the “within-network exchange” further, albeit to lesser extent, serves to economize on transaction costs of interacting with lawyers and monitoring (or minimizing opportunistic) their behavior. In this way, consumers of legal services govern their market transactions by embedding them in social relations. Lawyers’ networks, namely, their ties to prestigious members of the legal community, play a pivotal role in facilitating this process of embeddedness. Network oriented organizational scholars have paid substantial attention to a novel type of interorganizational behavior. Moving beyond Williamson’s (1981, 1985) markets and hierarchies paradigm, they have proposed “network forms of organizations” or what has also been characterized as “relational contracting” (Bradach & Eccles, 1989; Podolny & Page, 1998; Powell, 1990) as an alternative governance mechanism to price and authority emphasized by the transaction cost approach. The basic insight is that network embeddedness of organizations offers certain functional advantages, which neither the price mechanism (market) nor institutional authority (hierarchy) alone can afford. Through such measures as, facilitating fine-grained information transfers, imposing reciprocal obligations and discouraging opportunism, and building mutual trust, dependency, and joint-problem solving capacities, social networks serve to minimize transaction costs and thereby promote interorganizational transactions (e.g. Uzzi, 1996, 1999; see also Buskens et al., 2003 in this volume). What organizational scholars refer to as network form of governance or relational contracting clearly operates at the interpersonal level as well in the professional services market: lawyers and clients do not somehow act independently as if they exist in a social vacuum. The lawyer-client relationship is not purely driven by the economic laws of supply and demand but is fully embedded in the web of ties that link buyers and sellers through third parties or network brokers (i.e. notables). And the differential levels of this type of market embeddedness, we argue, accounts for some of the intraprofessional income stratification. Our main argument can be formally stated as follows:
252
HARRIS H. KIM AND EDWARD O. LAUMANN
Hypothesis 1. The greater the number of network connections (acquaintance ties) to notables, the greater the individual lawyer’s earnings advantage. We also argue that, in keeping with the growing empirical research on status and organizational life chances, access to notable connections interacts with the degree of audience uncertainty in influencing lawyers’ income. Stinchcombe’s (1965) insightful concept of liability of newness, which has attracted a tremendous amount of attention over the years especially from organizational ecologists, sheds much light on the contingent value of network-embedded status. Following Stinchcombe’s seminal idea, Hannan and Freeman (1989), in their systematic presentation of the population ecology perspective, postulate that younger organizations experience higher rates of mortality and demonstrate statistically significant age dependence in failure rates across several industries and organizational types. Older organizations are less vulnerable to environmental selection, since they come to possess “improved capabilities and more secure structural positions” that enhance their life chances vis-`a-vis their younger counterparts (Carroll & Hannan, 2000, p. 281). In looking at the legal profession, we suggest that the liability associated with newness applies to the level of individual lawyers and their status attainment as well. While we in no way see this process as inevitable, we believe it offers valuable analytic utility in making sense of the interaction between notable ties and altercentric uncertainty. For our purposes, the concept deals with the greater costs of evaluating the quality of younger lawyers. That is, it has to do with the external client perception. Young lawyers, by definition, lack professional experience or a proven track record from which potential consumers can judge how good they actually are or potentially can be. Hence, along with other signals of quality, network affiliates of prominence that can confer reputation benefits via endorsement help reduce this type of market uncertainty. This network-mediated process thus should produce greater benefits under conditions of higher uncertainty, i.e. for younger practitioners. This leads to our second, related, hypothesis to be tested: Hypothesis 2a. The earnings return on status is greater for younger (i.e. less experienced) lawyers. In the business of producing and selling legal work, the reputation of individual practitioners also derives in part from their institutional affiliation. As students of the legal services industry have pointed out (e.g. Abel, 1989; Galanter & Palay, 1991; Heinz & Laumann, 1982; Nelson, 1988), the legal profession has become highly fragmented and stratified over the years along organizational lines. Thus, along with age, organizational tenure (i.e. professional experience) serves as an important barometer of the overall ability and competence of individual lawyers.
Network Endorsement and Social Stratification in the Legal Profession
253
That is, organizational affiliation or tenure can be construed as another dimension of the level of client uncertainty. More specifically, shorter tenure implies greater costs of evaluating the quality of lawyers from the client’s point of view. Thus, ties to high-status peers ought to matter more for those who lack professional experience in the firm setting. Consequently, in addition to Hypothesis 2a, we further propose that: Hypothesis 2b. The earnings return on status is greater for lawyers with shorter organizational tenure. Estimating these interaction effects, as implied by the last two hypotheses, is critical in illustrating our theoretical claim that the benefits of network ties are, to large extent, symbolic in nature. In the literature on social networks and social stratification, as discussed above, much academic focus has been on how networks serve to receive outside information to ego that is located within. Our argument places the analytic focus on the reversed flow of information, by analyzing how network ties send signals of reputation and legitimacy (of especially unseasoned lawyers) to those located outside the egocentric network. In short, lawyers’ ties to elite partners pay because buyers infer the status of sellers from such network relations, in their attempts to reduce the information asymmetry they face in purchasing legal services. Moreover, ties to high-status, or legitimate, others matter more when it is more costly for alter(s) to make sense of the actual quality of ego’s product or service, whether the focal actor of uncertainty be an organization or a person. We examine below this critical contingency effect, on which our theoretical argument hinges, by looking at the determinants of lawyers’ income attainment.
DATA, MODEL, AND VARIABLES The data analyzed in this chapter come from face-to-face interviews with a 1995 probability sample of all lawyers practicing in the city of Chicago. The names of the respondents were obtained from the state’s official list of licensed lawyers, who are required to register with the Attorney Registration and Disciplinary Commission, an agency under the supervision of the Illinois Supreme Court. This survey was funded by the American Bar Foundation in order to examine the changes in the overall social structure of the Chicago bar, by comparing it to the earlier one taken in 1975 analyzed and reported by Heinz and Laumann (1982). Given the cross-sectional design of our data, unfortunately, we cannot conclusively settle the issue of causation. The causal sequence between income and social capital and status measures cannot be proved with the available empirical
254
HARRIS H. KIM AND EDWARD O. LAUMANN
snapshot of lawyers and their attributes. This critical issue has been explicitly raised by a number of previous studies of networks and stratification (e.g. Boxman et al., 1991; Burt, 1992; Podolny & Baron, 1997; Sandefur et al., 1999; Stoloff et al., 1999). The central question for us is whether network position or structure indeed precedes income attainment, since it’s possible that the latter, even if partially, causes rather than fully results from the former. Our concern here is similar to Burt’s, for whom the pivotal question is: Do the information and control benefits possible in a network give an advantage to a manager? Or, do advantaged managers build networks rich in information and control benefits? . . . [As he recognizes, the answer hinges on] how much of the association between structural holes and promotion happens before promotion (Burt, 1992, p. 173; italics added).
This too holds the key for our own analytic agenda. While it may be the case that more successful lawyers are more likely to have access to favorable social capital (status benefits) via notable ties, we reject the alternative explanation that the causal arrow necessarily extends from status attainment to social networks. Rather, in advancing our network argument, we assume the causation runs in the opposite direction, that is, earnings reflect network effects. This is because lawyers’ relationships with their colleagues and notables are, first of all, relatively long-term and most likely constructed prior to the year in which the lawyers’ income was reported (see Sandefur et al., 1999, pp. 229–231). Also, generally speaking, interpersonal network ties in the organizational setting do not fluctuate but are “quite stable” over time (Burt, 1992, p. 174). Thus while we acknowledge our data limitations and the need for more longitudinal studies on the topic of our interest, we believe our hypotheses nevertheless receive valid, albeit inconclusive, empirical scrutiny. Our conclusion is further supported by additional analyses using the two-stage least squares (2SLS) method. We find that the inclusion of instrumental variables does not significantly change the main findings from our OLS regression models. The survey sample contains 788 lawyers who were randomly selected from the state’s official list of licensed attorney. Each interview lasted on average one and a half hours. The response rate was around 82%. The sampling frame consists of all lawyers with offices in the City of Chicago. The survey subjects cover a wide spectrum, including solo practitioners, lawyers in private firms, corporate house counsel, government lawyers, public defenders, judges, and law professors, as well as those who were retired, unemployed, or engaged in occupations other than law (see Heinz et al., 1997 for a fuller description). Excluding judges and law faculty, there are 675 respondents in the data set who were practicing law full time at the time of the survey. This reduced sample comprises four distinct segments of the legal practice: private firm, corporate counsel, public sector, and self-employed. Of the practicing
Network Endorsement and Social Stratification in the Legal Profession
255
lawyers in the sample, there are 185 females (27%) and 490 males (73%). Only eight percent of them belong to minority groups – Asian, Latino, or African American. In terms of the legal practice settings, 15% work as solo practitioners, 61% work in private firms, the largest category, a much smaller 9% are found in government law offices, and the rest hold house counsel positions in corporations. Since only the private lawyers are concerned with recruiting clients, however, our analysis only deals with the specific subsample of respondents who work in the firm setting as well as those that are self-employed. The dependent variable in all our OLS regression models is annual earnings. We model the natural log of income to address the problem of outlying observations. A variety of lawyers’ personal and work characteristics affect the result of income attainment. As controls, we consider a host of variables in our models as proxies for human capital and demographic or ascriptive characteristics, including gender (1 = “male”), marital status (1 = “married”), race (1 = “white”), age and age squared (to control for the nonlinear effect), partner and specialist statuses (1 = “yes”), number of weekly hours worked, proportion of business clients (measured on a continuous percentage scale), class rank, law school standing, organizational size and tenure, and practice context (1 = “solo”). Scholars of the legal profession have observed time and again that the legal profession has traditionally excluded women and members of the minority groups. Though their entry has been steadily increasing, they have not enjoyed the relatively high earnings and other advantages associated with the privileged positions typically occupied by their white male counterparts (Epstein, 1993; Hagan et al., 1991; Hull & Nelson, 2000; Kay, 1997). The legal profession has also been extremely fragmented in terms of the academic credentials of its members. As Heinz and Laumann (1982) observed two decades ago, for starting lawyers the status of the law school one attended plays a pivotal role in shaping his or her legal career. One’s school status largely determines where one begins, and, as they explain, “where you begin powerfully shapes where you end up” (Heinz & Laumann, 1982, p. 197). The legal profession has also become increasingly bifurcated along client lines, as their well-known “dual-hemisphere” thesis suggests (Heinz et al., 1997; Heinz & Laumann, 1982). That is, the legal services industry has become split into two distinct camps: those that provide services for large corporate clients and those who work mainly for small or individual clients. In light of this growing trend, controlling for the percentage of business clients is critical in testing our status endorsement hypothesis, since this attribute is the single most important factor in determining the prestige level of legal practice types as well as legal specialties (Heinz & Laumann, 1982). By extension, a lawyer who services a relatively high proportion of corporate business clientele may be regarded to be more prestigious (of higher status) by both peers and clients. Organizational size is
256
HARRIS H. KIM AND EDWARD O. LAUMANN
also a powerful indicator of prestige. As Abel (1989, p. 183) puts it, it often serves as “an important surrogate for quality” of not only law firms but also their employees. Our regression models estimate the network effects net of these relevant control variables. To explain the controls in more detail, the variable for class rank, which measures the lawyer’s past academic performance in law school, consists of two dummy indicators: one for the top 10% of graduating class and another for the lower range of top 11% to 25%. The omitted category is the rank below the top 25% of class. Law school rank is divided into four categories: elite (e.g. University of Chicago), prestige (e.g. Northwestern), regional (e.g. University of Illinois), and local (e.g. De Paul). The local law schools constitute the baseline category. We treat the private firm lawyers as the reference group and compare them to the solo practitioners. Organizational size is the logged number of lawyers working in the same firm. It is recorded as “1” for the solo category. The key variable for our network endorsement argument has to do with the number of ties to elite members of the legal professional community. The total of 65 such members (notables) was selected after extensive consultation with informants who are very familiar with various aspects of the Chicago bar. An effort was made to list prominent lawyers representing a considerable range of social and professional categories. A more detailed account of the selection criteria is found in Heinz et al. (1997). A unique feature of these network data is that aside from the ties between the respondents and the notables, we also have information on the ties among the notables themselves. This additional information allows us to construct a network centrality measure to hierarchically differentiate the 65 notables in terms of status or power (see Wasserman & Faust, 1994, ch. 5). So instead of simply counting the number of ties to elite outside lawyers each respondent in the dataset may have, we can construct a more accurate measure of social capital by assigning weights to the notables based on their relative standings within this community of elites. In our study, we use the actor degree centrality (Wasserman & Faust, 1994, pp. 178–180) to construct the network weights, which is defined as the proportion of nodes that are adjacent to each actor.1 The survey data contain information on two types of notable linkages. During the face-to-face interview, respondents were asked to indicate the names of notables whom “they know well enough to call by their first name when you see them.” They were also asked to identify those from whom they believe they could receive legal advice free of charge. These two relationships thus constitute, respectively, “weak” and “strong” ties. Of the subsample, about 30% have no notable ties. Another 35% know from 1 to 3 notables, and about 10% of the respondents are connected to 10 notables. Less than 1% report acquaintance ties with as many as
Network Endorsement and Social Stratification in the Legal Profession
257
35 notables. Using this personal “acquaintance” measure over the alternative of the “advice” network allows a more conservative test of our main argument, as contained in Hypothesis 1, that the causal mechanism behind the effect of network ties on income is indeed elite sponsorship and concomitant status-enhancement. The age variable is the key proxy for work experience and, more importantly for our theoretical purposes, professional reputation, i.e. the level of altercentric uncertainty. The other measure of altercentric uncertainty is organizational tenure, measured as the number of years the respondent had been working for the law firm at the time of the survey. Since solo practitioners do not, by definition, work in a firm setting, they are dropped from the second part of the analysis when using the tenure variable. Organizational scholars conventionally use the age of organizations, along with their size, as a proxy for the level of uncertainty. Some have used the product term of status (or legitimacy) and age in their analyses to predict firm performance (e.g. Stuart, 1999; Stuart et al., 1999) and organizational failure (e.g. Baum & Oliver, 1991). The guiding assumption there is that status benefits or organizational legitimacy counteracts against the liability of newness (and/or “smallness”). Similarly, we anticipate a statistical interaction between the network-embedded status, or reputation, of a lawyer (measured as ties to notables) and his age and organizational tenure with respect to his income. Hypotheses 2a and 2b will be tested using these two interaction terms.
FINDINGS AND DISCUSSION Tables 1 and 2 contain the results of the regression analysis, using age and tenure as the measure of client uncertainty, respectively. With respect to the first regression using the full sample (N = 470), model 1 consists of just the control variables, including human capital and ascribed characteristics. Many of the variables included in this model reach the standard level of significance. Controlling for other variables, older lawyers enjoy higher income, though the effect of age on income is curvilinear, consistent with the conventional notion that the positive effect of professional experience on earnings tapers off eventually. Having the partner status increases one’s income, consistent with the fact that partners enjoy the unique benefits of profit sharing. Hours worked is significantly related to earnings as well. And so is the percentage of business clients. Class rank and law school prestige also both positively contribute to lawyers’ income. Working in the private firm or as internal counsel is associated with higher income vis-`a-vis working as a solo. And organizational size also significantly raises the earnings level, net of other variables. Consistent results are reported from the baseline model in Table 2 (N = 376) as well, which excludes solo practitioners.
258
HARRIS H. KIM AND EDWARD O. LAUMANN
Table 1. Unstandardized OLS Coefficients from the Regression of Logged Income on Notable Ties and Control Variables. Variables
Model 1
Model 2
Model 3
Coefficient
t-value
Coefficient
t-value
Coefficient
t-value
Male Married White Age Age2 Partner Specialist Hours Clients Top 10% Top 11–25% Elite law Prest law Reg law Solo Orgsize Notables Age × Notables
0.097 0.155∗ 0.099 0.124∗∗∗ −0.001∗∗∗ 0.513∗∗∗ 0.048 0.016∗∗∗ 0.001 0.255∗∗∗ 0.193∗∗ 0.359∗∗∗ 0.201∗ 0.149∗ −0.239∗ 0.115∗∗∗
1.27 2.22 0.762 5.94 −5.04 6.34 0.723 7.26 0.963 3.52 2.78 3.93 2.13 2.13 −2.37 3.67
0.102 0.164∗ 0.103 0.113∗∗∗ −0.001∗∗∗ 0.506∗∗∗ 0.024 0.015∗∗∗ 0.001 0.236∗∗∗ 0.197∗∗ 0.306∗∗ 0.189∗ 0.147∗ −0.218∗ 0.093∗∗ 0.001∗∗∗
1.39 2.43 0.821 5.61 −5.05 6.47 0.366 7.29 1.68 3.37 2.94 3.45 2.09 2.19 −2.25 3.06 5.87
0.113 0.162∗ 0.125 0.101∗∗∗ −0.000∗∗∗ 0.493∗∗∗ 0.022 0.015∗∗∗ 0.002 0.225∗∗ 0.191∗∗ 0.280∗∗ 0.184∗ 0.148∗ −0.233∗ 0.091∗∗ 0.003∗∗∗ −0.000∗
1.55 2.41 1.00 4.86 −3.94 6.33 0.344 7.13 1.84 3.23 2.87 3.16 2.03 2.20 −2.41 2.99 3.63 −2.54
Constant Adjusted R2 Number of cases
6.41∗∗∗ 0.541 470
∗
6.97∗∗∗ 0.572 470
6.92∗∗∗ 0.578 470
p < 0.05; ∗∗ p < 0.01; ∗∗∗ p < 0.001 (2-tailed tests).
Model 2 from Table 1 introduces the network variable that measures the degree of ties to notables. Including this variable improves the overall fit of the model. As hypothesized, the statistical results show that the number of notable ties, weighted by the indegree centrality measure based on the intra-notable network relations, is significantly related to income ( p < 0.001). Because these notables occupy influential brokerage positions by virtue of their status, they possess a great deal of power and capacity at their disposal, which can benefit those who are connected to them. A lawyer can expect to increase his or her status and reputation by virtue of affiliation with a high-status alter. In the market for legal services, where clients have great difficulty locating and evaluating able providers, notable connections can serve as a powerful source of social capital. Endorsement by a well-known lawyer may be seen as an effective form of social capital that elevates the legitimacy of a less prestigious lawyer in the eyes of a potential client. From the client’s point of view, relying on the opinions of
Network Endorsement and Social Stratification in the Legal Profession
259
Table 2. Unstandardized OLS Coefficients from the Regression of Logged Income on Notable Ties and Control Variables (w/o Solo Category). Variables
Model 1 Coefficient
Male Married White Age Age2 Partner Specialist Hours Clients Top 10% Top 11–25% Elite law Prest law Reg law Orgsize Tenure Notables Tenure × Notables Constant Adjusted R2 Number of cases ∗p
Model 2
Model 3
t-value
Coefficient
t-value
Coefficient t-value
0.064 0.794 0.075 0.956 0.081 0.532 0.131∗∗∗ 5.84 −0.001∗∗∗ −5.49 5.67 0.468∗∗∗ 0.107 1.45 0.013∗∗∗ 4.78 0.000 0.286 0.265∗∗ 3.43 2.60 0.189∗ 0.413∗∗∗ 4.33 0.244∗ 2.51 0.161∗ 2.14 0.102∗∗ 3.20 4.79 0.026∗∗∗
0.067 0.071 0.058 0.122∗∗∗ −0.001∗∗∗ 0.463∗∗∗ 0.095 0.013∗∗∗ 0.000 0.259∗∗ 0.205∗∗ 0.385∗∗∗ 0.236∗ 0.166∗ 0.075∗ 0.027∗∗∗ 0.001∗∗∗
0.857 0.987 0.311 5.60 5.62 5.83 1.32 4.88 0.302 3.49 2.95 4.20 2.61 2.35 2.32 5.05 5.71
0.077 1.02 0.064 0.874 0.049 0.219 0.110∗∗∗ 5.05 −0.001∗∗∗ 5.15 0.405∗∗∗ 5.12 0.085 1.21 0.013∗∗∗ 4.77 0.000 0.425 0.242∗∗ 3.32 0.212∗∗ 3.08 0.369∗∗∗ 4.08 0.238∗ 2.59 0.182∗ 2.53 0.071∗ 2.22 0.039∗∗∗ 6.28 0.002∗∗∗ 6.47 −0.000∗∗∗ −3.63
6.71∗∗∗ 0.558 376
6.97∗∗∗ 0.594 376
7.20∗∗∗ 0.608 376
< 0.05.∗ p < 0.05; ∗∗ p < 0.01; ∗∗∗ p < 0.001 (2-tailed tests).
reputable members of the profession is a sure way to economize on the costs involved in searching for a lawyer. From lawyers’ point of view, referral and status benefits of ties to notables allow them to construct and maintain more economically opportune provider-client relationships. Consequently, as the findings indicate, lawyers with greater access to notable ties enjoy greater earnings advantage. This argument receives additional support from the regression model (model 2) from Table 2, looking at the firm lawyers only. And these effects are clearly non-trivial vis-`a-vis other control variables. The relatively small metric coefficients from the regression tables may give the opposite impression, however. This is because one unit increase in the Notables variable is not a dyadic tie but a point increase in the indegree centrality measure. When we measure this variable as a raw (unweighted) score, i.e. a simple count of the number of ties, however, we get a regression coefficient of 0.029, controlling for the same variables. That is, ceteris paribus, an additional notable tie is associated
260
HARRIS H. KIM AND EDWARD O. LAUMANN
with a 2.9% increase in earnings, which gives a much greater causal magnitude than the reported value in model 2 (Table 1). When we consider the size of the beta coefficients, the relative strength of the network endorsement is clearer. Looking at the same model, for example, the value of the main network effect indicates that one standard deviation increase is associated with 0.201 std. dev. increase in the dependent variable. In comparison, having partnership status yields the coefficient of 0.278, the largest value in this model, followed by the number of hours billed (0.233). Profit sharing and work output (productivity), respectively, associated with these two covariates thus contribute most powerfully to lawyers’ income attainment. Knowing elite peers in the legal community rivals these factors in terms of causal impact. By comparison, graduating in top 10% of law school and coming from an elite institution both give the coefficient of 0.121, significantly lower than the observed main network effect. Model 2 from Table 2 also produces similar results. Model 3 in Table 1 contains the interaction term, which is included to test the prediction that the network endorsement process has a contingent impact on lawyers’ earnings. Its negative and significant coefficient ( p < 0.05) supports our hypothesis that under more uncertain conditions (i.e. when lawyers are younger and hence their quality more difficult to discern), status benefits from knowing powerful and well-known peers are greater. This is consistent with the prediction that the effects of knowing prominent members of the legal community should decline with lawyers’ age, that is, as the altercentric uncertainty level surrounding the quality of a focal actor declines with professional experience. Having high status network partners pays but only up to a certain point, after which its benefit begins to diminish. This is because as a lawyer gains more experience he or she leaves a more visible track record from which a potential client can more readily evaluate that lawyer’s quality. Aside from the reputation benefits, the referral benefits of notable ties also fall in value, to some extent, since a veteran lawyer’s dependence on them for new clients also diminishes, as his or her own experience and expertise increasingly speak for themselves. This contingency argument is borne out by additional evidence from Table 2. As the interaction coefficient in model 3 from the second set of regression analysis similarly indicates ( p < 0.001), returns on ties to notables also decline as a function of organizational tenure: network endorsement benefits are greater for novices than for veterans working in the law firm. Given the multicollinearity issue stemming from a high correlation between the status variable and the interaction term (0.98), we also estimated additional models using the mean-deviated values of age and notable ties. In their study of the network status effects on firm growth in the semiconductor industry, Podolny et al. (1997) faced a similar methodological problem. We pursued the same strategy they employed and found consistent results in support of our hypotheses
Network Endorsement and Social Stratification in the Legal Profession
261
using both regression models. This additional analysis, not reported below, further substantiates our claim that ceteris paribus the endorsement of lawyers made possible by the network affiliations with high-status others yields more substantial earnings benefits for younger, and hence less experienced, practitioners. We also ran additional models looking at three-way interactions, in order to further examine our key claim that notable ties really have to do with market signaling effects. We estimated two sets of regressions by disaggregating the data into two sets of subgroups based on the partner and specialist statuses. The statistical findings show that the interaction effects are significant only for the non-partners and specialists. In other words, returns to having notable ties decline with age, but this effect holds only for these two specific categories. This is because by the time a lawyer becomes a partner of a law firm, his already established reputation becomes much more transparent to the demand side of the market. As a result, the quality of a partner can be directly evaluated based on his own past performances. Hence social capital derived from knowing other high-status colleagues gradually loses its causal significance for the partner. On the other hand, the contingent effect holds only for the specialists because in the legal services market they are, by definition, much more difficult to spot and their abilities more difficult to gauge, compared with the generalists. When a client is in need of highly specific legal work, which can only be performed by a specialist, the network ties indeed pay off by providing the required referral benefits through the mechanism of market signaling. Generalists, on the other hand, are more easily found, since many of them are in fact interchangeable, i.e. are trained to provide comparable services. In light of this, the client is less likely to rely on network channels to evaluate and hire a generalist, and, as a result, the generalist benefits less from network-based endorsement. These additional analyses help capture and highlight an important network functionality: ceteris paribus, the focal actor’s success depends on relational ties not only because of what the network affiliates can directly provide for the focal actor but also because of what they can, even if indirectly, convey about ego’s unobservable quality.
CONCLUSION The market for legal services is not a simple amalgamation of anonymous exchanges between lawyers and clients. Rather, it consists of buyers and sellers who are embedded in networks of social relationships that have important ramifications for not only how legal services are purchased but, more importantly, what this means for intraprofessional stratification. Because clients typically lack the required knowledge or information regarding the true quality of potential providers, they seek pre-existing social ties or pay attention to available market
262
HARRIS H. KIM AND EDWARD O. LAUMANN
signals that can help eliminate the uncertainties they face in the marketplace. We have argued in this chapter that ties to notables, more specifically, endorsement and sponsorship by high-status network partners provide the function of reducing what Podolny (2001) calls altercentric uncertainty. Clients look for and rely on such information cues because they reduce the transaction costs involved in searching for the right service provider. Consumer transactions for legal commodities are thus socially embedded in this way, since it is a “cheap” way to deal with the market-related risks involved (DiMaggio & Louch, 1998; Kollock, 1994; see also Buskens et al., 2003 as well as Gulati & Wang, 2003 in this volume). As in the classic case of the market for “lemons” (Akerlof, 1970), buyers in the market for lawyers suffer from serious information asymmetries: sellers exaggerate or hide the actual quality of their products or services, and buyers are inadequately positioned to competently evaluate the validity of sellers’ claims. But unlike the undersocialized actors portrayed in the market for used cars or other economic models, as Granovetter’s (1985) programmatic theoretical piece reminds us, consumers make choices and decisions within the particular context of other economic actors. Social networks are not a mere “frictional drag” that gets in the way (Granovetter, 1985, p. 484), but play a crucial role in facilitating the process of market transactions. They are critical because, as we show in this chapter, networks can exert consequential influence on economic outcomes, in our empirical case, lawyers’ earnings. Lawyers with the right connections reap the symbolic benefits of status conferral as well as the more tangible benefits of client referral. And these benefits vary according to the uncertainty level characteristic of different lawyers. Differential access to notable members of the professional circle in turn produces differential opportunities and concomitant inequality in the market. Network scholars have long stressed how the prospects of individual mobility and stratification partly depend on the relational matrix underlying the focal actor. The increasing sociological usage of the concept of “social capital” has captured the essence of the adage that success really depends not just on what you know but who you know. As Burt explains, “competition is never perfect,” and the “[r]ate of return depends on the relations in which capital is invested” (1992, p. 10). The aim of this chapter was to illuminate this general theoretical point by analyzing the income attainment among a random sample of legal professionals. In doing so, however, it underscored a relatively unexplored aspect of the causal link between social networks and economic actions, that is, how network ties operate as a site where the construction and transmission of ego’s status takes place. Aside from serving as a conduit of information inflow, network alters (e.g. notables) also send out information about ego (e.g. lawyers) to relevant others (e.g. clients) residing outside the boundaries of egocentric network. Previous
Network Endorsement and Social Stratification in the Legal Profession
263
sociological research on interpersonal networks and individual-level stratification has mainly focused on the former functionality of social ties. As this chapter illustrates, social ties matter for another, related yet distinct, reason. Further investigating this other functionality can reveal much about the complex relationships between how, when, and why networks translate into economically relevant social resources, or, in certain instances, fail to do so.
NOTES 1. Based on this definition, a notable is considered to be more central to the extent that he receives more acquaintance ties from other notables. The following notation formally states the definition of this (directional) indegree centrality measure: C D (n i ) = d(n i )/(g − 1), where d(n i ) is the number of received ties for the ith actor and g is the group size. We also used two other standard network variables in lieu of the indegree measure and found consistent results supporting our hypotheses, namely Bonacich’s (1987) eigenvector centrality and Freeman’s (1977) betweenness centrality. The former takes the following expression: C B (n i ) = j
ACKNOWLEDGMENTS An earlier version of this chapter was presented at the Stratification Workshop, University of Chicago. We thank Ross Stolzenberg, the coordinator, and other members of the Workshop for their constructive comments. This research was funded by the American Bar Foundation. We acknowledge John Heinz and Robert Nelson for their support and encouragement. The helpful feedback on earlier drafts from the editors of this volume is also gratefully recognized.
REFERENCES Abel, R. (1989). American lawyers. New York: Oxford University Press. Akerlof, G. A. (1970). The market for ‘lemons’: Quality uncertainty and the market mechanism. Quarterly Journal of Economics, 84, 488–500.
264
HARRIS H. KIM AND EDWARD O. LAUMANN
Baker, W. E., Faulkner, R. R., & Fisher, G. A. (1998). Hazards of the market: The continuity and dissolution of interorganizational market relationships. American Sociological Review, 63, 147– 177. Baum, J. A., & Oliver, C. (1991). Institutional linkages and organizational mortality. Administrative Science Quarterly, 36, 187–218. Bian, Y. (1997). Bringing strong ties back in: Indirect ties, network bridges, and job searches in China. American Sociological Review, 62, 366–385. Bonacich, P. (1987). Power and centrality: A family of measures. American Journal of Sociology, 92, 1170–1182. Boxman, E. A. W., de Graaf, P. M., & Flap, H. D. (1991). The impact of social and human capital on the income attainment of Dutch managers. Social Networks, 13, 51–73. Bradach, J. L., & Eccles, R. G. (1989). Price, authority and trust: From ideal types to plural forms. Annual Review of Sociology, 15, 97–118. Burt, R. S. (1992). Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Burt, R. S. (1997). The contingent value of social capital. Administrative Science Quarterly, 42, 339– 365. Burt, R. S., & Knez, M. (1995). Kinds of third-party effects on trust. Rationality and Society, 7, 255–292. Buskens, V. (2002). Social networks and trust. Boston, MA: Kluwer. Buskens, V., Batenburg, R. S., & Weesie, J. (2003). Embedded partner selection in relations between firms. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 107–133). Oxford: Elsevier Science. Buskens, V., & Raub, W. (2002). Embedded trust: Control and learning. Advances in Group Processes, 19, 167–202. Carroll, G. R., & Hannan, M. T. (2000). The demography of corporations and industries. Princeton, NJ: Princeton University Press. DiMaggio, P., & Louch, H. (1998). Socially embedded consumer transactions: For what kinds of purchases do people most often use networks. American Sociological Review, 63, 619–637. Epstein, C. F. (1993). Women in law. Chicago, IL: University of Illinois Press. Fernandez, R. M., & Weinberg, N. (1997). Sifting and sorting: Personal contacts and hiring in a retail bank. American Sociological Review, 62, 883–902. Freeman, L. (1977). A set of measures of centrality based on betweenness. Sociometry, 40, 35–41. Gabbay, S. M., & Zuckerman, E. W. (1998). Social capital and opportunity in corporate R&D: The contingent effect of contact density on mobility expectations. Social Science Research, 27, 189–217. Galanter, M., & Palay, T. (1991). Tournament of lawyers. Chicago, IL: University of Chicago Press. de Graaf, N. D., & Flap, H. D. (1988). “With a little help from my friends”: Social resources as an explanation of occupational status and income in West Germany, The Netherlands, and the United States. Social Forces, 67, 452–472. Granovetter, M. (1973). Strength of weak ties. American Journal of Sociology, 78, 1360–1380. Granovetter, M. (1974). Getting a job. Cambridge, MA: Harvard University Press. Granovetter, M. (1985). Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91, 481–510. Granovetter, M. (1992). Problems of explanation in economic sociology. In: N. Nitin & R. G. Nohria (Eds), Networks and Organizations: Structure, Form, and Action (pp. 25–56). Boston, MA: Havard Business School Press.
Network Endorsement and Social Stratification in the Legal Profession
265
Greif, A. (1994). Cultural beliefs and the organization of society. Journal of Political Economy, 102, 912–950. Gulati, R., & Gargiulo, M. (1999). Where do interorganizational networks come from? American Journal of Sociology, 104, 1439–1493. Gulati, R., & Wang, L. O. (2003). Size of the pie and share of the pie: Implications of network embeddedness and business relatedness for value creation and value appropriation in joint ventures. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 209–242). Oxford: Elsevier Science. Hagan, J., Zatz, M., Arnold, B., & Kay, F. (1991). Cultural capital, gender, and the structural transformation of legal practice. Law & Society Review, 25, 239–262. Hannan, M. T., & Freeman, J. (1989). Organizational ecology. Cambridge, MA: Harvard University Press. Heinz, J. P., & Laumann, E. O. (1982). Chicago lawyers. New York: Russell Sage Foundation. Heinz, J. P., Laumann, E. O., Nelson, R. L., & Michelson, E. (1997). The changing character of lawyers’ work: Chicago in 1975 and 1995. Law & Society Review, 32, 751–776. Hull, K. E., & Nelson, R. L. (2000). Assimilation, choice or constraint? Testing three theories of gender differences in the careers of lawyers. Social Forces, 79, 229–264. Kay, F. (1997). Flight from law: A competing risks model of departures from law firms. Law & Society Review, 31, 301–335. Kollock, P. (1994). The emergence of exchange structures. American Journal of Sociology, 100, 313– 345. Lin, N. (1999). Social networks and status attainment. Annual Review of Sociology, 25, 467– 487. Lin, N. (2001). Social capital: A theory of social structure and action. New York: Cambridge University Press. Neckerman, K. M., & Fernandez, R. M. (2003). Keeping a job: Network hiring and turnover in a retail bank. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 299–318). Oxford: Elsevier Science. Nelson, R. L. (1988). Partners with power: The social transformation of the large law firm. Berkeley, CA: University of California Press. Podolny, J. M. (1993). A status-based model of market competition. American Journal of Sociology, 98, 829–872. Podolny, J. M. (1994). Market uncertainty and the social character of economic exchange. Administrative Science Quarterly, 39, 458–483. Podolny, J. M. (2001). Networks as the pipes and prisms of the market. American Journal of Sociology, 107, 33–60. Podolny, J. M., & Baron, J. (1997). Resources and relationships: Social networks and mobility in the workplace. American Sociological Review, 62, 673–693. Podolny, J. M., & Hsu, G. (2003). Quality, exchange, and Knightian uncertainty. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 77–103). Oxford: Elsevier Science. Podolny, J. M., & Page, K. L. (1998). Network forms of organization. Annual Review of Sociology, 24, 101–116. Podolny, J. M., & Phillips, D. J. (1996). The dynamics of organizational status. Industrial and Corporate Change, 5, 453–471. Podolny, J. M., & Stuart, T. E. (1995). A role-based ecology of technological change. American Journal of Sociology, 101, 1224–1260.
266
HARRIS H. KIM AND EDWARD O. LAUMANN
Podolny, J. M., Stuart, T. E., & Hannan, M. T. (1997). Networks, knowledge, and niches: Competition in the worldwide semiconductor industry, 1984–1991. American Journal of Sociology, 102, 659–689. Powell, W. W. (1990). Neither market nor hierarchy: Network forms of organization. Research in Organizational Behavior, 12, 295–336. Raub, W., & Weesie, J. (1990). Reputation and efficiency in social interactions: An example of network effects. American Journal of Sociology, 96, 626–654. Sandefur, R., Laumann, E. O., & Heinz, J. P (1999). The changing value of social capital in an expanding social system. In: Th. A. J. Leenders & S. M. Gabbay (Eds), Corporate Social Capital and Liability (pp. 217–236). Boston, MA: Kluwer. Stinchcombe, A. L. (1965). Social structure and organizations. In: J. G. March (Ed.), Handbook of Organizations (pp. 142–193). Chicago, IL: Rand McNally. Stoloff, J. A., Glanville, J. L., & Bienenstock, E. J. (1999). Women’s participation in the labor force: The role of social networks. Social Networks, 21, 91–108. Stuart, T. E. (1999). Interorganizational alliances and the performance of firms. Administrative Science Quarterly, 44, 315–349. Stuart, T. E. (2003). Governing strategic alliances. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 189–208). Oxford: Elsevier Science. Stuart, T. E., Hoang, H., & Hybels, R. (1999). Interorganizational endorsements and the performance of entrepreneurial ventures. Administrative Science Quarterly, 44, 315–349. Uzzi, B. (1996). The sources and consequences of embeddedness for the economic performance of organizations: The network effect. American Sociological Review, 61, 674–698. Uzzi, B. (1999). Social relations and networks in the making of financial capital. American Sociological Review, 64, 481–505. Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. New York: Cambridge University Press. Wegener, B. (1991). Job mobility and social ties: Social resources, prior job, and status attainment. American Sociological Review, 56, 60–71. Williamson, O. E. (1981). The economics of organization: The transaction cost approach. American Journal of Sociology, 87, 548–577. Williamson, O. E. (1985). The economic institutions of capitalism. New York: Free Press. Williamson, O. E. (1991). Comparative economic organization: The analysis of discrete structural alternatives. Administrative Science Quarterly, 36, 269–296.
INTERLOCKING JUDGES: ON JOINT EXOGENOUS AND SELF-GOVERNANCE OF MARKETS Emmanuel Lazega and Lise Mounier ABSTRACT This study stresses the importance of considering a “joint” governance of interfirm relations as an alternative to external governance (by the State) and self-governance (by the business community) of these relations. We argue that a broadly-conceived structural and organizational approach to economic institutions provides insights into this joint governance because it shows how such a system spreads the costs of control among several kinds of stakeholders. We look at how transactions between any two firms are regulated through jurisdiction by “consular” judges (i.e. judges elected through the local Chamber of Commerce) who indirectly represent other firms and industries in that market, and are therefore considered to be at the same time third parties and potential levers of influence acting on behalf of corporate interests. We study an empirical case of such joint governance: The Tribunal of Commerce of Paris (TCP). Following previous work on lateral control and leverage, we hypothesize that industries and/or companies that have a strong stake in the conflict resolution process will be more represented among the judges of this court than other industries and/or companies, and that judges who are socially active in the court that enforces this joint governance will be sought out for advice more than other judges, and thus gain influence on their peers by suggesting specific outcomes. The analyses of the composition The Governance of Relations in Markets and Organizations Research in the Sociology of Organizations, Volume 20, 267–296 Copyright © 2003 by Elsevier Science Ltd. All rights of reproduction in any form reserved ISSN: 0733-558X/PII: S0733558X02200105
267
268
EMMANUEL LAZEGA AND LISE MOUNIER
of the bench and of the advice network data collected in this court display an influence structure that confirms these hypotheses and that is likely to affect conflict resolution between businesses. It thus characterizes joint governance of markets as a complex set of social processes worthy of economic sociologists’ attention.
INTRODUCTION Businesses usually try as much as they can to participate in the governance of their markets. They try to shape their opportunity structure, to structure their environment, and to uphold the social mechanisms allowing them to cooperate. At the interorganizational level, at least two different sociological traditions deal with the issue of governance of markets, stressing either the formal and often exogenous aspects of this governance, or the informal and often endogenous character of self-governance. For the first tradition, a socio-legal one, exogenous governance or “regulation” (Ayres & Braithwaite, 1992; Hawkins, 1984; Hawkins & Thomas, 1984; Shapiro, 1984; Weaver, 1977) is provided by government agencies backed up by courts. These studies focus, for example, on the decision by government agencies to prosecute deviant companies. Such decisions are not obvious and they often come out of tradeoffs between official inspectors and company managers. This is especially the case when they face risks such as large-scale losses and layoffs, and sometimes bankruptcy, should the law be strictly enforced. The second tradition focuses on interfirm arrangements promoting self-governance benefits for firms in their interorganizational transactions and more informal conflict resolution mechanisms.1 Precisely because litigation is costly, firms prefer informal dispute resolution whenever possible, especially when they have long term continuing relationships (Macaulay, 1963; Raub & Weesie, 2000). Here the focus is on pressures to conform by one organization on the other. Pressures are based on resource dependencies and reputation (Raub & Weesie, 1993; see also Batenburg et al., 2003 as well as Voss, 2003 in this volume). Thus, each tradition focuses on a different kind of actor intervening in governance and incurring the largest share of costs of control: mainly the State and/or companies themselves – the latter sometimes through industry representatives or through selection of partners (Blumberg, 1997; see also Buskens et al., 2003 in this volume). In reality, the two governance, or conflict resolution, systems combine in various ways. One example of combination is provided by Ayres and Braithwaite in their analysis of “responsive self-regulation.” This analysis shows the existence of “enforcement pyramids” that exist between State regulatory agencies and corporate actors. Such pyramids express the possibility, for industry representatives
Interlocking Judges
269
and for law enforcement, to escalate from persuasion to warning letters to civil penalties to criminal penalties to license suspension and revocation. Actors know that such enforcement pyramids exist. They know that each way of enforcing contracts is only one way among several others, and that an escalation can be triggered. This is why, in spite of the costs of litigation, firms do use formal institutional litigation both as plaintiffs and as defendant (Cheit & Gersen, 2000; Dunworth & Rogers, 1996; Galanter & Epp, 1992) and conflicts do follow the disputing pyramid transforming informal complaints into court filings and formal judiciary decisions (Felstiner et al., 1980). Following the idea that the two forms of conflict resolution are connected, we further explore this connection. We think that it is possible to identify a form of “joint” governance or a combined regime of endogenous and exogenous conflict resolution in markets. We use the label “joint” because we argue that the governance mechanism is a combination of self-regulation and exogenous regulation, a combination in which costs of control are shared. The combinations of the two can take many hybrid forms. We can define the joint element in “joint governance” as the coexistence of several sources of constraint, both external and internal, that weigh on the actors in charge of solving the conflicts and enforcing the rules. These actors can be identified as third parties when they sit in judgment and make decisions about how to solve conflicts related to market activities between a first and a second party. Saying that such third parties are constrained by external and internal forces is equivalent to saying that they have to deal with influences coming from various stakeholders in the conflict resolution process, especially from stakeholders who share the costs of control. Thinking about joint governance in those terms follows both an organizational and a broadly conceived structural approach to economic institutions (Lazega & Mounier, 2002).2 Here we use these organizational and structural perspectives to build on this approach and to contribute to the study of economic and legal institutions that represent a combination of exogenous and self-governance of markets. Since the focus is on judges as third parties, courts are natural places to examine in order to identify joint forms of governance. Courts are indeed a locus of joint governance. They are not static institutions making a-temporal and purely rational decisions (Heydebrand & Seron, 1990; Wheeler et al., 1988). They are contested terrain, the prizes or objects of broader economic competition and conflicts that occur outside courthouses (Flemming, 1998). This is especially the case in courts in which judges are themselves business people elected by their local business community. In this chapter, we use such a case to focus on how organized interests try to shape the forum in which disputes between businesses are processed. In such “consular” commercial courts, we look at judges as both official third parties upholding legal rules and procedures in the conflict resolution process, and as unofficial and
270
EMMANUEL LAZEGA AND LISE MOUNIER
potential levers of influence representing their industry of origin, thus possibly favoring outcomes that do not hurt the interests of this industry. As levers, they could be in a position to weigh directly on judicial decisions, or indirectly if they can influence other judges sitting with them and gain their consent for specific outcomes. We use the fact that courts are usually organized in ways that stress collegial procedures among judges: for example three or five judges sit together in judgment of ongoing flows of cases, deliberating as equals and making judicial decisions. Previous structural work on leverage and allocation of costs of control in a collegial organization has shown that levers of influence are expected to be efficient if they concentrate large amounts of stakeholder expectations and investments on themselves, and if they are socially close to their targets.3 Convergence of stakeholder expectations on specific members creates a specific form of status that is particularly efficient in indirect control and conflict resolution. Social closeness increases access to the target and the lever’s chances of being listened to by the target, of gaining his/her consent for specific outcomes. Extending this previous work, our hypothesis is that joint governance is embedded in the formal structure and the social life of the organization that enforces it. We argue that the strongest the external stakeholders’ interests and expectations, the more they invest in shaping the courthouse, especially in selecting the judges. We also argue that the closer the judges (as levers of influence) are to other judges (as targets of influence) in the same court, the more influence they are likely to have on these other judges (i.e. on their targets). In other words, we hypothesize that industries and/or companies that have a strong stake in the conflict resolution process will be more represented among the judges than other industries and/or companies, and that judges who are socially active in the court will be sought out for advice more than other judges, and thus gain influence on their peers by suggesting specific outcomes. This structural approach to governance is applied to a case study, that of a consular commercial court, i.e. a court in which judges are elected by the local Chamber of Commerce and to which the State has delegated conflict resolution among businesses. To test these hypotheses, we use data on how the business community of Paris participates in operating its commercial court. In France, commercial courts are the main formal conflict resolution mechanism made available by French law to business people. We argue that using this organizational and structural approach may be important to understand, ultimately, who wins, what and when.4
JUDGES AS THIRD PARTIES AND POTENTIAL LEVERS With this organizational and structural approach to joint governance, it is possible to look at commercial courts as representing combined regulation. For example, when
Interlocking Judges
271
judges are elected from among local business people (“consular” judges), they can be thought of as third parties representing the State as well as representatives of their business community. They may not accept this idea of being “representatives” with a mandate from the industry that helped them become judges. But they are still expected to speak on behalf of this industry and its customs by the members of this industry. In a joint or combined regime, the parties involved in the conflict attempt to influence judicial decision making by influencing this third party. This is what occurs normally when judges listen to the arguments of lawyers representing each (first and second) party. But here judges represent not only the power and interests of the State but may sometimes represent that of other stakeholders sharing the cost of justice, including industries or companies, and may thus find themselves closer to either the first or the second party. Attempts at influencing what goes on in court from outside the court come through various angles. Flemming (1998) lists five such angles: external stakeholders can try to influence jurisdiction (the range of disputes over which the court has authority), positions (actors formally authorized to participate in the disposition of cases), resources (the capacity to influence the decisions of other actors), discretion (the range of choices available to actors), and procedures (rules governing courtroom processes). Parties involved in that contest may not be directly concerned by all the conflicts that are dealt with by the court, but they may have indirect concerns, material or symbolic, in the decisions of the court, and thus attempt to influence what goes on. Flemming’s categorizations are useful at this early stage of research. They help focusing on specific processes of influence, including ways of putting constraints on judges as third parties. In this chapter, we limit ourselves to the study of two such processes: position and resources (in Flemming’s vocabulary) and the relationship between the forms of influence they represent. By this we mean that we look at who is allowed to become a third party (or judge) and what kind of resources are made available to such third parties when sitting in judgment and participating in governance while solving conflicts between businesses. Looking at judges as constrained third parties will help in understanding the relationship between the two forms of influence. Indeed, collective actors involved in conflicts on markets, such as companies, whole industries (in class actions, for example) and even State administrations,5 may have strong incentives to influence who becomes a judge and what resources are available to these judges. For example, the more litigious these sectors are, the stronger their incentives to share the costs of conflict resolution. These collective actors are usually conceived of as external actors. Their incentives to influence the jurisdiction of the court are to protect their interests in the long run. They may do this by helping some of their own selected members in becoming a judge. The
272
EMMANUEL LAZEGA AND LISE MOUNIER
stronger their incentives, the more they can be expected to take care of their representation among the judges. Such judges, once in a position to solve conflicts between first and second parties, may have combined incentives to influence the jurisdiction of the court: on the one hand, they are supposed to represent the law and to proceed in all independence; on the other hand, they may represent, and therefore protect the interests of the organizations that supported their becoming a judge. Influencing who becomes a judge and what resources are available to these judges can be a very strong, although indirect, way of influencing the outcome of judicial decision making. In effect, collegial deliberation among judges is built into the jurisdictional institution. This deliberation relies heavily on knowledge management by the court. This brings us to the second process (in Flemming’s list above) by which influence characterizing joint governance is exercised in such organizations. One way to influence the behavior of judges is to try to set the premises of their decisions by attempting to control the information available to them while sitting in judgment. Knowledge management that helps the judges in their work relies on formal and informal quality control of decision making among judges. Formal quality control is organized by the procedural rules of adversarial building of the case, but also by collegiality of the committee of the judges sitting in judgment, and ultimately by resort to the President of the Chamber when there is a doubt. Informal quality control happens in the informal knowledge management, itself based on micro-politics of knowledge (Lazega, 1992). Judges coming from one sector of the economy may thus, as “judicial entrepreneurs” (McIntosh & Cates, 1997), attempt to keep particular definitions of problems alive, or promote the ideas, customs, rules, and interests that are commonplace in their sector, although not in others. Control over the premises of decisions can be assumed to influence the probability of who wins even if this framing control by players is almost invisible to outside observers. The law and the courts are aware of the fact that different types of actors in the environment of the court are involved in such influence attempts. Anticipating that the court will be the object of external attempts at influence, the legal system helps the judges in protecting themselves from such influences. It can do so, first, either by providing them with tenure (thus lowering their level of dependence upon these external sources of influence) or, if judges are elected, by trying to diversify the sources of influence so that they cancel each other out (Friesen et al., 1971). Second, if differences in access to resources exist between judges, the legal system tries to protect the court from the effects of such differences by providing each judge with the resources that he or she needs to make decisions. Example of such protections include a procedurally well defined filing system, or institutionalized collegiality
Interlocking Judges
273
of deliberation that helps judges in confronting their different definitions of the situation and reach a common frame of reference. Third, there are rules concerning conflicts of interests for judges: when they are too close to one of the parties, for example, when they sit in judgment of a competitor, sometimes even of a business in the same industry as their industry of origin, they must self-disqualify or, if this is discovered, be removed from the case by their hierarchy. However, a structural approach to joint governance raises the issue of the success of such procedural attempts in neutralizing external influences, especially when the judges are elected. Indeed, and to summarize, given the incentives identified above, we can expect influence on judges (as third parties) to take a form identified in the following hypotheses. First, the stronger the external stakeholders’ interests and expectations, the more likely they are to invest in shaping the courthouse, especially in selecting the judges. Second, the closer the judges (as levers of influence) are to other judges (as targets of influence) in the same court, the more influence they have on these other judges (i.e. on their targets). In other words, being active in the social life of the courthouse multiplies opportunities to meet, discuss decisions, and therefore of being sought out for advice, thus increasing one’s influence on other judges as targets. In this chapter we test these hypotheses using a case study of the Commercial Court of Paris. We do not have access to all the data that would be necessary for testing the efficiency of the influence structure that we expect. But we do have enough data to test, at least in part, the existence and shape of this influence structure.
FRENCH COMMERCIAL COURTS AS JOINT GOVERNANCE SYSTEMS SPREADING COSTS OF CONTROL In France, the role of commercial courts is to solve conflicts between economic actors, mainly businesses, and to exercise a form of discipline on market entry, exchanges, and exit. Institutional solutions found for such problems vary within and between countries. Data on French commercial courts are particularly useful to test our hypotheses because a specific form of combined governance of markets has existed in France for five centuries.6 The State has long been sharing its own judiciary power with the local business community. In effect, a special jurisdiction exists for commerce, and these commercial courts are operated by consular judges, the third parties in our vocabulary. Each judge acts both as an individual judge, and as a representative (but without any specific mandate) of the business community. They are unpaid judges elected for two or four years (for a maximum
274
EMMANUEL LAZEGA AND LISE MOUNIER
total of fourteen years) by the members of the Chamber of Commerce of their local jurisdiction. Today, judges at the local Tribunal de Commerce have exactly the same constituency as the local Chamber of Commerce. The two economic institutions support each other and maintain close ties. In this institutional design, the State, the industries or companies, and the individual judges share the costs of control. These judges sit in judgment one day a week, on a voluntary basis, to enforce the law among their peers, both in matters of commercial litigation (between consumers and businesses, or between businesses) and of bankruptcies. This court is thus both a formal legal institution and a social mechanism in its enforcement of the law and sanctioning of deviant behavior in the business community.7 Decisions made by the court can be challenged, as in any court, and be brought to the Court of Appeal. At the Court of Appeal, judges are no longer business people, but highly selected professional judges coming out of the very official Ecole Nationale de la Magistrature. A very small percentage of cases (around 5%), comparable to that of other jurisdictions, is challenged by the parties in French commercial courts. Thus this system represents a particular and efficient combination of external and self-governance of local business communities. In such a system, there are at least two broad types of such unpaid, voluntary, and consular judges: first, retired business people looking for status, an interesting activity and social integration;8 and second, younger professionals, whether bankers, lawyers, or consultants, who look for experience, status, and social contacts, sometimes for information on behalf of their employer (who keeps paying their salary one day a week while they are practicing as a judge at the Tribunal). If the individual judge is young enough, this can help build a relational capital (as explicitly stated in the flyers trying to attract new judges to the job) and open doors for future positions in economic institutions such as the Chambre de Commerce itself, arbitration courts, the Conseil Economique et Social (an advisory board to the Prime Minister), and other honors dispensed by the State apparatus. Indeed, for younger professionals, sitting as a judge at the TCP has traditionally been considered a “chore” that would be rewarded later on with seats on prestigious committees in economic institutions of the country (Lemercier, 2001). Various types of lucrative contracts and missions may also be awarded, on a discretionary basis, by the acting President of the Tribunal de Commerce to former judges to advise companies on a “prevention” basis. There are several justifications for this joint regime. First, it is a cheaper and faster form of justice than a system with professional judges. Business bears more of the costs of its own regulation, and backlogs and waiting time are much smaller than in traditional High Courts. For example, there is no jurisprudence. Second, inexperienced professional judges – who are civil servants – have been considered
Interlocking Judges
275
notoriously unable to understand the problems of companies and to monitor satisfactorily the behavior of company directors, particularly in the insolvency and bankruptcy mine fields (Carruthers & Halliday, 1998, p. 431). Third, business law often ignores idiosyncratic norms and customs (called usages in French commercial courts) based on industry traditional subcultures characterizing whole sectors. The argument is that efficient conflict resolution cannot ignore these bodies of rules and conventions (Favereau, 1994) that organize business practice differently in each traditional sector. Judges at the Tribunal de Commerce, since they are supposed to be experienced business people, are thus said to be specialized in their professional field and in a better position than tenured civil servants to know about these customs; to adjust them more quickly to unstable or changing business environments; and to be in a better position to foster regulatory innovations. Thus the French Tribunal de Commerce has specific features that make some aspects of the link between legal (exogenous) and social (endogenous) mechanisms in the governance of business more visible. The State alone does not, by itself, enforce and sanction. It needs the participation and investments by specific individual and corporate actors prepared to do their share and bear some of the costs of control. In effect, in this case, elected representatives (who may represent corporatist interests) perform a function usually considered to be a State function. This is a rather special case of joint governance (or “co-regulation”), understood as industry-associations self-regulation with some oversight and/or ratification by the State (Grabosky & Braithwaite, 1986, p. 83). In this system, the difficulty with representing both general and particularistic interests has always been the predicament of these judges.9 The public has always suspected that patronage appointments lead to politicized elections of judges, who then fail to distance themselves from their virtual “constituency,” i.e. the industry that endorsed their candidacy for the job. Especially in small town commercial courts, litigants’ confidence in the impartiality of the tribunal’s decision is often impaired. They fear that judicial control can be exercised by competitors. The institution, however, assumes that zealous judges will be entirely virtuous in spite of the short distance between regulator and regulatee (Black, 1984).10 In general, the business community wants speed and decisiveness from modern commercial courts, a low level of appeals, sharp segregation of politics and personal patronage from judicial decision making, and as much neutrality as possible. But with consular courts, entire sectors of this business community, as represented by their syndicats patronaux, may think that they sometimes have the possibility of preventing damage to the interests of their industry. Indeed, the syndicat’s interest is to be represented at the court for several reasons. First, this is a way of defending the customs of their occupation or profession. For example, consistent with the interests identified above, the financial industry sees arrangements for corporate
276
EMMANUEL LAZEGA AND LISE MOUNIER
liquidation or administration as very important to its practice of commercial lending. Bankers could handle their affairs concerning corporate rehabilitation outside of the courts if they perceive these courts to be incompetent or opposed to their interests. The courts must often conform to the expectations of the financial industry or lose much of their business (Carruthers & Halliday, 1998, p. 488). The “quality of justice” as defined by the financial industry (limit risk and permit failures, or extend credit and aid reconstruction) is a significant factor in the strategy of this industry. Second, it signals to the constituency that the leaders of their syndicat are working hard at promoting the interests of the profession. Increases in the number of judges sent to the commercial court makes the apparatchiks look good. Thus, from the perspective of each industry, consular judges are more than simple judges. They are judicial entrepreneurs representing the sensitivity of the syndicats patronaux and organized interests that helped them into the courthouse in the first place by endorsing their candidacy. This is true in spite of the fact that judges are now managers in medium sized or large companies. They rarely own many shares in their employer. Large banks sometimes control their group. They are no longer the baker, or the jeweler who owned a shop around the corner. They are entrepreneurs because they should identify problems and push for solutions that make sense in their own business community.11 As seen above, having judges of one’s sector sitting in judgment may represent minimally a guarantee that one’s way of doing business will be respected, if not a leverage or a damage control instrument. The judges themselves do not entirely share this view. Officially, in our interviews, they declare that – once a judge – they are independent, without any mandate from their industry of origin, and think of themselves as entirely impartial. Nevertheless, the system itself represents a typical joint governance regime in which institutions, the business community, and individual citizens accept to share the costs of control for the reasons outlined above.
FIELDWORK AT THE TRIBUNAL DE COMMERCE DE PARIS (TCP) Fieldwork was conducted at the TCP. This court is one of the four large commercial courts in the Paris region (along with the courts of Nanterre, Bobigny and Cr´eteil), where around 220,000 businesses are officially registered. The court includes twenty generalist and specialized chambers (such as bankruptcy, unfair competition, company law, European community law, international law, multimedia and new technologies, etc.) which handle around 12% of all the commercial litigation in France, including large and complex cases (that do not go to arbitration courts). The list of judges included 157 names, that of the 147 judges that were
Interlocking Judges
277
active at the end of 2000, plus 10 “wisemen,” i.e. former judges who asked to remain in the tribunal as advisers after they decided to resign as judges; they are thus former judges to whom sitting judges can turn for advice in complex cases. Sociodemographic characteristics of the judges show that 87% of the judges are men. Average age is 59 (minimum 36, maximum 78, standard deviation eight years). On average, women judges are slightly younger than their male colleagues (53 versus 60). 52% of the judges have been with the Tribunal for ten years or more; 36% were elected between 1991 and 1995; 12% since 1996. 38% are retired. Positions occupied by judges in their industry and occupation of origin (or former occupation) include CEOs (25%), vice-presidents, and top executives of all kinds. Among the youngest judges, there are more professionals such as company counsels, accountants, and consultants. Most of the time, they work for large business groups12 or medium-sized companies (that the judges do not like to name); they prefer to be discreet about their professional affiliation. Most judges have diplomas from top French business schools (HEC, ESSEC, ESC, INSEAD, IAE, etc.), law schools, elite engineering schools (specialized or generalist). Fieldwork was carried out in two phases. A first phase during which we collected ethnographic information and observations on the operations of the court. We attended court hearings and even a couple of normally closed deliberations about bankruptcy cases. We also interviewed representatives of all the professions operating such courts (mainly judges, but also lawyers, representatives of the attorney general, liquidators, bailiffs, etc.). Several judges and “wisemen” accepted in depth interviews. The ethnography of this court included looking at the ties between the TCP and economic institutions of the city: there is, for example, an annual common golf tournament between the TCP and the Chambre de Commerce de Paris. And Hautes Etudes Commerciales (HEC), the top French business school, in which a sizable proportion of judges in large commercial courts were trained, is also managed by the Chambre de Commerce de Paris. A second phase allowed us to interview all the judges about various issues of interest to the presidency of the court, and to include a name generator about advice seeking in a twenty-minute questionnaire. This phase lasted three weeks, with the President’s team helping with scheduling the interviews with the judges. A letter signed by the President of the Court helped in persuading the judges to answer our questions. Response rate reached 97%. We were well received by the judges who were interested in talking about their voluntary participation in running a State institution. This in spite of, or because of, a controversial reform of commercial courts that was discussed in Parliament at the time (Jean, 2000). Almost all the active judges were thus interviewed. Additional research allowed us to reconstitute the professional biography of 137 of these active judges.
278
EMMANUEL LAZEGA AND LISE MOUNIER
The organizational functioning of the TCP is complex. It is not our purpose to describe it here in more detail. As mentioned above, several kinds of professionals operate these courts together: consular judges, clerks, business lawyers, representatives of the attorney general, bailiffs, experts of all kind, professional liquidators and/or administrators (for companies that are on the brink of bankruptcy but could perhaps be brought back to life). Judges are allocated across the large number of generalist and specialized chambers. The minimal distinction in terms of specialties is between bankruptcy and litigation. Different procedural rules are legally attached to each. But the litigation bench is then subdivided in many types of subspecialties mentioned above. There is a President of each Chamber who reports to the President of the Tribunal. In each chamber, three – sometimes five – judges sit together in judgment, processing cases collegially while listening to the conflicting parties, a series of clerks, lawyers, and other professionals, as in any other court.
ANALYSES Over-Representation of the Financial Industry among the Judges Recall that, according to the justification of this system of joint governance, the selection of judges should produce a representation including as many sectors as possible, especially in large commercial courts such as that of Paris. At the time of the study, economic sectors represented by the judges (i.e. in which they work or in which they used to work) were indeed very diverse. Thus, in complex cases, intelligence about one specialty could be made available to the court from the judges coming from that specialty. We hypothesized, however, that some industries or companies would invest more than others in judicial entrepreneurship and incur a larger share of the costs of control because they have an interest in doing so. Analyses show that this hypothesis is confirmed. Although in theory all the syndicats patronaux can present candidates to the elections of consular judges on an annual basis to fill in the vacancies created by a 10% turnover at the court, in reality all did not. Some did so much more systematically than others. Twenty-nine percent of the judges came from the financial industry.13 Forty-four consular judges were currently employed by the financial industry or had been employed by it in the past. This industry promotes several candidates for election to the job each year.14 The financial industry is clearly over-represented, in relative terms, at the TCP. In effect, the sector of financial activities represents 3% of the active population in France15 and 5.1% in Paris where the services industries are over-represented compared to the rest of France.16 In terms of value added to the economy per branch (chained prices for previous year, 1995 basis), the share of the financial
Interlocking Judges
279
services industry in the total value added to the French economy was 5.3%.17 These descriptive results show that judges coming from the financial industry are clearly potential levers of that industry and that influence over other judges (targets) in the court would mean that the financial industry is the biggest threat to this court’s independence. This industry is traditionally a very litigious sector (Cheit & Gersen, 2000). The business docket in France, as probably in most countries, is dominated by contract disputes and debt collection issues. A sizable portion of this docket involves the financial industry for obvious reasons, which provides a strong incentive to invest in judicial entrepreneurship – for example to ensure damage control in cases involving high levels of credit. An example can be provided to illustrate the interests of this industry in incurring a larger than average share of the costs of control. In 1985, French bankruptcy law was modified by France’s Socialist government in order to give priority to employees’ interests over that of creditors. The new law required judges to make a decision about the possible survival of the company, were it to be better managed (Gu´eroult et al., 1993). If they decided that the company – and its jobs – could be saved, they would name an administrator to run it and bring it back to life within a given time frame. If they decided that the company could not be saved, they would order it to be liquidated. When judges choose an administrator over a liquidator, creditors are in danger of losing even more money than if the company were simply liquidated. Bank and financial institutions are often creditors themselves and they stood to lose enormous amounts when the new law was passed. For nine years, the French financial sector tried to lobby for a change in the law. In 1994, a failed attempt at such legal changes pushed the sector to change its strategy, invest in penetrating the courts, and increase the number of consular judges coming from its ranks. Thus the financial industry has high amounts of resources at stake in commercial litigation and bankruptcy. It is willing to play for the rules and has an interest in trying to shape the court and impose its industry norms and practices over those of other industries. The priorities of the financial sector (such as preserving high value of assets, high level of claims, and high sensitivity to the impact of corporate failures on the economy) can thus be defended in both the litigation and the bankruptcy bench. One of the likely influence processes that characterize joint governance is thus detected in the selection of judges themselves (i.e. the “positional” effect in Flemming’s vocabulary). In spite of the French business community’s attempts at diversifying the origins of judges through a complex election procedure, only a few industries actually invest in judicial entrepreneurship. Why do others neglect it – in spite of their incentives to exercise influence on judges sitting as third parties and potentially used as levers? The answer may be that large companies
280
EMMANUEL LAZEGA AND LISE MOUNIER
go to arbitrage (no publicity) and perhaps no longer care about public commercial courts. Or perhaps that small syndicats do not have the necessary clout to lobby effectively, nor enough resources to share the costs of control. All sectors of the business community cannot participate equally actively in the contests and attempts to shape the courthouse from outside. The potential influence of each in the struggle over this kind of contested terrain varies with the resources available to promote candidates for the jobs of consular judge. Resources available are not the same in the banking industry and in less well organized sectors, such as retail.
Advice Networks and Selective Premise Setting Using this information about “positional” influence, we can now test our second hypothesis by looking at the extent to which (a) judges socially close to their targets have more influence than others, i.e. control resources necessary to other judges’ work; and (b) judges coming from this over-represented financial industry are in a position to exercise this influence. We examine this form of control of resources by looking at centrality in the advice network among all the judges of the court. Indeed, an indirect way of looking at this kind of influence is to focus on advice interactions between judges, which we assume to be equivalent to interactions setting the premises of judicial decisions. The advice network among judges, and its premise-setting efficiency, are considered to be a bridge between structure and decision making (Lazega, 1992). In effect, patterns of advice seeking in the court show who is prepared to listen to whom when framing and defining problems at hand in the judicial decision making process. Therefore, we look at the ways in which these judges transfer and exchange advice, then try to appreciate the ways in which contextual factors can influence this process. We can measure the capacity of an industry to set the premises of such decisions by looking at the centrality of its representatives in the advice network among all the judges, and then at the determinants of this centrality. Data on advice seeking among judges were collected using the following name generator: “Here is the list of all your colleagues at this Tribunal, including the President and Vice-Presidents of the Tribunal, the Presidents of the Chambers, the judges, and ‘wisemen.’ Using this list, could you check the names of colleagues whom you have asked for advice during the last two years concerning a complex case, or with whom you have had basic discussions, outside formal deliberations, in order to get a different point of view on this case.” Our high response rate allowed us to reconstitute the complete advice network (outside formal deliberations) among judges at this courthouse, and thus to measure each judge’s centrality in this network.18
Interlocking Judges
281
Given that only 10% of all choices in this network are reciprocal, the existence of an informal pecking order and an informal hierarchy among judges is very likely. Table 1 presents simple correlations between the judges’ centrality scores and other attributes showing that being socially active in the organization and coming from the financial industry are clearly not correlated to high centrality in the advice network. Thus, apparently, our hypothesis is not confirmed. However, in order to control for other possible effects on centrality, we included these attributes along with several other characteristics of the judges in a regression model predicting centrality in the advice network among judges. In addition to two main variables representing the judges sector of origin (having worked in the financial industry, having worked in a large business group), a series of control variables were added to the model. Seniority, measured by the number of years an individual has served as a judge, can be understood as “experience” and help a judge wield influence independent of the sector of origin. Specialty could also have an effect on premise setting; for example, secrecy is usually more strongly required from judges sitting on the bankruptcy bench. In addition, the other judges in the commercial court may not be the single source of advice and influence. Being well connected and open to the business community can attract colleagues who need economic advice. The same is true for being well connected and open to professional judges in other courts (High Courts, Court of Appeal). Such external ties can attract colleagues who need legal advice. The same is not true with being well connected and open to the office of the attorney general; surveillance and influence by the Ministry are not always welcome in consular courts. Being in activity (as opposed to retired) may also have an effect on centrality in the advice network since retired judges may have more time and be more available than professionally active ones to discuss issues at length. Being member of the State elite (the “noblesse d’Etat,” i.e. coming from the elite French engineering and administration grandes e´ coles, having been trained at ENA or Polytechnique) means that one has connections in high places and thus might wield some authority and influence among fellow consular judges. Table 2 presents the analysis controlling for these effects.19 Results of this analysis provide a picture different from that of the simple correlation table. They show that our hypothesis is confirmed. Controlling for the other variables, being active in the social life of the court does have a robust effect (which remains significant in most models) on centrality in the advice network, and thus on the capacity to set the premises of other judges’ (targets) decisions. The sector of origin of a judge, particularly coming from the banking industry, has no effect on being central. Bankers may be over-represented at this court, but they do not exercise strong indirect influence through premise setting in this organization. In effect, in order to exercise such indirect influence, judges should
282
Table 1. Correlations between Independent Variables Characterizing the Judges of the Tribunal de Commerce de Paris in Table 2.
Pearson correlation coefficients, N = 146.
2
3
4
5
6
7
8
9
10
11
– 0.49 0.34 −0.02 0.06 −0.08 0.15 0.20 0.02 −0.18 0.16 0.08
– 0.50 0.13 0.09 −0.18 0.00 −0.03 0.28 −0.27 −0.00 −0.13
– −0.02 −0.03 −0.17 −0.00 −0.08 0.01 −0.12 −0.11 −0.09
– 0.03 −0.08 −0.00 −0.09 0.22 0.05 −0.13 −0.04
– 0.01 −0.02 −0.11 0.09 −0.17 0.08 −0.03
– −0.06 0.01 0.00 −0.06 0.09 −0.02
– 0.08 0.18 0.18 0.03 0.16
– 0.05 −0.02 0.07 0.07
– 0.15 −0.06 0.20
– −0.05 0.33
– 0.10
EMMANUEL LAZEGA AND LISE MOUNIER
1. Centrality in advice network 2. Number of years as a judge 3. President of Chamber 4. Bankruptcy (vs. litigation) 5. Coming from financial sector 6. Coming from non-financial large groups 7. Seeking advice from business community 8. Seeking advice from professional judges 9. Seeking advice from attorney general 10. Professionally active (vs. retired) 11. “Noblesse d’Etat” 12. Active in intra-TCP social life
1
Interlocking Judges
283
Table 2. Who Sets Whose Premises? Linear Regression Model Measuring the Effect of the Judges’ Characteristics on Their Centrality in the Advice Network. Variables Intercept Number of years as a judge Specialty (Bankruptcy vs. litigation) Past employment in financial sector Past employment in large business groups Seeks advice in business community Seeks advice from professional judges Seeks advice from attorney general’s office Currently employed (vs. retired) Member of the noblesse d’Etat (ENA, X) Active in the social life of the court
Parameter Estimate
Standardized Estimate
−3.24 (1.16) 0.67 (0.09) −0.17 (0.78) 0.40 (0.69) 0.11 (0.59) 1.47 (0.64) 4.19 (1.47) −1.64 (0.68) −0.66 (0.66) 1.70 (1.11) 2.29 (0.96)
0.00 0.55 −0.01 0.04 0.01 0.15 0.19 −0.18 −0.07 0.10 0.17
Standard deviations in parenthesis; N = 145; R 2 = 0.38.
also be strongly integrated in the courthouse, in its social life, have and use ties outside the courthouse, consult with professional judges. We know from Table 1 that they do not do so more than other judges, which explains in part their lack of potential influence. Notice that, in addition to being socially active in the court, being senior and seeking advice from other sources (in the business community and among professional judges) are also good predictors of potential influence in this court. Notoriety of consular judges can be built inside the small microcosm of the courthouse by investments in ties to other judges whether within or outside this specific courthouse. However, seeking advice from the attorney general (who represents the State directly inside the TCP) is significant and negative: the more contact judges have with the attorney general and its representatives, the less these judges are sought out for advice by their peers. In sum, the more socially active the judges within the court, the more open to discussions with the business community and with the legal environment – but the less open to discussions with representatives of the State – the more influential these judges are at the TCP.
“Les Commer¸cants D´etestent la Banque” The bankers’ lack of detectable potential influence may thus be explained by at least two factors. First, seniority measured by the number of years spent as a judge is the strongest determinant of centrality in the advice network. This is partly correlated with being a Pr´esident de Chambre, which explains the strength of this effect. Most senior judges, i.e. often also retired ones, tend not to be from the financial sector.
284
EMMANUEL LAZEGA AND LISE MOUNIER
Professionally active judges, often bankers, may have less seniority at the court, less time to interact with other judges, socialize with them, and thus less centrality as advisors. But, second, the social integration of bankers at the court is relatively weaker than that of judges coming from other industries precisely because the banking industry has so many representatives on the bench. Being perceived to be so strong in terms of “position” may be counterproductive in terms of influence. The banking industry is defensively perceived either as defending its corporatist interests – not the general interest – or as trying to substitute for the State, once the latter has withdrawn from direct control of the economy. This was illustrated to us by a judge (No. 128) who is himself a former banker: “Les commer¸cants d´etestent la banque” (shopkeepers hate bankers). Another indicator is that judges coming from the financial sector are almost systematically those elected with the smallest number of votes.
Bankers Do Not Seem to Coordinate with One Another One reason that may account for the lack of positive relationship between Flemming’s two forms of influence (the more control of “position,” the more control of “resources”) may be that, at the TCP, the judges coming from the financial sector did not coordinate their activities. First, homophilous choices among these judges are not more frequent than choices of advisors among colleagues from other specialties. Second, the set of judges coming from the financial industry does not seem to be very organized, but rather segmented and dispersed. This shows in the picture representing a reduction of the advice network at the TCP. An analysis of the same advice network approximating structural equivalence among judges provides a simplified overview of the system. Figure 1 presents this reduced network. The density table from which it is derived is presented in the appendix. The network breaks down into six positions of approximately structurally equivalent judges in this Tribunal. Four of these (positions One, Two, Three and Five) include relatively small proportions of judges coming from the financial sector. The last two (positions Four and Six) include larger proportions of such judges. We therefore examine these two sets of positions separately. To simplify, position One was made up of 20 judges, 25% of whom came from the financial sector; the others belonged to large, non-financial business groups, were recently elected although not necessarily young, and did not participate in the social life of the courthouse. Members of this position did not seek advice from anybody in the other positions and were not sought out. Position Two included 25 judges, 16% of whom bankers. The position included four Pr´esidents de Chambre, and its members, coming from various sectors, had less than average centrality
Interlocking Judges
285
Fig.1 Advice Seeking among Judges: The Reduced Network. Note: Pattern of relationships between positions of approximately structurally equivalent judges in the advice network for all members. Position One includes 20 judges, position Two 25 judges, position Three 51 judges, position Four 7 judges, position Five 21 judges, and position Six 22 judges. One banker was lost to the “residual” category. The most central positions (Four, Five, and Six) are the positions including a high proportion of bankers and other representatives of the financial industry. Members of positions Four and Six (internal densities of 0.26 and 0.41 respectively), overall, tend to consult each other. The density table from which this representation of the structure is derived (using a high cutoff of 2.5) is presented in the appendix. Densities for each position are included in the diagonal of this table.
scores in the advice network. They tended not to seek advice from anybody in the other positions. They were more recent judges than members of position One, but they were also more sociable. They were sought out slightly more than position One members. Position Three was the largest in the courthouse. It was made up of 50 judges (among whom 14% were bankers), a third of whom with a diploma form Hautes Etudes Commerciales, the top French business school, and a career in large French business groups. They tended to be the youngest judges on average, consulted a lot with the outside economic world, had an intense social activity within the courthouse, but weak centrality scores as advisors. Members of this position tended to seek advice from positions Four, Five, and Six. Position Five included 21 judges (29% of whom were bankers), who were the most central and senior advisers in the organization, and included eight Pr´esidents de Chambre.
286
EMMANUEL LAZEGA AND LISE MOUNIER
They had been judges for nine years on average, had high average centrality scores in the advice network. Their difference with position Six is that they did not participate in the social life of the courthouse, did not consult with the business community, and did not seek advice from anyone. Positions Four and Six also included judges from the financial sector. Position Four included 22 judges (41% of whom were bankers), the oldest of the court on average, the most senior, and with high centrality scores in the advice network. This position included six Pr´esidents de Chambre. They were more active than average in the courthouse’s social life. They included also four members of the French noblesse d’Etat. Members of this position tended to seek advice from positions Five and Six. Position Six included seven medium seniority judges (43% of whom were bankers) with, on average, medium-level centrality scores. Members of this position tended to seek advice from positions Four and Five. This picture suggests that, at the TCP, judges coming from the financial industry are spread across different and often low-density positions. Assuming that there were coordinated influence efforts among these judges, these efforts did not systematically succeed in placing bankers at the heart of the indirect influence structure. Even if they did coordinate in an invisible way, they had to share their influence with other influential judges such as former high-level, non-banking business executives and industrialists, as well as representatives of the noblesse d’Etat. High internal densities are rare in positions in which the financial sector is most strongly represented. Status competition and the social discipline characterizing this organization seem to have defused, in part, the bankers’ threat to this court’s independence.
DISCUSSION AND CONCLUSION In sum, this chapter identifies a joint level of governance by using a broadly conceived structural and organizational approach to conflict resolution in markets. Joint governance is defined as a form of conflict resolution in which external stakeholders attempt to control commercial courts indirectly by using judges in official courts – i.e. traditional third parties in conflict resolution – as levers of influence to further their interests. Relying on structural theories of leverage, we assume that indirect control of that type requires a two-step leverage system. In the first step, first or second parties (or their representatives) must be elected judges in this court; in the second step these judges must influence other judges so as to gain their consent for specific decisions. The two steps that we use to reason about joint governance and the sharing of costs of control are equivalent to two dimensions of influence that organized
Interlocking Judges
287
business interests try to exercise on commercial courts. First, the selection of actors formally authorized to participate in the disposition of cases, particularly the judges. And second, the capacity to set the premises of other judges’ decisions. Premise setting for judicial decision making is key to our approach. Solutions to problems depend on how problems are defined or framed. Framing highlights certain dimensions of a problem and downplays others (Lazega, 1992). Definitions of a problem that gain early acceptance among members of an organization, including judges in a courthouse, are likely to subsequently dominate in the definition of a solution. Sentencing is often about the kind of information that should be included in pre-sentence reports, i.e. about definition of the situation and reframing of problems. Events reported in these files are interpreted and judges attach meaning to them. Some of these judges have the authority or position to have their views taken seriously, and their views can be decisive with respect to who is likely to win. To examine more closely the nature of joint governance, we hypothesized, first, that the strongest the external stakeholders’ interests and expectations, the more likely they are to invest in shaping the courthouse, especially in selecting the judges; and, second, that the closer the judges (as levers of influence) are to other judges (as targets of influence) in the same court, the more influence they have on these other judges (i.e. on their targets). Based on the case of a large contemporary French commercial court, we confirmed these hypotheses. In effect, we can measure investments by an industry by looking at the number of consular judges that it has sponsored; we can measure the capacity of an industry to set the premises of judicial decisions by looking at the centrality of its representatives in the advice network among the judges, and then at the determinants of this centrality. We find that the financial industry is clearly over-represented on the bench and that judges with potential influence on judicial decision making in that court are judges who have spent more years on the bench than other judges, who are active in the social life of their court – in addition to having ties to judges in other courts and to people with responsibilities in the economy at large. We also show that, in spite of its interests and efforts, the financial and banking sector which places on the bench a high proportion of judges coming from its ranks, tries and seems to fail to exercise indirect control on commercial courts and thus on conflict resolution in markets. Consular judges coming from this sector are not more central than judges coming from other sectors in the advice network connecting the judges in this court. Thus they are not able to reach a position in which they can systematically set the premises of other judges’ decisions and to perform the second step of the control mechanism. Investments made by this highly organized economic sector for the promotion of consular judges in this specific Tribunal de Commerce are not supplemented with relational investments by these individual judges within the microcosm of the courthouse. Attempts to
288
EMMANUEL LAZEGA AND LISE MOUNIER
control the Tribunal de Commerce by the financial industry is thus both evident and challenged from within the court. Our broadly conceived structural approach is limited in two ways. First, we only have indirect evidence of member judges’ influential status. Direct observation of such challenges and conflicting influences was not accessible to us since deliberation among judges (as for a jury) is secret. Being sought out for advice by one’s peers is certainly an indicator of potential influence, but not an indicator of actual influence. Conflicts of definition of the situation between representatives of different industries are not captured. This is a partial test for our hypotheses, but one for which we were able to collect data. Second, social closeness of judges was not measured by reconstitution of a complete friendship network, only by eliciting information about participation in the social life of the court as a whole. More precise data on friendship ties in the organization would have been more conclusive. Finally, bankers who seem disorganized may actually be so organized that they look disorganized. There are many characteristics of the judge’s activities within and outside the court that we do not know about and which could be a basis for coordination. One example is membership in the Free Masonry, which is not public information available to us. Identifying the effects of such variables would also be important to understand how the business community shares the costs of control, affects the operations of the court and thus structures the probability of who wins in conflict resolution and joint governance. Indeed, it is difficult, without a systematic analysis of cases and their outcome (i.e. data that are not accessible at this stage of the research), to test the extent to which social relations affect actual conflict resolution or to understand who actually wins, what and when. We do not have docket data on the outcome of judicial decisions made by judges whose industry of origin is identified. Constraints on data collection prevented us from extracting from this case study a clearer view of the truly original mechanism of joint conflict resolution that characterizes this business community. Also, lack of litigation rates by industry and lack of access to detailed court statistics prevent us from providing a direct and conclusive test for our hypotheses. More information should be collected in order to understand more deeply this combination of external and self-governance of business. In addition, our results, at this stage, raise questions of generalizability. First, if the proportion of bankers is the same in most large commercial courts and if their lack of centrality as advisers is not confirmed, then the divide and tension between finance and the rest of the French business community can be said to structure the kind of joint governance observed here. Second, other kinds of “courts,” such as independent administrative authorities (Frison-Roche, 1999) or alternative dispute-resolution organizations such as arbitration courts (Dezalay & Garth, 1996) also belong to a scene in which civil society participates in the operations
Interlocking Judges
289
of State institutions or institutions endowed with official power. They should be examined in the same way to confirm or disconfirm the generalizability of our results. Obviously more research is needed to characterize the various combinations of external and self governance of business. Further docket research should be able to reconsider the effect obtained here and shed more light on joint governance. Nevertheless, we believe that our approach adds value to current studies of market governance, particularly to conflict resolution in markets. External governance and self-governance are usually studied separately. Here we look at a system in which the two are combined. This joint governance can take forms that are different from the much heralded Anglo-Saxon responsive self-regulation (see Ayres & Braithwaite, 1992, and their “Benign Big Gun”). In France, a complex and old system of cooperation between the State, local Chambers of commerce, and citizens, produces consular commercial courts and specific ways of sharing the costs of control. The latter are likely to operate under many influences coming from the business community in ways that reflect the importance of specific actors (for example, the financial industry), but also in ways that point to a variety of leverage processes. The important actors have the resources to incur the costs of placing many consular judges on the bench. But staying capacity is not everything. The existence of long term relational investments in ties to other judges coming from other sectors of the economy shows that additional influence mechanisms must be examined to understand joint governance. In conclusion, economic actors such as interdependent firms spend time and resources trying to structure their environment, their opportunity structure, and the governance mechanisms that constrain them.20 These efforts are often built into the operations of economic institutions, especially institutions representing joint governance. Institutional economics and sociology have looked at contexts in which economic activities and competition take place as legally and culturally defined (Fligstein, 2002; Milgrom et al., 1990; North, 1990; Reynaud, 1989; Swedberg, 1993; Williamson, 1985, 1996). Work presented in this chapter also suggests that economic sociology could benefit more systematically from organizational and structural studies of joint governance by looking at economic institutions such as Chambers of commerce, commercial courts, and many others – just as economic history does (Hirsch, 1985; Lemercier, 2001).
NOTES 1. See also Lazega (1994) and Lazega and Mounier (2002) on some of the social and organizational mechanisms of this self-regulation. About new arbitration courts and the role of corporate law firms in their promotion, see Dezalay and Garth (1999). In big international
290
EMMANUEL LAZEGA AND LISE MOUNIER
business, to find a quick and discrete solution to a conflict, large multinationals prefer arbitration in most countries, including France. 2. Since Max Weber’s 1924 study of the German 19th century B¨orse (Weber, 2000), many organizational sociologists’ work on economic institutions have greatly contributed to economic sociology. A few researchers, such as Baker (1984) in his study of the Chicago Options Exchange, have added a structural spin to such organizational approaches to economic institutions (for a theoretical basis, see White, 2002; for a summary, see Swedberg, 1994). 3. But not too close to their targets, in which case other processes are triggered by the organizational control regime (Lazega, 2000, 2001; Lazega & Krackhardt, 2000; Lazega & Lebeaux, 1995; Wittek, 1999). 4. We also follow here the tradition of structural analyses of business such as that of interlocking directorates (Stokman et al., 1985) and derived approaches to relationships between the political and economic spheres (Pizarro, 1999). 5. As in so many situations where they are supposed to be the prototypical third parties, State organizations have combined incentives as well. They have vested interests related to their own involvement in market activities. Therefore, for institutions of conflict resolution, such as courts dealing with market litigation, it may make a strong difference whether the State is represented by professional judges who are civil servants, or whether it delegates its judicial powers to elected judges representing local civil society and/or the business community. 6. “Pure” self-regulation, such as arbitration and its more secretive operations also exists as a formal avenue of conflict resolution for businesses. But it is usually limited to large multinational companies prepared to pay large sums, particularly when they want to keep a litigation case secret. 7. For the characteristics of the French system of commercial courts as an institution of combined external and self-governance, see, for example, Hirsch (1985), Ithurbide (1970), Szramkiewicz (1989). Today, there are 191 commercial courts in France, around 3,000 consular judges making approximately 300,000 judicial decisions each year, of which 50,000 concern insolvency issues (Ministry of Justice figures for 1998; www.justice.gouv.fr/publicat/tc1807.htm). 8. Two months after our fieldwork was carried out, French commercial courts went on strike, for the first time in their history, against a proposed 2001 reform of their system. At the TCP, we were surprised to see so many judges haunting the corridors of the building during weeks of strike: many were retired, “all dressed up with no place to go” as their younger colleagues would say. 9. The same is true in English and American history: “In bankruptcy cases, delay permits the remaining assets of the bankrupt corporation to run down, or to forestall rapid reorganization, thus impairing chances for corporate turnarounds” (Carruthers & Halliday, 1998, p. 474). There is thus a need for rapid judgments by an empowered court. In the past, patronage, political partisanship, and conflicts of interests bedeviled bankruptcy courts. A system of mutual accommodations and exchanges of favors brought all the players in the system into a tight coalition of mutually protective practitioners. Judges were necessarily part of this “bankruptcy ring” (Carruthers & Halliday, 1998, p. 480). 10. This lead to attempts at reforms of commercial jurisdictions. The French commercial justice system is also changing, with the State trying to increase its control over the Tribunal de Commerce. Since 1981, the attorney general has an office at the Court and may participate in the judges’ deliberations, especially in bankruptcy proceedings (which are not public). In 2001 and 2002, a new law was unsuccessfully debated at the National Assembly and the
Interlocking Judges
291
Senate which aimed at introducing professional and tenured judges in the midst of consular judges. This “mixity” is presented by reformers as a means to reassure businesses involved in court proceedings by improving control over the controllers (Colcombet & Montebourg, 1998; Gaudino, 1998). In fact, it is worth noticing that this reform reproduced a paradoxical situation. When the French State used to be an even more active member of the business community (than it is today), it left commercial justice in the hands of this community. After it began to withdraw from direct control of the economy, since the 1980’s, it also began to try to increase its control over the judicial process in commercial courts. The justification for this change in policy is to reassure European and global investors who want to be certain that they will receive a fair treatment in French consular commercial courts. This explanation, however, must be taken with caution. In spite of a decade of selling some of its holdings, the French State’s direct ownership in the economy remains enormous, a long time caract´eristique of French capitalism. 11. Sometimes, they speak out about even broader and politically-charged policy issues, such as “Should commercial courts handle bankruptcy cases?” Many in French commercial courts are in favor of getting rid of such cases, which are precisely the cases that attract suspicions upon the activity of consular judges. 12. The classification of the top 500 French and European non-financial groups in France (Enjeux, Les Echos, November 2000) shows that 42 groups out of these 500 are mentioned as employers in the biographies of these consular judges. 38% of the judges have had one, two, or three such large groups as employers during their personal career. 13. NAF 60, code 65. NAFs are the French equivalent of Standard Industrial Classification codes. 14. For example, 21 were elected as candidates of the Association fran¸caise de banque and 5 as candidates of the Association fran¸caise de soci´et´es financi`eres. Among the financial institutions that were the employers of sitting judges (at the TCP alone), BNP-Paribas had sent seven judges, Suez had sent four, Soci´et´e G´en´erale four, Cr´edit Lyonnais four, and Cr´edit Commercial de France four. 15. Source: Enquˆete Emploi 2000, Institut national de la statistique et des e´ tudes e´ conomiques, CD-ROM Version. 16. Source: Institut national de la statistique et des e´ tudes e´ conomiques, Mensuel No. 202, Octobre 2001, Ile-de-France a` la page: “Gros plan sur l’emploi francilien en 1999.” 17. Source: Institut national de la statistique et des e´ tudes e´ conomiques, Comptabilit´e nationale, 2001 (www.insee/fr/indicateur/cnat annu/tableaux/t 1201 25 4.htm). In 1998, there were 1237 credit establishments in France, with 25428 bank tellers, and 714730 employees. 18. Additional data were also collected during interviews with the judges. But we were not authorized to elicit other network information, for example about friendship ties among judges (a question considered to be highly intrusive in French society in general, and in this organization in particular). We were allowed to ask questions about the following variables, which will be used in the regression model presented in Table 1: Did the judge seek advice, during the past two years, from former judges of the Tribunal of Commerce, from personalities in the business community, from professional judges, from the office of the attorney general? We were also allowed to ask questions about participation in the social life organized within the courthouse (dinners, trips, or conferences organized by the President, by their Chamber, by their promotion, or by their syndicat), and about opinions concerning current reform discussed at the National Assembly.
292
EMMANUEL LAZEGA AND LISE MOUNIER
19. Some of the characteristics of these judges were strongly correlated with others, as shown in Table 1. Therefore, we did not include all of them in the regression equation. For example, number of years as a judge is strongly correlated with being a Pr´esident de Chambre, or with age. Also, we do not know much about the judges. For example, several judges told us that membership in local chapters of the Free Masonry was an important variable in order to explain advice seeking behavior. That variable is not available to us for inclusion in this model. 20. Sometimes to escape from the pressure of competitive markets: As well described by Schumpeter, entrepreneurs often try to dump market competition on others, and to look for niches in which rents and discounts are available. Researchers have shown that businesses often do this illegally, through various forms of collusion (see, for example, Reiss, 1984; Stone, 1975; Vaughan, 1983, 1999). But in many countries they do so perfectly legally (with respect to the laws of these countries) and openly.
ACKNOWLEDGMENTS This study was funded by the Groupement d’Int´erˆet Public “Droit et Justice” at the French Ministry of Justice. We would like to thank Jean-Paul Jean for his help and advice, and Nathalie Busiaux, Karima Guenfoud and Claire Lemercier for help in carrying out fieldwork. We are grateful to Ronald Breiger, Bertrand Du Marais, Florence Jany-Catrice, Philippa Pattison, St´ephane Saussier, Harrison White, and the editors and reviewers of this volume for helpful comments on a first version of this chapter.
REFERENCES Ayres, I., & Braithwaite, J. (1992). Responsive regulation: Transcending the deregulation debate. Oxford: Oxford University Press. Baker, W. (1984). The social structure of a national securities market. American Journal of Sociology, 89, 775–811. Batenburg, R. S., Raub, W., & Snijders, C. (2003). Contacts and contracts: Dyadic embeddedness and the contractual behavior of firms. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 135–188). Oxford: Elsevier Science. Black, D. (1984). Social control as a dependent variable. In: D. Black (Ed.), Toward a General Theory of Social Control (pp. 1–36). New York: Academic Press. Blumberg, B. F. (1997). Das management von technologiekooperationen. Partnersuche und verhandlungen mit dem partner aus empirisch-theoretischer perspektive. Amsterdam: Thela Thesis. Burt, R. S. (1991). STRUCTURE 4.1. Center for the Social Sciences, Columbia University, New York. Buskens, V., Batenburg, R. S., & Weesie, J. (2003). Embedded partner selection in relations between firms. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 107–133). Oxford: Elsevier Science. Carruthers, B. G., & Halliday, T. (1998). Rescuing business: The making of corporate bankruptcy law in England and the United States. Oxford: Oxford University Press.
Interlocking Judges
293
Cheit, R. E., & Gersen, J. E. (2000). When businesses sue each other: An empirical study of state court litigation. Law & Social Inquiry, 25, 789–816. Colcombet, F., & Montebourg, A. (1998). Les tribunaux de commerce: Une justice en faillite? Paris: Rapport de la commission d’enquˆete de l’Assembl´ee Nationale. Dezalay, Y., & Garth, B. (1996). Dealing in virtue: International commercial arbitration and the construction of a transnational legal order. Chicago, IL: University of Chicago Press. Dunworth, T., & Rogers, J. (1996). Corporations in court: Big business litigation in U.S. federal courts: 1971–1991. Law & Social Inquiry, 21, 497–592. Favereau, O. (1994). R`egles, organisation et apprentissage collectif: Un paradigme non standard pour trois th´eories h´et´erodoxes. In: A. Orl´eans (Ed.), Analyse Economique des Conventions (pp. 113–138). Paris: Presses Universitaires de France. Felstiner, W. L. F., Abel, R. L., & Sarat, A. (1980). The emergence and transformation of disputes: Naming, blaming, claiming. Law & Social Inquiry, 15, 631–654. Flemming, R. B. (1998). Contested terrains and regime politics: Thinking about America’s trial courts and institutional change. Law & Social Inquiry, 23, 941–965. Fligstein, N. (2002). The architecture of markets. Princeton, NJ: Princeton University Press. Friesen, E. C., Gallas, E. C., & Gallas, N. M. (1971). Managing the courts. New York: Bobbs-Merrill. Frison-Roche, M.-A. (1999). Droit, Finance, Autorit´e: Les modes de r´egulation juridique propres aux autorit´es de march´es financiers. Paris: Rapport a` la Mission de recherche Droit et Justice, Minist`ere de la Justice. Galanter, M., & Epp, C. E. (1992). A beginner’s guide to the litigation maze. Business Economics, 27, 33–38. Gaudino, A. (1998). La Mafia des tribunaux de commerce. Paris: Albin Michel. Grabosky, P., & Braithwaite, J. (1986). Of manners gentle: Enforcement strategies of australian business regulatory agencies. Oxford: Oxford University Press. Gu´eroult, A., Lamotte, H., & du Marais, B. (1993), Les garanties et le cr´edit aux entreprises. Rapport du Conseil National du Cr´edit, Paris, Editions de la Banque de France. Hawkins, K. O. (1984). Environment and enforcement. Oxford: Oxford University Press. Hawkins, K. O., & Thomas, J. M. (Eds) (1984). Enforcing regulation. Boston, MA: Kluwer. Heydebrand, W., & Seron, C. (1990). Rationalizing justice: The political economy of federal district courts. Albany: SUNY Press. Hirsch, J.-P. (1985). Les deux rˆeves du commerce. Entreprise et institution dans la r´egion lilloise (1780–1860). Paris: Editions de l’EHESS. Ithurbide, R. (1970). Histoire critique des tribunaux de commerce. Paris: LCDG. Jean, J.-P. (2000). La r´eforme des tribunaux de commerce. In Regards sur l’actualit´e. Paris: La documentation franc¸aise. Lazega, E. (1992). Micropolitics of knowledge: Communication and indirect control in workgroups. New York: Aldine-de Gruyter. Lazega, E. (1994). Les conflits d’int´erˆets dans les cabinets am´ericains d’avocats d’affaires: concurrence et auto-r´egulation. Sociologie du Travail, 35, 315–336. Lazega, E. (2000). Enforcing rules among peers: A lateral control regime. Organization Studies, 21, 193–214. Lazega, E. (2001). The collegial phenomenon: The social mechanisms of cooperation among peers in a corporate law partnership. Oxford: Oxford University Press. Lazega, E., & Krackhardt, D. (2000). Spreading and shifting costs of lateral control in a law partnership: A structural analysis at the individual level. Quality and Quantity, 34, 153– 175.
294
EMMANUEL LAZEGA AND LISE MOUNIER
Lazega, E., & Lebeaux, M.-O. (1995). Capital social et contrainte lat´erale. Revue Fran¸caise de Sociologie, 36, 759–777. Lazega, E., & Mounier, L. (2002). Interdependent entrepreneurs and the social discipline of their cooperation: Structural economic sociology for a society of organizations. In: O. Favereau & E. Lazega (Eds), Conventions and Structures in Economic Organization: Markets, Networks, and Hierarchies. Cheltenham: Edward Elgar Publishers. Lemercier, C. (2001). La Chambre de Commerce de Paris, 1803–1852. Un “corps consultatif” entre repr´esentation et information e´ conomiques. PhD thesis. Paris: Ecole des Hautes Etudes en Sciences Sociales. Macaulay, S. (1963). Non-contractual relations in business: A preliminary study. American Sociological Review, 28, 55–67. McIntosh, W. V., & Cates, C. L. (1997). Judicial entrepreneurship: The role of the judge in the marketplace of ideas. Westport, CT: Greenwood Press. Milgrom, P. R., North, D. C., & Wiengast, B. R. (1990). The role of institutions in the revival of trade: The law marchant, private judge, and the champagne fairs. Economics and Politics, 2, 1–23. North, D. (1990). Institutions, institutional change, and economic performance. Cambridge: Cambridge University Press. Pizarro, N. (1999). Appartenances, places et r´eseaux de places: la reproduction des processus sociaux et la g´en´eration d’un espace homog`ene pour la d´efinition des structures sociales. Sociologie et soci´et´es, 31, 143–161. Raub, W., & Weesie, J. (1993). Symbiotic arrangements: A sociological perspective. Journal of Institutional and Theoretical Economics, 149, 716–724. Raub, W., & Weesie, J. (2000). The management of durable relations. In: J. Weesie & W. Raub (Eds),The Management of Durable Relations: Theoretical and Empirical Models for Households and Organizations (pp. 1–32). Amsterdam: Thela Thesis. Reiss, A. J., Jr. (1984). Selecting strategies of social control over organizational life. In: K. O. Hawkins & J. M. Thomas (Eds), Enforcing Regulation (pp. 23–35). Boston, MA: Kluwer. Reynaud, J.-D. (1989). Les r`egles du jeu. Paris: Armand Colin. Shapiro, S. P. (1984). Wayward capitalists: Target of the securities and exchange commission. New Haven, CT: Yale University Press. Stokman, F. N., Ziegler, R., & Scott, J. (Eds) (1985). Networks of corporate power. Cambridge: Polity Press. Stone, C. D. (1975). Where the law ends: The social control of corporate behavior. New York: Harper & Row. Swedberg, R. (1993). Economics and custom. Journal of Institutional and Theoretical Economics, 149, 204–209. Swedberg, R. (1994). Markets as social structures. In: N. Smelser & R. Swedberg (Eds), Handbook of Economic Sociology (pp. 255–282). Princeton, NJ: Princeton University Press. Szramkiewicz, R. (1989). Histoire du droit des affaires. Paris: Montchrestien. Vaughan, D. (1983). Controlling unlawful organizational behavior. Chicago, IL: University of Chicago Press. Vaughan, D. (1999). The dark side of organizations: Mistake, misconduct, and disaster. Annual Review of Sociology, 25, 271–305. Voss, T. (2003). The rational choice approach to an analysis of intra- and interorganizational governance. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 21–46). Oxford: Elsevier Science.
Interlocking Judges
295
Weaver, S. (1977). Decision to prosecute: Organization and public policy in the antitrust division. Cambridge, MA: MIT Press. Weber, M. (2000). Die B¨orse. Translated by S. Lestition in Theory and Society, 29, 303– 337. Wheeler, S., Mann, K., & Sarat, A. (1988). Sitting in judgment: The sentencing of white collar criminals. New Haven, CT: Yale University Press. White, H. C. (2002). Markets from networks: Socioeconomic models of production. Princeton, NJ: Princeton University Press. Williamson, O. E. (1985). The economic institutions of capitalism. New York: Free Press. Williamson, O. E. (1996). The mechanism of governance. Oxford: Oxford University Press. Wittek, R. (1999). Interdependence and informal control in organizations. Amsterdam: Thela Thesis.
APPENDIX Table A1 is the density table for the advice network among the judges of the Commercial Court of Paris, produced by Structure 4.1 (Burt, 1991). In this density table, cell i, j is the average relation from someone occupying position i to someone in position j. The average relation between any two people in the network is 0.101. The two image matrices in Table A2 are derived from the density table in Table A1. The cutoff for the first matrix is the overall density of the network (cell i, j = 1 if cell i, j density is greater than the average relation in the network between any two people) and a high density of 2.5 for the second matrix. This second matrix is used to represent the structure in Fig. 1.
Table A1. Density Table for the Advice Network among the Judges of the Commercial Court of Paris. Position
1
2
3
4
5
6
Residual
1 2 3 4 5 6
0.000 0.000 0.002 0.000 0.000 0.000
0.001 0.000 0.115 0.103 0.001 0.106
0.000 0.000 0.027 0.013 0.000 0.011
0.000 0.000 0.495 0.539 0.000 0.558
0.004 0.005 0.437 0.450 0.006 0.449
0.000 0.000 0.269 0.316 0.000 0.256
0.000 0.000 0.367 0.393 0.000 0.313
Residual
0.000
0.176
0.069
0.396
0.371
0.291
0.000
296
EMMANUEL LAZEGA AND LISE MOUNIER
Table A2. Two Images Matrices for the Advice Network among the Judges of the Commercial Court of Paris. Position 1 2 3 4 5 6
123456
123456
000000 000000 010111 010111 000000 010111 Cutoff at overall density
000000 000000 000111 000111 000000 000111 Cutoff at 2.5, used for Fig. 1
KEEPING A JOB: NETWORK HIRING AND TURNOVER IN A RETAIL BANK Kathryn M. Neckerman and Roberto M. Fernandez ABSTRACT The literature on job networks predicts that employees referred through networks would be better matched and mentored and thus would have lower turnover. However, existing research on this question has neglected the ways in which network effects are contingent upon firm organization. Using the personnel records of a large retail bank, we examine the relationship between network recruitment and turnover among new employees. There was no significant difference between network referrals and non-referrals, but referrals eligible for the employee referral program did have lower turnover. These results are explicable in light of the bank’s organization.
INTRODUCTION Recruitment through the personal networks of current employees is widespread in American business. Network recruitment represents an inexpensive way for firms to generate an applicant pool; more important, employers believe these networks will bring them better-qualified applicants (Granovetter, 1995). Recruitment networks may be particularly important for providing information about difficultto-observe qualities such as an applicant’s honesty, reliability, and “fit” with corporate culture (Chatman, 1991; Coverdill & Finlay, 1998). An added benefit, some believe, is that the referring employee may help train and mentor the newly-hired
The Governance of Relations in Markets and Organizations Research in the Sociology of Organizations, Volume 20, 299–318 © 2003 Published by Elsevier Science Ltd. ISSN: 0733-558X/PII: S0733558X02200117
299
300
KATHRYN M. NECKERMAN AND ROBERTO M. FERNANDEZ
worker, and may even exert pressure for reliable work performance (Bailey & Waldinger, 1991). Use of recruitment networks seems to offer employers a means to gain information about their applicants and social control over new employees. The implication is that network recruits should be better employees. There is surprisingly little evidence on this question, however, and the few existing studies provide mixed results. One reason is that network recruitment may actually comprise a number of distinct actions; because surveys typically collapse these actions into one or two simple measures, the mixed empirical findings are not surprising. Underlying this measurement problem is a theoretical one. As we argue, the role and effect of recruitment networks is contingent on the firm’s organizational structure, which activates some functions of networks and suppresses others. Previous research has done little to specify how network effects are contingent on firm characteristics, in part because of data limitations: most studies of job networks are based on surveys of workers which provide little information about the employing firms. In this chapter, we report new evidence from the personnel records of a large retail bank. Although our data do not permit us to compare network effects across varying organizational structures, our knowledge of the bank’s organization allows a refinement of hypotheses about the impact of network recruitment. We examine a key employment outcome: turnover. Matching theory would predict that network referrals would have lower turnover than non-referrals because networks are a conduit for information that improves match quality. The “social enrichment” perspective (Fernandez et al., 2000), which emphasizes post-hire processes of informal training, mentoring, and social control, leads to the same prediction. However, as we discuss below, elements of the bank’s organization – namely a highly selective hiring process and a large and geographically dispersed structure – limit some of the mechanisms through which recruitment networks are theorized to influence employment outcomes. The bank’s organization, in other words, is likely to dampen the effect of network referral on employment outcomes such as turnover. On the other hand, the bank’s employee referral program creates a distinction among network referrals – between “claimed” and “unclaimed” referrals – based on the intermediary’s willingness to stake time and reputation on their behalf, a distinction likely to reflect the intermediary’s judgment of match quality. Consistent with our expectations, we found no significant effect of network referral on turnover. Among the “claimed” referrals, however, turnover rates were significantly lower than among unclaimed referrals and non-referrals, a result that is accounted for entirely by the fact that claimed referrals were less likely to be fired. Indeed, unclaimed referrals were significantly more likely than claimed referrals or non-referrals to quit; this finding reinforces the idea that claiming is a strong signal of match quality. The results suggest the value of more careful specification of firm characteristics in analyses of network recruitment.
Keeping a Job: Network Hiring and Turnover in a Retail Bank
301
BACKGROUND Social embeddedness provides a resource for coping with the uncertainty inherent in economic transactions. Economic actors engaged in search commonly use networks as sources of third-party information about prospective “partners,” whether those partners be employees, suppliers, or partners in joint ventures (see Granovetter, 1995; Rees, 1966; as well as Buskens et al., 2003; Gulati & Wang, 2003 in this volume). The use of referral networks in hiring is widespread across industry sector and occupational level, driven in part by employers’ interest in obtaining a more carefully selected applicant pool and in getting more intensive information about those applicants. Referral networks may be particularly important in providing information about “unobservable” characteristics, such as honesty or dependability, which are difficult to assess through the usual screening methods. Referral networks may also give applicants information about the job. Intermediaries are in a position to provide a more realistic and nuanced picture of working conditions, opportunities for promotion, and other important features of the job (Halaby, 1988; Jovanovic, 1984; Wanous, 1980). In particular, intermediaries can provide information about more difficult-to-observe aspects of the job and work context, including information the firm has an interest in concealing. Matching theory suggests that network referral has implications for job outcomes such as turnover (Jovanovic, 1979). For matching theory, productivity is a function of match quality, or the match between skills and other qualities of the worker and distinctive features of the job. Jovanovic describes jobs as “experience goods”: match quality becomes evident only post-hire when productivity can be observed; it cannot be known with certainty before the new worker is hired. Search theory posits, however, that search can reduce uncertainty and improve match quality (Mortensen, 1986; Stigler, 1962). Informal methods, such as network recruitment, typically allow more intensive search and therefore are more likely to reduce uncertainty about match quality (Barron & Bishop, 1985). Search and matching theory imply, then, that network referrals would have higher starting wages, slower wage growth, and lower turnover (Simon & Warner, 1992). As Buskens and Raub (2002) note, networks also provide resources for “control,” providing economic actors more assurance that their partners will not default on their commitments. When an economic transaction is socially embedded, such a default harms reputation and may even lead to stronger sanctions such as social ostracism (Raub & Weesie, 1990). In the employment context, referral networks are commonly believed to function as well as they do because of these resources for control. The intermediary’s concern for reputation within the firm should constrain him or her to be cautious about referring applicants and candid in making recommendations. An intermediary who consistently refers poor candidates or
302
KATHRYN M. NECKERMAN AND ROBERTO M. FERNANDEZ
exaggerates their quality is likely to lose credibility and, potentially, future opportunities to recommend new workers (Granovetter & Tilly, 1988; Grieco, 1987; Saloner, 1985). In addition, an intermediary concerned about reputation may also monitor the performance of the new hire and exert social control if necessary to ensure diligent work. Of course, the interests of firm and intermediary do not wholly coincide. The intermediary is embedded not only in the firm but also in private networks of friends and relatives. So positioned, he or she may have obligations to these personal associates that outweigh loyalty to the firm and concern about reputation, and feel socially constrained to refer those associates even if they are of poor quality. Once the friend is hired, the interests of intermediary and firm may become more congruent: whether motivated by loyalty to the friend or by a more self-interested preference for working with friends, the intermediary may provide informal training and mentoring that boosts the productivity of the new worker. However, anecdotal accounts suggest that here, too, interests of firm and intermediary may diverge, as co-workers with personal ties may cover for one another or share unsanctioned strategies for making the job easier (Bailey & Waldinger, 1991; Blau, 1985; Wenger, 1998). Employers can capture some of the benefits of referral networks even if the reputation incentive is absent or weak – in other words, even if the intermediary has little or no positive reason to act on behalf of the firm. As mentioned previously, personal networks tend to be homophilous. An employer who recruits through personal networks, particularly by selecting the referrals of high-performing employees, can expect thereby to improve the applicant pool even if the intermediary is unconcerned about reputation. Moreover, even in the absence of the reputation incentive, the employer might benefit from informal training and mentoring between a new employee and his or her intermediary. Casual information-sharing pervades sociability among co-workers (Chetkovich, 1997; Wenger, 1998). Nonetheless, firms clearly gain more from recruitment networks when intermediaries are motivated to act on behalf of their employer. Given this, it is not surprising that many firms use employee referral programs to structure incentives for intermediaries (Halcrow, 1988). Although the details of these programs vary, in general they provide a financial reward and, perhaps, some social and professional recognition for employees who refer a successful applicant. Often the referral bonus is made contingent on the tenure of the new hire, which provides an added incentive not only for referring well-matched applicants but also for post-hire processes that promote training, mentoring, and social integration for the new hire. But note that the referral bonus by itself provides no incentive for selective referral; employees increase their chances of getting a bonus by referring more applicants, and do not bear the added screening costs that result from high-volume referral.
Keeping a Job: Network Hiring and Turnover in a Retail Bank
303
The higher are screening costs, the more firms have an interest in sanctioning employees for referring poorly-matched applicants, most often by designing the employee referral program to activate the reputation incentive. The theoretical literature implies that employees recruited through networks will be better matched and will have a more supportive post-hire environment, and thus will have lower turnover than employees recruited through other means. The empirical evidence has been uneven, however. Early studies reported that network referrals had lower turnover than other employees (Caldwell & Spivey, 1983; Decker & Cornelius, 1979; Gannon, 1971; Reid, 1972), but more recent studies controlling for employee qualifications and skills have found mixed results. Using data from the Employment Opportunities Pilot Project (EOPP), Coverdill (1998) and Sicilian (1995) both reported lower turnover among network referrals; analyzing a survey of engineers, Simon and Warner (1992) found the same result. Datcher (1983), however, in an analysis of male household heads from the Panel Study of Income Dynamics (PSID), found among white, non-college males that network referral was associated with higher rates of turnover; referral did reduce the turnover of black workers and of more educated workers. Finally, in an analysis of the personnel records of a large bank, Fernandez et al. (2000) reported no relation between network referral and turnover. As Coverdill (1998) argues, these mixed results may reflect heterogeneity in network recruitment. Sometimes network recruitment reflects a fairly intensive involvement by the intermediary, who might provide information to the employer about the applicant and to the applicant about the firm; shape the outcome of the hiring process by coaching the applicant or by exerting influence on the employer; or provide post-hire training and mentoring. On the other hand, network recruitment could mean only a casual mention of a vacancy at the firm. Theoretically, it is apparent that these different actions have different implications for match quality and, ultimately, for turnover. The lack of fine-grained data on network recruitment, however, has impeded empirical study. In most household surveys, network recruitment is indicated with only one or two questions about whether or not a friend or relative provided information or helped the respondent get the job. In employer surveys, network recruitment is typically measured as one of a menu of recruitment sources, with little information collected about the process. In neither case do we get much leverage on heterogeneity in the process of network recruitment. Like most researchers, we lack direct measures of the process of network recruitment, but we take advantage of the way that organizational structure constrains the role of recruitment networks. We start from the premise that the mechanisms through which referral networks improve job-employee matches are contingent upon specific features of organizational structure. For these
304
KATHRYN M. NECKERMAN AND ROBERTO M. FERNANDEZ
mechanisms to operate, the intermediary must have good information about the vacant position. In addition, the intermediary must have an incentive to use this information in a way that benefits the firm. This incentive in turn depends on sanctions the employer deploys with respect to the intermediary. Finally, the more extensive and efficient are alternative, usually formal, means of providing information to applicant and firm, and of training and mentoring new employees, the less influence networks are likely to have (e.g. Fernandez et al., 2000). The site for our research is a large retail bank located in the western United States, the same site studied in Fernandez and Weinberg (1997). (Fernandez et al., 2000 reference a different site of the organization.) This bank used an employee referral program to encourage network recruitment. Current employees were offered a bonus of $150 to $2,000 – the amount varied with the rank of the new hire – if they referred an applicant who was hired and retained through the 90-day probation period. About 8% of applicants were network referrals. Previous research with these data suggest that these referral networks served as a conduit for information both to the firm and to the applicant. Network referrals tended to apply at more auspicious times – when there was a lower ratio of applicants to positions – and to have experience and skills that were better suited to the job for which they applied. Even taking these factors into account, however, network referrals were significantly more likely to be interviewed and, conditional on being interviewed, more likely to receive a job offer (Fernandez & Weinberg, 1997). Several features of the bank’s organization are important for our purposes. First, the bank’s hiring process was highly selective; during our study period, only 6.5% of applicants received job offers. Second, the bank is large and decentralized: recruitment and hiring were handled primarily through a regional clearinghouse, and from there new employees were dispersed across more than 80 branch banks. Unlike many firms, where a network recruit can expect to have regular contact with the employee who referred him or her, only about 10% of the network recruits in our sample worked at the same branch as their referral source. Finally, the bank’s employee referral program activated the reputation incentive – but only for intermediaries who chose to participate. To be eligible for the program, the referring employee had to submit a referral form along with a resume or application for the prospective employee; in addition, the referring employee’s name had to be listed on the employment application and on another form by the time the new employee started work – in other words, before it was obvious whether or not the new employee would perform well. We call their submission of the referral form claiming the referral. Not all network recruits were claimed by their intermediaries at the firm: only two-thirds had a contact willing to invest time and risk reputation by filing the employee referral program paperwork.
Keeping a Job: Network Hiring and Turnover in a Retail Bank
305
These features have implications for the effect of network recruitment on match quality and on post-hire processes that lower turnover. The difference between network referrals and non-referrals is likely to be attenuated in this context. The highly selective hiring process is likely to attenuate differences in match quality, at least in those traits that can be assessed via application and interview. In addition, because new employees are dispersed across more than 80 branch banks, the applicant usually cannot get branch-specific information about supervisors or co-workers, nor can the intermediary take an active role in post-hire training and social control. Both the selective hiring process and the decentralized structure of the bank are likely to reduce the differences in turnover between network referrals and non-referrals. In addition, the bank’s employee referral program creates a new distinction, that between claimed and unclaimed referrals. Because the act of claiming a referral imposes some time costs and involves a risk to reputation – by explicitly linking the intermediary’s name with that of the applicant – the intermediary is unlikely to take this step unless he or she views the applicant as relatively likely to be successful. Intermediaries vary, of course, in their cost of time and in the value they place upon reputation at the firm; thus the act of claiming is a stronger signal for some intermediaries than for others. On average, however, claimed status should signal that the new hire is perceived to be relatively well-matched for the job, with the implication that claimed referrals should have lower turnover than unclaimed referrals.
HYPOTHESES Our hypotheses are based on the research literature discussed above as tailored to the specific context of the bank. For simplicity, we assumed that the effect of network homophily is too small to be detected in our small dataset and can be ignored for purposes of hypothesis formation. For instance, we assumed that in the absence of other mechanisms driving an effect of network referral on turnover, that the effect of network homophily on match quality and therefore on turnover would be so small as to be undetectable. As discussed above, we predicted that highly-selective hiring and the decentralization of bank employees would severely limit the effect of networks on turnover. We also expected that the employee referral program would motivate intermediaries to distinguish among applicants and to claim those they believed to be better matched to the job and the firm. Therefore we hypothesize that: Hypothesis 1. There is no difference in turnover between network referrals and non-referrals.
306
KATHRYN M. NECKERMAN AND ROBERTO M. FERNANDEZ
Hypothesis 2. Claimed referrals have lower turnover than unclaimed referrals. Although we have no direct measure of the quality of the intermediary’s information about the job, we do have access to a crude proxy: the intermediary’s performance rating. We assume that employees who do superior work have a firmer grasp of the job and the workplace culture, and thus are better able to judge who is a good match for the organization. Their referrals, therefore, should have lower turnover. We hypothesize: Hypothesis 3. Network referrals claimed by highly-rated intermediaries have lower turnover than network recruits referred by intermediaries with lower performance ratings. Finally, we consider the effect of network referral on voluntary and involuntary turnover. Most research on turnover elides this distinction or considers only voluntary turnover; for an exception, see Coverdill (1998), who finds that network referral has a similar, negative effect on both voluntary and involuntary turnover. Here we assume that voluntary turnover is driven primarily by poor information to the applicant, who has accepted a job that is later revealed to be unsatisfactory. Of course, a dissatisfied worker might perform so badly that he or she is terminated; however, the stigma attached to being fired seems likely to discourage such action. Ordinarily one would expect network recruits to have lower rates of voluntary turnover because they have better information prior to accepting the job (cf. Fernandez et al., 2000). However, in this case networks can provide only limited information to the applicant because of the bank’s decentralized structure. If network recruitment influences voluntary turnover primarily through information provided to the applicant, then we hypothesize that: Hypothesis 4. There is no difference in voluntary turnover between network referrals and non-referrals. Given this premise, there is no reason to suppose that voluntary turnover is sensitive to the actions or characteristics of the intermediary: Hypothesis 5. There is no difference in voluntary turnover between claimed and unclaimed referrals. Hypothesis 6. There is no difference in voluntary turnover between referrals with high-performing intermediaries and those with lower-performing intermediaries. Involuntary turnover, we assume, is driven primarily by poor information to the employer: it reflects a “hiring mistake” that would have been prevented with better information. Here, network referrals and non-referrals may not differ by
Keeping a Job: Network Hiring and Turnover in a Retail Bank
307
much: again because of the bank’s large and decentralized structure, in which the hiring office is separated geographically and organizationally from the branch banks, the flow of information from intermediary to the personnel office is likely to be limited. The impact of network recruitment on turnover occurs primarily, we expect, because the intermediary is selective about whom he or she recommends. Because the reputation incentive makes the intermediary more selective in recommending applicants, claimed referrals should be fired less often than unclaimed referrals. And, because high-performing intermediaries are likely to have a better understanding of job requirements than lower-performing intermediaries, fewer of their referrals should be fired. We hypothesize that: Hypothesis 7. There is no difference in involuntary turnover between network referrals and non-referrals. Hypothesis 8. Claimed referrals have lower involuntary turnover than unclaimed referrals. Hypothesis 9. Referrals with high-performing intermediaries have lower involuntary turnover than referrals with lower-performing intermediaries.
DATA AND METHODS Our research uses data on recruitment and turnover from more than 80 branches of a large retail bank, all located in the western U.S. The data were collected and made available by the bank’s Western Region Human Resources group (hereafter WRHR), which is the regional clearinghouse for applicants to the bank. The WRHR maintains a computer database tracking the progress of recent job applicants from initial inquiry to the hiring decision, supplemented by paper files including resumes, cover letters, and application forms. Another computer database includes information on the status and characteristics of current and past employees. Our dataset includes 256 employees in the four entry-level jobs for which WRHR had nearly complete information: Customer Service Representatives (CSR), Personal Bankers (PB), Business Bankers (BB), and Mortgage Consultants (MC). CSR is an hourly teller position in the retail bank. PB is a salaried position involving sales to individual customers. BBs are similar to PBs, but serve commercial customers. MC is a commissioned sales position in the commercial bank. During the time of our study, the starting pay for these jobs ranged from about $16,000 to about $45,000, with BBs starting toward the higher end of the pay range and CSRs starting toward the lower end.
308
KATHRYN M. NECKERMAN AND ROBERTO M. FERNANDEZ
Our dataset includes information for new hires who had inquired at the WRHR about one of the four entry-level jobs between January 1993 and March 1995. We tracked these new employees in the employee database until October 1996, so we have at least 18 months of data for everyone included in the sample. We measure turnover at six-month intervals, constructing dummy variables indicating whether an employee had left by six months, 12 months, and 18 months after the date of hire. In addition, we used the reasons for departure to code variables indicating voluntary and involuntary separation. In five cases the decision to separate appeared to be mutual, and these cases were coded “0” on both quit and fire measures. For the measures of network referral, we drew from several of the bank’s databases. The simplest measure is a dummy variable, “referral,” coded “1” for new hires who reported that they had contacted WRHR at the suggestion of a current employee at the bank; the bank’s employment application form prompts for this information, which is then keypunched into the bank’s applicant database. Among the 256 new hires, 45% reported that they had been referred by a current employee. In addition, we distinguished among these self-reported referrals on the basis of whether the contact had signed up for the employee referral program. Among the self-reported referrals in our sample, 66% had been referred by a bank employee who completed the paperwork necessary to be eligible for the bonus program; we call these “claimed” referrals. Finally, by linking the referral database to information on current employees, we got information on the performance rating of the referring employee. In annual performance reviews, the bank assigned a three-category rating with 1 being the highest; we constructed dummies indicating the rating of the referral source, collapsing ratings of 2 and 3 (only three cases had a rating of 3). Among claimed referrals, 43% had a referral source rating of 1, and 57% had a referral source rating of 2 or 3. The performance rating is missing for unclaimed referrals. Our control variables include measures that are conventionally included in models of turnover. The job characteristics include three dummy variables indicating occupation (MC, BB, and PB), with CSR as the omitted category, and the natural log of the job’s starting wage. In addition, using the resumes and job application forms in the bank’s paper files, we constructed measures of the employee’s skills and credentials. Our choice of variables reflects information about selection criteria provided by WRHR recruiters. The variables include years of work experience, divided into experience inside and outside the financial services industry, and evidence of computer skills and foreign language expertise, each of these coded 1/0. Years of bank experience squared captures the decaying value of work experience. WRHR advised that candidates for the four entry-level jobs should have either a college background or five years experience in financial services industry, and that the bank was wary of over-qualified candidates.
Keeping a Job: Network Hiring and Turnover in a Retail Bank
309
Based on this advice, we recoded years of education as a dummy variable, with 1 = 13–16 years of education and 0 = other, and included a term interacting banking experience with college education. Computer skills, foreign language skills, bank experience, and college background indicate skills that make the employee a better match for these bank jobs, at least according to WRHR, while bank recruiters looked less favorably upon non-bank work experience. These variables correspond to those used in an earlier analysis of hiring at the same bank, in which most had significant effects (Fernandez & Weinberg, 1997). Finally, the analysis includes several background measures for the new employees. About 12% of new hires had applied to the bank more than once; we indicate these repeat applicants with a dummy variable. We also include a dummy variable indicating gender (female = 1). Missing data for the dependent variable and for these control variables reduced the sample size from 256 to 227 (and to 217 for the analyses distinguishing voluntary from involuntary turnover), of whom 123 were non-referrals and 104 were referrals. Table 1 provides means for the total sample and for referrals and non-referrals separately. These figures show that half of the new hires were
Table 1. Means for Independent Variables for Total Sample and for Network Referrals and Non-Referrals. Non-referrals Number of cases Job characteristics Mortgage consultant Business banker Personal banker Customer service representative Full-time Starting wage
123
Referrals 104
Total 227
0.146 0.081 0.252 0.520 0.821 13.24
0.077 0.029 0.433∗ 0.462 0.788 14.06
0.115 0.057 0.335 0.493 0.806 13.33
Skills and credentials College background Bank experience (years) Non-bank experience (years) Computer experience Foreign language skills
0.797 4.805 3.821 0.659 0.366
0.798 3.462∗ 3.284 0.615 0.365
0.797 4.189 3.575 0.639 0.366
Other characteristics Repeat applicant Female
0.138 0.675
0.096 0.663
0.119 0.670
+
p < 0.10; ∗ p < 0.05; ∗∗ p < 0.01.
310
KATHRYN M. NECKERMAN AND ROBERTO M. FERNANDEZ
CSRs, with most of the rest hired as PBs. Most had a college background and some computer experience, and they averaged more than four years experience in the financial services industry. Most were female. Referrals and non-referrals differed significantly on only two characteristics: compared to non-referrals, referrals were more likely to work as personal bankers and had less banking experience.
RESULTS We first examine the zero-order relationships between referral source and turnover (see Table 2). Referrals had lower turnover rates than non-referrals did, and the difference widened with time from 3.7 percentage points at six months to 7.0 percentage points at 18 months. None of these differences was statistically significant, however. To facilitate comparison with the multivariate results, the lower part of Table 2 reports coefficients for the referral variable in analyses predicting turnover. Like the descriptive statistics, these coefficients indicate that referrals had lower turnover than non-referrals, but again the differences do not reach statistical significance. In the probit analyses reported in Table 3, we added our measures of job characteristics, skills and credentials, and other control variables. Few of these additional variables had significant effects on turnover. Business bankers and employees with foreign language skills tended to have lower turnover rates. Repeat applicants also had lower turnover, consistent with the interpretation that they were highly motivated to work at the bank and, once hired, were unlikely
Table 2. Effects of Network Referral on Turnover (Standard Errors in Parentheses). Non-referrals % turnover within 6 months % turnover within 12 months % turnover within 18 months
22.0 36.6 40.7
Coefficient in probit analyses predicting: Turnover within 6 months (Standard error) Turnover within 12 months (Standard error) Turnover within 18 months (Standard error) Number of cases
– – –
+
p < 0.10; ∗ p < 0.05; ∗∗ p < 0.01.
Referrals 18.3 29.8 33.7 −0.131 (0.191) −0.187 (0.173) −0.185 (0.171) 227
Total 20.3 33.5 37.4 – – –
Keeping a Job: Network Hiring and Turnover in a Retail Bank
311
Table 3. Effects of Network Referral on Turnover, Controlling for Characteristics of Job and Employee (Standard Errors in Parentheses). Dependent variable Turnover within 6 months
Turnover within 12 months
Turnover within 18 months
Network referral Referral
−0.194 (0.211)
−0.273 (0.187)
−0.282 (0.184)
Job characteristics Mortgage banker Business banker Personal banker Log of starting wage
0.049 (0.361) −1.475+ (0.818) −0.206 (0.300) −0.334 (0.445)
0.108 −1.213* −0.077 0.149
0.141 −0.465 0.625 −0.120
(0.305) (0.517) (0.240) (0.284)
−0.122 −0.042 −0.028 0.049 −0.557* 0.003 0.014
(0.370) (0.075) (0.031) (0.216) (0.224) (0.003) (0.045)
−0.446 (0.325) −0.084 (0.069) −0.012 (0.026) 0.021 (0.188) −0.357+ (0.188) 0.004 (0.003) 0.035 (0.042)
−0.231 −0.071 −0.023 0.030 −0.380* 0.004 −0.007
(0.320) (0.068) (0.026) (0.184) (0.184) (0.003) (0.041)
−1.004* (0.487) 0.129 (0.230) −1.141 (1.071)
−0.663* (0.321) −0.011 (0.200) 0.366 (0.739)
−0.809* (0.317) −0.130 (0.195) 0.791 (0.731)
Skills and credentials College background Bank experience Non-bank experience Computer experience Foreign language skill Bank experience squared Bank experience × College background Other characteristics Repeat applicant Female Constant −2 Log-likelihood Number of cases +p
208.712 227
(0.311) (0.600) (0.243) (0.290)
254.498 227
264.232 227
< 0.10; ∗ p < 0.05; ∗∗ p < 0.01.
to quit. To be consistent with other analyses of turnover, we also tried alternative specifications using years of education and combining the two experience measures into one, but the results did not vary substantially from those reported here. The bank’s highly selective hiring may explain why employee characteristics have so little effect on turnover. Bank recruiters are likely to screen out, on the basis of exactly these observable characteristics, applicants they deem unlikely to be stable and satisfactory employees; these hiring practices truncate variation in employee characteristics and bias downward our estimates of the effects of these variables on turnover. Including the control variables makes little difference to the estimated effect of network referral. The coefficients for referral become larger but remain insignificant, consistent with Hypothesis 1.
312
KATHRYN M. NECKERMAN AND ROBERTO M. FERNANDEZ
Table 4. Effects of Network Referral on Turnover, Distinguishing Claimed Status and Performance Rating of Referring Employee (Standard Errors in Parentheses). Dependent variable Turnover within 6 months Claimed status Claimed referral Unclaimed referral Number of cases
−0.419+ (0.246) 0.217 (0.293) 227
Turnover within 12 months −0.679** (0.224) 0.434 (0.266)
−0.714** (0.220) 0.488+ (0.266)
227
227
Claimed status and performance rating of referring employee Claimed referral, rating 1 −0.415 (0.336) −0.672* (0.301) Claimed referral, rating 2 or 3 −0.422 (0.300) −0.685* (0.274) Unclaimed referral 0.217 (0.293) 0.434+ (0.266) Number of cases
227
Turnover within 18 months
−0.671* (0.296) −0.748** (0.268) 0.487+ (0.266)
227
227
Note: Analyses reported in this table include all control variables listed in Table 3. + p < 0.10; ∗ p < 0.05; ∗∗ p < 0.01 (one-tailed test).
Next we examine the effect of participation in the employee referral program. The top panel of Table 4 reports selected coefficients from probit analyses in which we substituted variables indicating claimed and unclaimed referral for the single variable indicating network referral. (The analyses reported in Table 4 include all the control variables included in earlier multivariate analyses; the effects of these variables were essentially unchanged in the new models.) By 12 months from date of hire, claimed referrals had significantly lower turnover than either non-referrals or unclaimed referrals. In fact, the coefficient for unclaimed referrals was positive and marginally significant at 18 months, suggesting that an employee’s failure to claim a referral may actually be a negative signal about a new hire. Consistent with our second hypothesis, new employees whose contacts in the bank were willing to claim them had lower turnover than other new hires, even those who said they had been recruited through networks. These results rule out an interpretation of claimed referrals’ lower turnover as reflecting only social control through the probation period. An intermediary seeking to qualify for the referral bonus might press the new hire to perform well and not to quit, but there is no financial incentive for this action beyond the 90-day probation period. If this incentive were paramount, the difference between claimed and unclaimed referrals should be evident in the early months of employment and disappear later. In fact, the reverse is true. At three months after the start date,
Keeping a Job: Network Hiring and Turnover in a Retail Bank
313
there was no significant difference between claimed and unclaimed referrals, and the difference at six months is only marginally significant; it becomes larger and more significant with time. We also examined whether the referrals of higher-performing employees would have lower turnover. In the analysis reported in the lower panel of Table 4, we substituted dummy variables indicating the performance rating of the referring employee for the variable indicating claimed referral. (Recall that we have no information on performance rating of the referral source for the unclaimed referrals.) Contrary to Hypothesis 3, we find no evidence that the performance rating of the referring employee is related to the turnover of the new hire he or she recommended; the coefficients indicating higher- and lower-performing intermediaries are statistically indistinguishable. Our last analyses distinguish between voluntary and involuntary turnover. Table 5 reports coefficients for the referral variables only. Voluntary turnover was twice as common as involuntary: 22% of new hires quit within the first 18 months of employment, while only 10% were fired. Network referrals were no less likely than non-referrals to quit: 27% of referrals quit within 18 months, compared to only 18% of non-referrals, and in multivariate analysis the difference was statistically insignificant, which is consistent with Hypothesis 4. Thirty-nine percent of unclaimed referrals quit within 18 months, compared to 22% of claimed
Table 5. Effects of Network Referral on Voluntary and Involuntary Turnover by 18 Months (Standard Errors in Parentheses). Dependent variable Quit within 18 months 1. Referral “Claimed” referrals “Unclaimed” referrals Referral source rating 1 Referral source rating 2 or 3 Unclaimed referrals Number of cases
2.
Fired within 18 months 3.
4. −0.846** (0.312)
0.197 (0.207) −0.047 (0.239) 0.640* (0.284) −0.019 (0.307) −0.070 (0.291)
217
217
0.640* (0.284) 217
203
Note: Analyses reported in this table include all control variables listed in Table 3, with one exception: the analysis of being fired excludes the business bankers. + p < 0.10; ∗ p < 0.05; ∗∗ p < 0.01.
314
KATHRYN M. NECKERMAN AND ROBERTO M. FERNANDEZ
referrals and 18% of non-referrals. In the multivariate analysis, unclaimed referrals were significantly more likely to quit (column 2, Table 5) contradicting Hypothesis 5. As before, there were no differences by performance rating of the referring employee (Hypothesis 6). The results for unclaimed referrals were unexpected and raise questions about our assumption that voluntary turnover reflects a deficiency in the applicant’s information. It is difficult to tell a story in which unclaimed network referrals have less information than non-referrals. We speculate that these results reflect two underlying processes. The first is that claiming is a strong signal of match quality. All else equal, claimed referrals should be better matched than other employees, and unclaimed referrals should be worse matched; the fact that intermediaries passed up an opportunity to earn a substantial reward presumably reflects significant doubts about the suitability of these new hires. Second, we speculate that network referral is a rough proxy for the extent of personal networks: all else equal, referrals quit more often than non-referrals because their wider network brings them information about more alternative opportunities at other firms. Jointly, these two propositions predict the pattern of results we find: for claimed referrals, the effect of match quality offsets the effect of extensive personal networks, and we observe no difference in voluntary turnover between claimed referrals and non-referrals; for unclaimed referrals, both (poor) match quality and extensive personal networks should increase voluntary turnover. We cannot test this ad hoc explanation with our data, but it remains useful to keep in mind for future research. Descriptive statistics on involuntary turnover showed that fewer referrals than non-referrals were fired over the 18-month period: 4% compared to almost 16%. Column 4 in Table 5 shows that the effect of referral remains significant when other factors are controlled – a result inconsistent with Hypothesis 7. Claimed and unclaimed referrals also differed significantly, which confirms Hypothesis 8. Thirteen percent of unclaimed referrals were fired within 18 months, while not one of the claimed referrals was fired. Without any variation on the dependent variable among claimed referrals, we could not estimate the effects of claimed and unclaimed referral in a multivariate analysis. Therefore, Hypothesis 9 cannot be tested. But in the earlier analyses, introducing control variables had little effect; the difference between claimed and unclaimed referrals would probably persist if we could control for the other factors. These results have a straightforward interpretation in the matching theory framework: that referring employees are good judges of human resources, tending to claim employees who are indeed better-matched and therefore less likely to be fired. At least in the descriptive statistics, however, performance rating of employees appeared to have no effect (results not shown).
Keeping a Job: Network Hiring and Turnover in a Retail Bank
315
SUMMARY AND CONCLUSION In this chapter we examined the relationship between network recruitment and the turnover of new bank employees. We found no evidence of differences in turnover between non-referred and referred employees overall. But when we distinguished among network referrals on the basis of whether the referring employee had gone through the process necessary to become eligible for the bank’s employee referral program, we found that these “claimed” referrals had significantly lower turnover rates than either the “unclaimed” referrals or the non-referrals. In addition, we found that referred employees were less likely than non-referrals to be fired; this difference was due entirely to the record of the claimed referrals, none of whom were fired. There was no difference in quit rates between referrals and non-referrals, although we found that unclaimed referrals were more likely than others to quit. And we found no effect on turnover of the performance rating of the referring employee. These findings are explicable in light of the bank’s organizational structure. A highly-selective hiring process narrows differences between referrals and non-referrals, making it less likely that significant differences in turnover would emerge. In addition, the bank’s structure, in which few new employees work in the same branch as their intermediaries, precludes some pre-hire and post-hire network processes that would ordinarily reduce turnover. Finally, the employee referral program creates a distinction among referrals by asking intermediaries to commit time and risk reputation in order to receive a bonus; in effect it provides an indirect measure of the intermediary’s evaluation of the applicant, an evaluation that appears from our results to be merited. Effectively the bank’s structure appears to have suppressed most functions of recruitment networks, with the exception of their role in selecting well-matched applicants, a role that is signaled and perhaps strengthened by the employee referral program. Our work implies that the role and impact of network processes are dependent upon organizational structure, and provides a possible explanation for the mixed findings in empirical studies of the relationship between recruitment networks and job outcomes. Unlike our research, for instance, some studies find that network referrals are less likely to quit (Coverdill, 1998; Sicilian, 1995; Simon & Warner, 1992). But these studies are based on samples of employees at a wide range of firms, including some small and/or spatially-consolidated settings in which network contacts can take a more active role in social integration and can provide better employee-side information. More broadly, our work contributes to research on the relationship between search and selection procedures, on the hand, and outcomes of partnerships on the
316
KATHRYN M. NECKERMAN AND ROBERTO M. FERNANDEZ
other. This relationship is seldom examined in research on network effects in other domains. In the small literature on employment networks, there is some evidence on the relationship between recruitment networks and outcomes such as wages, occupational position, and tenure/turnover. Much of this work has focused on questions of information and influence (Boxman et al., 1991; Bridges & Villemez, 1986; Corcoran et al., 1980; Lin et al., 1981; Marsden & Hurlbert, 1988; Montgomery, 1991; Wegener, 1991). We broaden it by highlighting the importance of mechanisms of control, particularly the firm’s use of the reputation incentive. Both theoretical reasoning and empirical results indicate that the interests of firms and intermediaries do not entirely coincide, and that the disparity of interests is to some extent dependent upon incentives structured by the firm. In the firm we studied, the spatial decentralization of the firm is likely to have diminished the reputation incentive – because potential intermediaries have spatial and organizational locations distant from the hiring office – making it particularly important to reinforce that incentive through the formal mechanism of the employee referral program. There may be analogues in other contexts as well: research on buyer-supplier relationships, for instance, or on other kinds of between-firm transactions would benefit from considering the implications of firm organization for the structure of information flows and incentives, rather than taking for granted that firms can be treated as unitary.
ACKNOWLEDGMENTS The authors would like to acknowledge the generous financial support of the Citibank Behavioral Sciences Research Council. At Stanford, Emilio Castilla, Paul Moore, Damon Philips, Alison Siskin, and Nancy Weinberg provided exceptional research assistance. For useful comments and suggestions on earlier versions of this chapter, we thank David Card, Toby Parcel, Canice Prendergast, and Nancy Weinberg.
REFERENCES Bailey, T., & Waldinger, R. (1991). Primary, secondary, and enclave labor markets: A training systems approach. American Sociological Review, 56, 432–445. Barron, J. M., & Bishop, J. (1985). Extensive search, intensive search, and hiring costs: New evidence on employer hiring activity. Economic Inquiry, 23, 363–382. Blau, G. (1985). Relationship of extrinsic, intrinsic, and demographic predictors to various types of withdrawal behavior. Journal of Applied Psychology, 70, 442–450.
Keeping a Job: Network Hiring and Turnover in a Retail Bank
317
Boxman, E. A. W., de Graaf, P. M., & Flap, H. D. (1991). The impact of social and human capital on the income attainment of Dutch managers. Social Networks, 13, 51–73. Bridges, W. P., & Villemez, W. J. (1986). Informal hiring and income in the labor market. American Sociological Review, 51, 574–582. Buskens, V., Batenburg, R. S., & Weesie, J. (2003). Embedded partner selection in relations between firms. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 107–133). Oxford: Elsevier Science. Buskens, V., & Raub, W. (2002). Embedded trust: Control and learning. Advances in Group Processes, 19, 167–202. Caldwell, D., & Spivey, W. A. (1983). The relationship between recruiting source and employee success: An analysis by race. Personnel Psychology, 36, 67–72. Chatman, J. (1991). Matching people and organizations: Selection and socialization in public accounting firms. Administrative Science Quarterly, 36, 459–484. Chetkovich, C. A. (1997). Real heat: Gender and race in the urban fire service. New Brunswick, NJ: Rutgers University Press. Corcoran, M., Datcher, L., & Duncan, G. J. (1980). Information and influence networks in labor markets. In: G. J. Duncan & J. N. Morgan (Eds), Five Thousand American Families: Patterns of Economic Progress (pp. 1–38). Ann Arbor, MI: Survey Research Center, Institute for Social Relations, University of Michigan. Coverdill, J. E. (1998). Personal contacts and post-hire job outcomes: Theoretical and empirical notes on the importance of matching methods. Research in Social Stratification and Mobility, 16, 247–269. Coverdill, J. E., & Finlay, W. (1998). Fit and skill in employee selection: Insights from a study of headhunters. Qualitative Sociology, 21, 105–127. Datcher, L. (1983). The impact of informal networks on quit behavior. Review of Economics and Statistics, 65, 491–495. Decker, P. J., & Cornelius, E. T., Jr. (1979). A note on recruiting sources and job survival rates. Journal of Applied Psychology, 64, 463–464. Fernandez, R. M., Castilla, E. J., & Moore, P. (2000). Social capital at work: Networks and employment at a phone center. American Journal of Sociology, 105, 1288–1356. Fernandez, R. M., & Weinberg, N. (1997). Sifting and sorting: Personal contacts and hiring in a retail bank. American Sociological Review, 62, 883–902. Gannon, M. J. (1971). Sources of referral and employee turnover. Journal of Applied Psychology, 55, 226–228. Granovetter, M. (1995). Getting a job: A study of contacts and careers (2nd ed.). Chicago: University of Chicago Press. Granovetter, M., & Tilly, C. (1988). Inequality and the labor process. In: N. J. Smelser (Ed.), Handbook of Sociology (pp. 175–221). Newbury Park, CA: Sage. Grieco, M. (1987). Keeping it in the family: Social networks and employment chance. London: Tavistock. Gulati, R., & Wang, L. O. (2003). Size of the pie and share of the pie: Implications of network embeddedness and business relatedness for value creation and value appropriation in joint ventures. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 209–242). Oxford: Elsevier Science. Halaby, C. N. (1988). Action and information in the job mobility process: The search decision. American Sociological Review, 53, 9–25. Halcrow, A. (1988). Employees are your best recruiters. Personnel Journal, 67, 42–49.
318
KATHRYN M. NECKERMAN AND ROBERTO M. FERNANDEZ
Jovanovic, B. (1979). Job matching and the theory of turnover. Journal of Political Economy, 87, 972–990. Jovanovic, B. (1984). Matching, turnover, and unemployment. Journal of Political Economy, 92, 108– 122. Lin, N., Ensel, W. M., & Vaughn, J. C. (1981). Social resources and strength of ties: Structural factors in occupational status attainment. American Sociological Review, 46, 393–405. Marsden, P. V., & Hurlbert, J. S. (1988). Social resources and mobility outcomes: A replication and extension. Social Forces, 66, 1038–1059. Montgomery, J. D. (1991). Social networks and labor-market outcomes: Toward an economic analysis. American Economic Review, 81, 1408–1418. Mortensen, D. T. (1986). Job search and labor market analysis. In: O. Ashenfelter & R. Layard (Eds), Handbook of Labor Economics (pp. 849–919). Amsterdam: North Holland. Raub, W., & Weesie, J. (1990). Reputation and efficiency in social interactions: An example of network effects. American Journal of Sociology, 96, 626–654. Rees, A. (1966). Information networks in labor markets. American Economic Review, 56, 559–566. Reid, G. (1972). Job search and the effectiveness of job-finding methods. Industrial and Labor Relations Review, 25, 479–495. Saloner, G. (1985). Old boy networks as screening mechanisms. Journal of Labor Economics, 3, 255–267. Sicilian, P. (1995). Employer search and worker-firm match quality. Quarterly Review of Economics and Finance, 35(special issue), 515–532. Simon, C. J., & Warner, J. T. (1992). Matchmaker, matchmaker: The effect of old boy networks on job match quality, earnings, and tenure. Journal of Labor Economics, 10, 306–329. Stigler, G. J. (1962). Information in the labor market. Journal of Political Economy, 70, 94–105. Wanous, J. P. (1980). Organizational entry: Recruitment, selection, and socialization of newcomers. Reading, MA: Addison-Wesley. Wegener, B. (1991). Job mobility and social ties: Social resources, prior job, and status attainment. American Sociological Review, 56, 60–71. Wenger, E. (1998). Communities of practice: Learning, meaning, and identity. Cambridge: Cambridge University Press.
THE WITHIN-JOB GENDER WAGE GAP, SWEDEN 1970–1990 Trond Petersen, Eva M. Meyersson Milgrom and Vemund Snartland ABSTRACT We report three findings in a comprehensive study of hourly wage differences between women and men working in same occupation and establishment in Sweden in 1970–1990. (1) Within same occupation and establishment in 1990, women on average earn 1.4% less than men among blue-collar workers, 5.0% less among white-collar employees. This occupation-establishment level wage gap declined strongly from 1970 to 1978. (2) For white-collar employees, occupational segregation accounts for much of the wage gap, establishment segregation for little. For blue-collar workers both types of segregation are important. (3) The within-occupation gaps are small, below 4% and 7% for blue- and white-collar workers.
INTRODUCTION Wage differences between men and women caused by discrimination from employers can come about by several mechanisms. In a first, women are differentially allocated to occupations and establishments that differ in the wages they pay. This may involve discrimination in the matching process at the point of hire, in subsequent promotions, and in firings.1 In a second, women are paid lower wages
The Governance of Relations in Markets and Organizations Research in the Sociology of Organizations, Volume 20, 319–353 © 2003 Published by Elsevier Science Ltd. ISSN: 0733-558X/PII: S0733558X02200129
319
320
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
than men in the same job or occupation in the same establishment. This we refer to as “within-job wage discrimination.” In a third, female-dominated occupations receive lower wages than male-dominated ones, although skill requirements and other wage relevant factors are the same. This is the issue addressed by comparable worth initiatives. These mechanisms differ. It may be the case that men and women are paid the same when they do the same work for the same employer. It may also be the case that equally qualified men and women are hired and promoted at the same rates once they apply for the same jobs and seek the same promotions. It may just be the case that men and women rarely do the same work for the same employer and rarely apply for the same jobs or seek the same promotions, be that caused by preference, family constraints, or vestiges of past discrimination. Differences in pay could then arise, not due to discrimination in wage setting within jobs, or in hiring and promotions, but due to the jobs men and women seek and hold being paid at different rates. When those different rates cannot be justified by objective features of the jobs, such as skill requirements, some researchers argue there is comparable worth discrimination. One conjecture currently accepted by many researchers and policy makers is that wage differences are today less a question of within-job wage discrimination and more a matter of who gets which jobs and how female-dominated occupations are valuated in the market. For the U.S. case, but also with relevance for Europe and elsewhere, Treiman and Hartmann (1981, pp. 92–93) write: Although the committee recognizes that instances of unequal pay for the same work have not been entirely eliminated, we believe that they are probably not now the major source of differences in earnings.
Lazear (1991, pp. 13–14) provides the same sentiment: My view is that hiring is most important; promotion is second; and wages are third.
Petersen and Saporta (2003) outline reasons why within-job wage discrimination is unlikely to be a major source of the wage gap today, focusing on the ease of observing such discrimination, relative clarity of evidence, and availability of a complainant to press charges. But in order to confirm the accepted conjecture one needs unusual data: on men and women working in the same occupation or type of job in the same establishment, and on how their employers pay them, focusing on the possibly discriminatory acts. Other than in case studies of single establishments it is difficult to assemble such data and hence to assess the extent of within-job wage discrimination.
The Within-Job Gender Wage Gap, Sweden 1970–1990
321
Except for one study covering the U.S., as well as one study of Norway, it has not been established that men and women receive more or less equal pay within given jobs or occupations in given establishments, or that within-job wage discrimination is not important. Petersen and Morgan (1995) analyzed wage differences between men and women employed in the same detailed occupation and establishment, using data collected by the U.S. Bureau of Labor Statistics covering about 1.5 million employees in the period 1974–1983. Within given occupation-establishment units, wage differences were relatively small: on average women earned 1.7% less than men among blue-collar and clerical employees, while on average 3.1% less in seven professional and three administrative occupations. Hence, in the U.S. within-job wage discrimination is not a central source for the gender wage gap.2 In Norway, the situation in 1984 and 1990 was quite similar, with a within-job wage gap of about 3.3% and 4.5% among blue- and white-collar workers respectively (Petersen et al., 1997).3 Against this background and the results for the U.S. and Norway, this article therefore reports a comparative and similar but more comprehensive analysis of data from the private sector in Sweden. We use data on entire populations of establishments in several important sectors of the Swedish economy, covering about 40% of employees in the private sector. We have access to individual-level wage data at the occupation-establishment level, so that we can compare men and women working in the same occupation in the same establishment. We analyze data from six years in the period 1970–1990, including 1978 and 1980, one year before and after the passing of the equal pay act in 1979, thus being able to address some crucial historical trends.4 Sweden is distinctive and of general interest in several ways, including its compressed wage distributions and extensive family policies, on which we elaborate in the next section. With access to data for a 20-year period we may address changes over time and discuss how these might be related to institutional changes in the period. While our errand is empirical, aiming to document the situation for equal pay for equal work in Sweden, the issues addressed have broader implications. The setting of wages within firms is central to governance of the employment relationship. The ability to recruit, retain, and motivate employees depends significantly on a firm’s wage structure. For gender, this gets strongly regulated by the law, frequently monitored by parties internal and external to the firm, perhaps being the single pay issue receiving the most attention the last thirty years. Few decisions within firms have the potential to engender as much conflict, especially when they lead to real or perceived gender inequality. The organizational governance issue then, in practically all Western nations, concerns how to pay men and women doing the same work. Our empirical analysis documents the wage outcomes of
322
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
this governance and thus the possible sources of wage inequality between men and women. Does it arise in the governance relationship that sets wages for the same work for the same employer? Or does it arise elsewhere, in the allocation to jobs through hiring and promotion and in setting of wages for female-dominated occupations? Knowing here exactly what is the case, rather than guessing at what might be the case, is essential for informing organizations what to do next in the area of gender equality and where the law possibly should be extended. Four issues will be addressed: (1) What is the wage gap at the occupation-establishment level? (2) What is more important for the overall wage gap, segregation on establishments or segregation on occupations? (3) What are the changes over a 20-year period in (1)–(2) above, including the role of changes in wage inequality? (4) How does what is reported in (1)–(2) compare to the U.S. and Norwegian experiences, the only countries with similar studies? In the following section, we outline the broader institutional features of the Swedish case, including a comparison to Norway and the U.S. In the two subsequent sections, we describe the data and the methods. A section on wages and positions addresses an important preliminary issue in the Swedish context: the extent to which there is variation in wages among employees once they work in the same occupation in the same establishment. Without such variation, there could not be a gender wage gap at the occupation-establishment level, and there would be no point to our analysis. Thereafter, the main results are presented. The last section provides conclusions and discussion.
INSTITUTIONAL FRAMEWORKS Sweden is in several ways a strategic research site for investigating these questions: Its equal pay laws, similar to those in other advanced contemporary societies, its compressed wage and salary distributions, and its extensive family policies. The latter two may impact the gender wage gap. We outline each of these institutional arrangements and make comparisons to the U.S. and Norway.
Equal Pay Legislation An Equal Pay Act was passed in 1979, becoming effective in 1980 (see SOU, 1993, pp. 49, 172). This was similar to the corresponding U.S. Equal Pay Act of
The Within-Job Gender Wage Gap, Sweden 1970–1990
323
1963 and Title VII of the 1964 Civil Rights Act. It also is similar to the Norwegian Equal Pay Act of 1978 and legislation in the European Union (Ellis, 1991; Petersen et al., 1997). The 1979 law made within-job wage discrimination illegal.5 But both before and after 1979 Sweden has pursued vigorous policies to diminish wage differences between men and women. As early as 1960 the Central Confederation of Employers (SAF) and the main labor union (LO) agreed to abolish all separate wage lists for male and female employees within their domain. This agreement was gradually implemented and completed by 1965 (see SOU, 1993, pp. 169, 264). The same process occurred in Norway, with agreement in 1961 and implementation between 1963 and 1967 (see NOU, 1997, p. 78). The comparison of conditions in the period 1970–1990 and especially of the years 1978 and 1980, immediately before and after passing of the equal pay act in 1979, will be particularly interesting. The sex equality laws in the three countries are thus similar. There are however several differences in the legal systems and enforcement of the laws. Perhaps most important is that penalties imposed on employers found guilty of discrimination are larger in the U.S. A major difference is that it is difficult or even impossible to bring class-action suits to the courts in Sweden and Norway, resembling the British system (NOU, 1997, pp. 90, 134; SOU, 1993, p. 54).
Wage Inequality Sweden is a society with strong egalitarian traditions, allowing for much less inequality in pay than the U.S., but more than Norway (see Fritzell, 1991).6 The countries are at opposite ends with respect to wage and income inequality. The distribution of market rewards before taxes may be more unequal in Sweden than Norway. But Sweden has a more progressive tax system, so that disposable income after taxes and transfers is more equal than in Norway.7 In their comparison of Japan, Sweden, and the U.S., Verba et al. (1987, p. 363) conclude: “the most egalitarian group in the United States favors a wider income gap than that favored by the most conservative group in Sweden.” The perhaps most clear expression of the aversion against inequality one finds in the system of solidaristic wage bargaining in Sweden. It was particularly strong in the 1950–1983 period, with attempts to minimize wage differences between groups as well instituting the principle of equal pay for equal work across firms and sometimes even equal pay for all (e.g. Edin & Richardsson, 2002). But since 1983, when the central bargaining system started to dissolve, there has been a move toward less solidaristic and less rigid wage policies (SOU, 1993, pp. 76–78).
324
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
We thus ascertain whether the Swedish aversion against inequality also translates into smaller wage differences between the sexes at the occupation-establishment level, addressing how the overall wage system intersects with gender inequality. But we also address how the decline in solidaristic wage bargaining impacted the wage gap at various levels: overall, within occupations, within establishments, and within occupation-establishment units.
Family Policies The Swedish concern for gender equality has had a pronounced impact in the area of family policies.8 Since the 1970s Sweden has had more extensive policies than any other country, with extensive maternity as well as paternity leave and universal provision of child care with a strengthening of policies since 1979 (see Kamerman, 1988, 1991).9 Norway has less extensive family policies than Sweden, and the U.S. is at the opposite end. As an example, take job-protection during absences in the period before and after childbirth. Norway and Sweden have had such policies for a long time, while the U.S. only since 1993 with the passage of the Family and Medical Leave Act (FMLA), requiring medium and large employers to provide 12 weeks of unpaid parental leave. Many employers provided such benefits voluntarily and some states had such provisions but nothing universal occurred before 1993, unlike most other industrialized countries. Family policies clearly facilitate combining careers with family and children, which helps women workwise and as such are important. But their effects on the gender wage gap are less clear, neither at the overall, occupation, or occupationestablishment level.
Implications of Institutional Frameworks These three broad institutional features – equal pay laws, wage inequality, and family policies – should in principle facilitate the position of women in employment and careers. There is however disagreement about the extent to which this has occurred in Sweden. Some scholars argue that there is a major impact of the equal pay laws, of solidaristic wage bargaining, and of family policies on women’s position in the labor market and the gender wage gap (Gustafsson & Lantz, 1985; L¨ofstr¨om, 1989, 1991). Others claim that changes in women’s relative wages to a large extent are unrelated to these institutional changes, stressing instead such factors as technical changes and the demand for work requiring various levels of qualification which in turn may affect wage dispersions and the wage gap (Svensson, 1992, 1995).
The Within-Job Gender Wage Gap, Sweden 1970–1990
325
With respect to the Swedish Equal Pay Act, its likely impact on the wage gap at the occupation-establishment level is to make it small, as in Norway and the U.S., both with comparable legislation. From an organizational governance viewpoint, the main difference arises in law enforcement, where in the U.S. penalties are higher and it is possible to bring class-action suits. In the area of gender inequality the U.S. simply has stricter external regulation of the firm. This may lead to more vigilance from employers concerning equal pay in the U.S. than in Sweden and Norway, perhaps resulting in a lower within-job wage gap in the U.S. With respect to the compressed wage distribution, it could have multiple and contradictory effects. One is that the overall wage gap is likely to be lower than elsewhere, given that women in most countries are concentrated in the lower-paying jobs. Blau and Kahn (1996) reported a lower overall wage gap in Sweden than in eight other Western countries. The lower inter-occupational wage inequality could lead to a smaller role for occupational sex segregation in explaining the gender wage gap. Another effect is that the lack of spread in pay between occupations also gives fewer incentives for women to move into more lucrative positions, potentially leading to more occupational sex segregation (Blau & Kahn, 1996, p. S40, Table 3). This would result in a bigger role of occupational sex segregation for the gender wage gap in Sweden, the opposite effect of the low inter-occupational wage inequality. Additionally, intra-occupational wage inequality is likely lower in Sweden. This could further increase the effect of occupational sex segregation on the gender wage gap. The compressed wage distribution could even have an impact on the wage gap at the occupation-establishment level. In contrast to the U.S., merit and individually negotiated pay are unusual. For example, all professorial salaries were identical within and between universities up until 1990. This could in turn result in a lower gap at the occupation-establishment level in Sweden. What about the possible effects of the extensive family policies? It is difficult to assess how such policies impact women’s employment and careers. The firstorder effect is straightforward. The policies make it easier to combine family and work, as when there is public provision of childcare and when jobs are protected during absences surrounding childbirth (B. Hoem, 1993). But such policies also change incentives for families, making it cheaper to have children. So there may be a second-order effect on fertility, which in turn interacts with employment and career opportunities for women. For example, following the Swedish maternity leave provisions introduced in 1985, which gave 72 weeks of leave at 90% pay, there was a sharp increase in fertility, with Sweden for a period having the highest fertility rate in Western Europe (e.g. J. M. Hoem, 1993). The lower costs of having children thus led to an increase in the number of children born per women, which in turn may have been detrimental to employment and career development. It
326
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
has in fact been suggested that the high level of maternity benefits is part of the reason why Sweden appears to have a more sex-segregated labor market than other countries (B. Hoem, 1995; Stoiber, 1990).
DATA The wage data were collected and compiled by the Swedish Employers’ Confederation (SAF), from their database on wage statistics, assembled from establishment-level personnel records. Compared to the Norwegian and the U.S. studies these data are even more extensive and detailed and contain information for all blue- and white-collar workers in every industry (except insurance and banking) in the private sector within the SAF domain. Member firms have been providing information to the database from 1970 to 1990, once or twice a year. The data have been used for inputs in the yearly wage negotiations and are monitored not only by SAF but also by the labor unions. They should be very reliable compared to standard sample surveys with personal reports of pay rates, hours worked, and occupation. The establishment characteristics include the following: detailed industry code; size (the number of employees); region and area within region. For each employee surveyed, information was obtained on age, hours worked, part- versus full-time employed, method of wage payment (incentive- or time-rated), union status, and a detailed description of job content, usually a four-digit code. We refer to the job content information as occupational codes, but for the blue-collar workers it could as well be called job titles. For white-collar employees we also know the length and type of education. The occupational codes for the blue-collar workers are industry specific and detailed, typically corresponding to the titles used in collective agreements. They comprise 674 titles. For the present purpose we have broken these down even further, distinguishing between jobs done during the day, on continuous shift, on night shift, etc., so as not to compare day with night workers, even though they may be in the same occupation. This yields altogether 1,849 occupations for the blue-collar workers. The white-collar occupations are less detailed. There are 51 broad occupational groups. Within each group a further distinction is made with respect to the level of difficulty in the job, from 1 (low) to 7 (high), which we refer to as ranks.10 Not all occupations span the entire 7 ranks, some start higher than rank 1 and some do not have the top ranks 5–7. The cross-classification of 51 occupational groups and 7 ranks yields 275–286 occupation-by-rank groups (varying by year), which we for short refer to as occupations. This so-called BNT-code was developed first
The Within-Job Gender Wage Gap, Sweden 1970–1990
327
in 1955 and has been revised several times since (SOU, 1993, p. 204). Its main purpose was to aid in the collection of wage statistics, not to set wages for jobs and individuals. It is not unlike the salary grade level structure used in many large U.S. organization (e.g. Spilerman, 1986), where a salary grade level indicates such things as the level of responsibility and qualification associated with the position, but without a strong tie between the grade level and the actual salary itself, though a clear correlation exists. The data cover practically the entire occupational spectrum, including managers and professionals. The Chief Executive Officer and members of executive teams are excluded. An overview of the data is given in Table 1.11 For 1990 we have information on 643,349 and 391,997 blue- and whitecollar employees respectively. Among the blue-collar workers there were 1,849 occupations, 23,544 establishments, and 87,640 occupation-establishment units. The wage data are reported in an unusually detailed manner. For each individual, the wages (as well as hours worked), are reported separately for those earned during regular hours and those earned during overtime hours. For blue-collar workers, the wages are given in hourly units, while for white-collar workers they are given as monthly pay, from which we get hourly wages by dividing by hours worked. The partition of the wage data into the part earned on regular hours and the part earned on overtime is very important, and unusual, as in the Norwegian data (Petersen et al., 1997). It makes the wage data less prone to bias than virtually every other study used for assessing wage discrimination. Men usually work more overtime hours than women, either due to preference or to better access to overtime hours, and overtime hours are usually paid at a higher rate. The present data do not conflate pay at regular and overtime hours, focusing only on wages during regular hours. Our information on how employers pay their employees is thus quite precise, not meshing labor supply and other adaptation by employees with pay rates actually paid to men and women. The three-tiered multilevel structure of the data is essential. First we have the establishment level. Second, within each establishment, we have the occupation level with information on all occupations, or in the white-collar data, the occupation-byranks. Third, within each occupation in each establishment, we have information on all employees, the individual level. This structure allows us to compute the wage gap at various levels. For example, at the establishment level, we can compare the wages of men and women working in the same establishment, but not necessarily in the same occupation. The most relevant part of our analysis compares men and women working in the same occupation and establishment, thus making so-called within-job comparisons.
328
Table 1. Documentation of Data for Blue- and White-Collar Workers, by Year, in Sweden. Nf
Nm
%f
No
Ne
Noe
Nb
w ¯
w ¯f
w ¯m
1
2
3
4
5
6
7
8
9
10
11
Blue-collar workers 1990 643,349 1985 626,601 1980 676,323 1978 646,466 1975 644,540 1970 583,963
188,540 179,235 185,648 167,589 171,183 139,146
445,809 447,366 490,675 478,857 473,357 444,817
29.7 28.6 27.4 25.9 26.6 23.8
1,849 2,070 2,482 1,926 1,832 1,438
23,544 24,165 24,916 23,939 19,290 18,049
87,640 89,334 95,917 94,401 86,227 80,592
23 22 22 20 18 19
64.10 44.60 29.15 26.05 19.02 11.25
58.99 41.08 26.70 23.79 17.21 9.70
67.69 46.01 30.07 26.83 19.68 11.74
White-collar workers 1990 391,997 1985 380,513 1980 381,702 1978 367,207 1975 351,459 1970 299,154
135,581 124,423 117,798 110,460 100,050 73,318
256,416 256,090 263,904 256,747 251,409 222,472
34.6 32.7 30.9 30.1 28.4 24.8
280 279 281 271 345 256
22,031 20,669 19,769 18,457 15,894 13,779
146,940 145,070 148,461 144,309 135,340 108,121
32 32 31 34 36 40
92.71 63.03 44.06 37.19 29.09 17.09
74.63 50.03 34.56 28.93 21.83 11.46
102.27 69.35 48.30 40.74 31.98 18.94
Sector
Note: N = total number of employees, N f = number of women, N m = number of men, %f = percent women, N o = number of occupations, N e = ¯ = mean wage, w ¯ f = women’s number of establishments, N oe = number of occupation-establishment units, N b = number of industries, w mean wage, and w ¯ m = men’s mean wage.
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
N
The Within-Job Gender Wage Gap, Sweden 1970–1990
329
METHODS We report the relative wages between men and women at various levels, with technical details given in the appendix. Separately for blue- and white-collar workers we first compute the average female wage as percentage of the average male wage, where, for example, the number 88% means that on average women earn 12% less than men. This we refer to as the relative wages. The relative wages we then decompose in four ways, separately for blue- and white-collar employees. We start by computing separately for each industry, occupation, establishment, and occupation-establishment unit the average female wage as percentage of the average male wage. This can only be done for industries, occupations, establishments, and occupation-establishment units that are sex integrated. For example, at the occupation-establishment level we compute the average female wage as percentage of the average male wage in each sex-integrated occupation-establishment unit. Next, we compute the average of these relative wages across the sex-integrated units within a level. For example, at the occupation-establishment level, we compute the average of the relative wages at that level across the sex-integrated occupation-establishment units. This gives what we also refer to as the withinjob relative wages. These computations give the average relative wages for each of four levels: industry, occupation, establishment, and occupation-establishment. The wage gap itself obtains as 100 minus the relative wages. We additionally report the percent of the raw wage gap explained separately by each of the four levels.12 The average wage gap at the occupation-establishment level gives an estimate of an upper bound on the amount of within-job wage discrimination, the quantity of greatest interest here. But also the within-occupation and within-establishment gaps are of interest, as they document the extent to which differential distribution of men and women on occupations and establishments can account for the overall gender wage gap. There is a question of what the appropriate weighting scheme is when computing the wage gap, especially at the occupation-establishment level. Ours is chosen primarily for conceptual reasons, not related to statistical or other considerations. For within-job wage discrimination, the relevant decision-making unit is the employer, who potentially discriminates or pays men and women different wages for the same work. The central goal of the analysis is to assess the extent to which the wage gap is due to such within-job wage discrimination, by employers against men and women in specific occupations. The relevant sampling unit or unit of analysis is therefore the match of the employer and an occupation, not a collection of employees. The central statistics is how employers on average pay men and women working in the same occupation for the same employer, describing what employers on average do within occupations, not what employees on average experience.13
330
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
Sampling employees, for example by a random sample of those in the labor market, conflates employee choices and adaptations with the actions taken by employers. Employees would obviously be a better sampling unit if the focus were on adaptations to family obligations rather than on potential discrimination by employers. For the white-collar employees, where we have information on several individual-level characteristics, we also report results from regression analysis, using so-called fixed-effects estimates. Here we control for the occupationestablishment unit by a dummy variable for each such unit, allowing the estimates to be interpreted as within-job differences. In addition to sex we control for educational level, age, and part-time status. The goal is to assess how the gender wage gap at the occupation-establishment level gets modified once we take into account the relevant individual-level characteristics. The technical details are given in equation (9) in the appendix.14
WAGES AND POSITIONS One issue requires attention before presenting our results. In Sweden, Norway and in many European countries, firm-internal wage structures are quite rigid: To each position a fixed wage or salary is often attached. This is foreign to practices in the U.S., where within given occupations or jobs for the same employer, there typically is a wide range in pay. One may thus object to our research that once we focus on wages at the occupation-establishment level there is by definition or by practice no variation in pay. Everyone will receive the same pay, so our analysis becomes tautological. We therefore report here the percentage range in wages at the occupation and the occupation-establishment levels. We first computed the percent by which the highest was above the lowest wage in each occupation and each occupationestablishment unit. Thereafter we took the average of this percent across all occupations and all occupation-establishment units.15 The results are reported in Table 2 , for blue- and white-collar workers in panels A and B respectively, separately for each year. Columns 1 and 2 give the average of the percentage ranges within occupations, first for all occupations and next for sex-integrated ones. Columns 3 and 4 give the same averages at the occupationestablishment levels, first for all and next for sex-integrated units. Table 2 shows in a striking way that there is considerable variation in wages at the occupation and the occupation-establishment levels, in all years. The variation is always larger at the occupation level. It is also larger in units that are sex integrated, at the occupation and the occupation-establishment levels. The range is three to twelve times larger at the occupation than the occupation-establishment level.
The Within-Job Gender Wage Gap, Sweden 1970–1990
331
Table 2. Average Sample Range (in Percent) of Hourly Wages within Occupation and Occupation-Establishment for All and for Sex-Integrated Units, Sweden. Sample range (in percent) Occupation
Number of units (N)
Occup.-estab.
Occupation
Occup.-estab.
All
Integ.
All
Integ.
All
Integ.
All
1
2
3
4
5
6
7
8
110.06 83.67 110.28 118.71 208.33 159.07
19.94 16.19 17.37 17.09 27.29 29.74
28.10 23.49 25.00 25.89 42.95 49.46
1,728 1,911 2,209 1,762 1,669 1,329
1,202 1,247 1,182 990 936 745
54,933 54,870 59,187 57,647 53,517 50,116
16,704 14,554 14,197 12,532 11,436 8,529
White-collar workers 1990 296.41 315.85 1985 195.81 206.79 1980 204.46 222.05 1978 211.81 229.09 1975 217.24 247.67 1970 274.68 316.24
24.14 21.17 20.15 20.59 24.49 31.82
34.71 30.65 29.02 29.08 34.58 53.27
276 275 276 271 336 256
251 246 232 225 263 191
58,341 56,431 56,831 54,546 50,612 40,747
16,416 13,628 11,890 10,971 9,907 7,733
Sector
Blue-collar workers 1990 93.96 1985 70.69 1980 80.06 1978 90.44 1975 146.22 1970 128.48
Integ.
Note: The figures represent the average percentage ranges of wages at the occupation and occupationestablishment levels, calculated for occupations and occupation-establishment units with two or more employees, separately for all units and for sex-integrated units. We first computed how many percent the highest wage was above the lowest wage in each occupation and each occupationestablishment unit. Thereafter we took the average of this percent across all occupations and all occupation-establishment units. Consider white-collar workers in 1990, with 276 different occupations employing two or more persons (see column 5). The average sample range (in percent) for these 276 occupations is 296.41 (column 1). ‘Occup.-estab.’ stands for occupationestablishment and ‘Integ.’ stands for integrated, meaning sex integrated.
At the latter level, the most relevant here, the average of the percentage range among blue-collar workers in 1990 was 20% across all units, while 28% among sex-integrated units. The corresponding numbers among white-collar employees were 24% and 35%. The highest paid person on average earned 20–35% more than the lowest paid, a considerable range, in principle allowing a wage gap at the occupation-establishment level as big as the overall gap in the labor market. The overall wage gap could thus have been caused by wage differences at the occupation-establishment level. The range in wages is higher among white- than
332
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
blue-collar workers, perhaps reflecting the coarser occupational titles among the former. At the occupation-establishment level, the range among blue-collar workers declined from 1970 to 1985, but increased between 1985 and 1990, when it reached a higher level than in 1978 but lower than in 1970. The pattern is similar at the occupation level, except that there was an increase in the range between 1970 and 1975. Among white-collar workers the range at the occupation-establishment level declined from 1970 to 1978, was stable from 1978 to 1985, but then increased between 1985 and 1990, when it reached a higher level than in 1975 but not as high as in 1970. At the occupation level, the pattern over time is much the same, decreasing from 1970 to 1985, when it reached its lowest level, but then increasing from 1985 to 1990, when it reached as high or a higher level than in 1970. At both levels and for both groups of employees, there was thus first a decrease and then an increase in the range of wages. As noted in the section on institutional frameworks, solidaristic wage bargaining broke down in 1983. The higher amount of wage dispersion in the second half of the 1980s most likely reflects this. We have shown that there is variation in pay at the occupation-establishment level also in Sweden. Our detailed occupational titles are thus not synonymous with wage levels. The analysis that follows is hence not tautological.
THE WITHIN-JOB WAGE GAP The Wage Gap in 1990 Earlier investigations have shown that the average wage of females is 20–25% below that of men in Sweden (e.g. Chen & Edin, 1994; Edin & Richardsson, 2002; Gustafsson, 1988; le Grand, 1991; Rosenfeld & Kalleberg, 1991; SOU, 1993; Westerg˚ard-Nielsen, 1994). We focus on employees in the private sector, showing a similar gap: In 1990, women on average earned 12.8% and 27.8% less than men among blue- and white-collar workers, respectively. Tables 3 and 4 report averages of the relative wages as well as various measures of dispersion for blue- and white-collar workers respectively, for each of six years between 1970 and 1990. For each year, column 1 gives the average female wage as a percentage of average male wage: overall (line 1), and by industry, occupation, establishment, and occupation-establishment respectively (lines 2–5). Equations (1)–(5) in the appendix were used for computing the ratios. Column 2 gives in lines 2–5 the percentages of the raw wage gap explained by industry, occupation, establishment, and occupation-establishment respectively, from equation (6) in the appendix. Columns 3–5 give standard deviations and minimum and maximum
The Within-Job Gender Wage Gap, Sweden 1970–1990
333
Table 3. Women’s Wages Relative to Men’s (in Percent), for Blue-Collar Workers, by Overall, Industry, Occupation, Establishment, and Occupation-Establishment, Sweden. Year
Mean
% Ex.
St. dev.
Min.
Max.
N
Women
Men
6
7
8
1
2
3
4
5
1990 Overall Industry Occupation Establishment Occ.-establishment
87.16 90.10 96.61 95.84 98.63
22.9 73.6 67.6 89.3
4.22 8.77 12.46 9.72
80.95 55.37 24.27 35.19
95.59 171.80 189.23 224.13
643,349 22 1,202 9,808 16,704
188,540 188,540 188,117 177,323 153,375
445,809 442,289 415,201 333,895 220,454
1985 Overall Industry Occupation Establishment Occ.-establishment
89.28 90.40 97.05 95.88 99.07
10.5 72.5 61.6 91.3
5.75 7.42 11.03 8.91
71.77 49.62 34.48 31.05
96.47 138.55 206.56 226.66
626,601 21 1,247 9,353 14,554
179,235 179,235 178,460 165,884 138,063
447,366 443,466 407,099 325,987 202,572
1980 Overall Industry Occupation Establishment Occ.-establishment
88.78 89.32 96.01 94.16 98.24
4.8 64.4 48.0 84.3
5.59 8.55 11.47 9.51
73.73 66.53 36.17 31.24
95.45 232.98 380.37 341.50
676,323 20 1,182 9,257 14,197
185,648 185,648 184,355 170,800 136,757
490,675 484,193 433,273 350,908 211,518
1978 Overall Industry Occupation Establishment Occ.-establishment
88.67 88.45 96.23 93.97 98.05
– 66.7 46.8 82.8
7.75 8.58 12.58 9.27
63.34 63.74 28.44 38.55
95.53 172.90 693.29 334.51
646,466 19 990 8,738 12,532
167,589 167,589 166,308 154,074 118,961
478,857 475,750 415,470 334,167 193,142
1975 Overall Industry Occupation Establishment Occ.-establishment
87.45 87.72 94.44 90.19 96.76
2.2 55.7 21.8 74.2
6.60 11.96 13.23 12.72
70.88 44.63 17.35 17.27
94.75 202.92 341.85 553.84
644,540 17 936 7,505 11,436
171,189 171,183 169,293 159,245 120,344
473,357 470,229 428,611 347,673 196,814
1970 Overall Industry Occupation Establishment Occ.-establishment
82.61 81.71 92.48 86.57 94.91
– 56.8 22.8 70.7
8.41 11.16 15.74 14.96
61.46 54.68 14.53 38.24
80.70 149.94 243.38 268.15
583,963 19 745 6,044 8,529
139,146 139,146 137,076 127,044 89,768
444,817 444,817 392,644 293,349 150,766
Note: The numbers obtain as follows. The raw relative wages is reported in the first line and obtains as women’s average wages as a percentage of men’s average wages. The raw relative wages is then decomposed in four ways. First, for line 2, we calculate, separately for each industry, women’s average wages as a percentage of men’s average wages. These percentages may only be calculated for integrated industries, where both men and women work. Second, we compute the mean of these percentages, which gives the number presented in line 2. Analogous with the industry numbers we calculate by occupation in line 3, by establishment in line 4, and by occupation-establishment in line 5. ‘% Ex.’ denotes the percent explained of the raw wage gap by each of the four levels separately, while ‘St. dev.’, ‘Min.’, and ‘Max.’ denote the standard deviation, the minimum, and maximum values respectively of the numbers used to compute column 1.
334
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
Table 4. Women’s Wages Relative to Men’s (in Percent), for White-Collar Workers, by Overall, Industry, Occupation, Establishment, and Occupation-Establishment, Sweden. Year
Mean
% Ex.
St. dev.
Min.
Max.
N
Women
Men 8
1
2
3
4
5
6
7
1990 Overall Industry Occupation Establishment Occ.-establishment
72.97 72.97 93.24 75.43 95.00
– 75.0 9.1 81.5
3.91 8.00 15.11 13.93
63.29 46.12 27.45 21.86
85.04 147.30 258.60 256.40
391,997 32 251 15,002 16,416
135,574 135,574 135,567 130,274 52,157
256,398 256,398 254,361 246,355 92,606
1985 Overall Industry Occupation Establishment Occ.-establishment
72.14 73.55 93.91 75.70 95.49
5.1 78.1 12.8 83.8
3.57 7.78 13.76 13.03
66.99 61.16 24.43 14.15
81.33 134.60 180.10 219.90
380,513 32 246 13,767 13,628
124,422 124,422 124,375 119,888 41,384
256,090 256,090 252,956 244,856 77,172
1980 Overall Industry Occupation Establishment Occ.-establishment
71.56 73.41 93.10 75.58 95.51
6.5 75.7 14.1 84.2
4.45 8.19 13.04 12.59
65.11 25.91 28.63 28.63
84.43 117.30 221.00 216.10
381,702 31 232 13,319 11,887
117,783 117,783 117,774 114,234 35,744
263,894 263,894 257,335 252,338 66,594
1978 Overall Industry Occupation Establishment Occ.-establishment
71.01 72.73 92.64 74.57 95.97
5.9 74.6 12.3 86.1
3.87 7.53 13.02 13.06
65.30 52.79 24.80 26.70
83.91 121.10 250.00 234.80
367,207 34 225 12,263 10,970
110,741 110,741 110,741 107,654 33,431
257,492 257,492 247,167 244,926 59,271
1975 Overall Industry Occupation Establishment Occ.-establishment
68.26 70.57 92.07 71.96 94.61
7.3 75.0 11.7 83.0
4.43 9.22 13.64 14.82
53.89 41.19 21.04 32.36
77.20 131.60 263.80 246.40
351,459 36 263 10,901 9,896
101,184 110,184 101,125 98,317 30,262
255,304 255,304 247,031 242,336 55,593
1970 Overall Industry Occupation Establishment Occ.-establishment
60.94 62.92 89.35 62.56 89.85
5.1 72.7 4.1 74.0
3.98 11.15 15.41 18.60
53.52 50.58 15.89 37.06
74.35 140.20 212.10 272.20
299,154 40 191 8,605 7,646
72,217 72,217 72,217 70,765 28,230
222,103 222,103 206,587 209,628 41,866
Note: The numbers obtain as follows. The raw relative wages is reported in the first line and obtains as women’s average wages as a percentage of men’s average wages. The raw relative wages is then decomposed in four ways. First, for line 2, we calculate, separately for each industry, women’s average wages as a percentage of men’s average wages. These percentages may only be calculated for integrated industries, where both men and women work. Second, we compute the mean of these percentages, which gives the number presented in line 2. Analogous with the industry numbers we calculate by occupation in line 3, by establishment in line 4, and by occupation-establishment in line 5. ‘% Ex.’ denotes the percent explained of the raw wage gap by each of the four levels separately, while ‘St. dev.’, ‘Min.’, and ‘Max.’ denote the standard deviation, the minimum, and maximum values respectively of the numbers used to compute column 1.
The Within-Job Gender Wage Gap, Sweden 1970–1990
335
values for the numbers that were used to compute the figures in column 1. Column 6, denoted N, gives in line 1 the total number of employees in the year, while lines 2–5 give the number of sex-integrated units, that is, the number of units where both men and women are employed in the same industry, same occupation, same establishment, and same occupation-establishment respectively. The number of women and men used for computing the ratios in column 1 are given in columns 7–8. To illustrate, consider the occupation-establishment level (line 5) for blue-collar workers in 1990. There are 16,704 sex-integrated occupation-establishment units (column 6), employing 153,375 women and 220,454 men (columns 7–8), a total of 373,829 workers. From line 1 column 6 we see further that there were 643,349 blue-collar workers in 1990. Hence, a total of 269,520 (= 643,349–373,829) or 42% of the workers are excluded from the computation of the wage gap at the occupation-establishment level because they worked in units that were entirely sex segregated. The total number of occupation-establishment units is 87,640 (Table 1, column 7), but only 16,704 are sex integrated (Table 3, line 6, column 6). Focusing on blue-collar workers in 1990, there are three striking results in Table 3. The first is that the wage gap is quite small when we compare men and women working in the same occupation for the same employer: 1.4%. The second result is that occupational segregation is somewhat but not much more important for the gender wage gap than establishment segregation, meaning that also differential allocation of men and women on establishments accounts for a portion of the wage gap. At the occupation and establishment levels the gaps were 3.39% and 4.16% respectively, small differences. The percentages of the gap explained by occupational and establishment segregation are 73.6 and 67.6 respectively. This is similar to results for blue-collar workers in Norway: Occupation explained 93% and establishment 91% of the wage gap (Petersen et al., 1997). The low gap at the establishment level and the corresponding high percentage of the wage gap explained by establishment segregation may in part be due to a narrow range of blue-collar occupations represented in each establishment. The third result is that the within-occupation gap is relatively small, less than 4%. This reflects that within an occupation, wage levels are rather uniform across establishments. So even if men and women are differentially distributed on establishments, this does not necessarily translate into a large wage gap as long as occupation is held constant. We return to this in our concluding discussion. Turning to the white-collar employees, Table 4 shows in 1990 a wage gap at the occupation-establishment level of 5.0%, with an overall gap of 27.8%, both bigger than among blue-collar workers. The occupational classification among white-collar employees is coarse: 280 positions account for the entire occupational spectrum across diverse industries. Among white-collar workers the role of occupational segregation is more important than among blue-collar workers,
336
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
whereas establishment segregation is of little importance. At the occupation level the gap is also small, less than 7%. The conclusion is straightforward: Within-job wage discrimination is in Sweden, as in Norway and the U.S., not a central force in explaining the gender wage gap.
Changes over Time in the Wage Gap For Sweden we have a consistent and long time series for the wage gap, for every five years from 1970 to 1990 plus for 1978, the year prior to the passing of the Swedish equal pay act in 1979. Focusing first on blue-collar workers, it is striking that the major changes in the occupation-establishment level wage gap occurred between 1970 and 1978, when it dropped from about 5% to its current level of about 1.5%, with a small but steady decline from 1978 to 1990. Much the same is the case for white-collar workers, where the major drop occurred between 1970 and 1975, except that here the occupation-establishment gap never became as small as among blue-collar workers. As argued above this is most likely an artifact of the data; the occupational classification is too crude. Thus at the occupation-establishment level, the major changes in the wage gap occurred prior to passing of the law making within-job wage discrimination illegal. One may speculate whether employers adapted to anticipated changes in legal environments, coming into compliance with the law prior to its implementation. At the occupation and establishment levels, the wage gap also declined from 1970 to 1990. Among blue-collar workers the decline was from 7.5% to 3.4% at the occupation and from 13.4% to 4.1% at the establishment level, with corresponding declines among white-collar employees from 10.6% to 6.8% and from 37.4% to 24.6%.
Regression Analysis of the Wage Gap Table 5 reports estimates of the sex effect from so-called fixed-effects regression analysis among white-collar employees, where the fixed-effect pertains to the occupation-establishment unit. We include controls for three individual-level characteristics in addition to sex: educational level, age, and part-time status. See equation (9) in the appendix. In the analysis, only occupation-establishment units that are sex-integrated contribute toward estimating the regression coefficients. The results are comparable to those obtained in the last line in each year of
The Within-Job Gender Wage Gap, Sweden 1970–1990
337
Table 5. Fixed-Effects Regression Estimates of the Effect of Being Female on Wages, Controlling for Educational Level, Age, and Part-Time Status, among White-Collar Workers by Year (Estimated Standard Errors in Parentheses). Year
1990 1985 1980 1978 1975 1970
Fixed-effects
From fixed-effects estimatesa
Estimates
Wage penalty
1
2
−0.0460 (0.0011) −0.0423 (0.0011) −0.0433 (0.0013) −0.0397 (0.0014) −0.0504 (0.0016) −0.0794 (0.0022)
−4.50 −4.14 −4.24 −3.89 −4.92 −7.63
Relative wages 3 95.50 95.80 95.76 96.11 95.08 92.34
From Table 4b Wage penalty 4 −5.00 −4.51 −4.49 −4.03 −5.39 −10.15
Relative wages 5 95.00 95.49 95.51 95.97 94.61 89.85
Note: The results are from fixed-effects regression estimates with controls for sex, educational level (one continuous variable from a low of 1 to a high of 7), age represented by age and age-squared, parttime status (one dummy variable), and dummy variables for each occupation-establishment unit. The part-time status variable is not included in the regression for 1970 because the information on it in that year was not reliable. The dependent variable is the natural logarithm of wages. Only sex-integrated occupation-establishment units are included in the analysis. If one includes in the regression computations all occupation-establishment units, not only sex-integrated units, with within-unit variation in at least one of the four individual-level characteristics, this changes the estimates only marginally. In five of the six years the sex effect then becomes marginally smaller, with estimates from 1990 to 1970 of: −0.0456 (0.0011), −0.0426 (0.0012), −0.0418 (0.0013), −0.0385 (0.0013), −0.0484 (0.0015), −0.0781 (0.0021). a Column 3 obtains as exp (b ) × 100, where b FE FE is the fixed-effect estimate from column 1. Column 2 obtains as 100 minus the result in column 1. b Column 5 comes from column 1 in Table 4, from the last line in each year. Column 4 obtains as 100 minus the result in column 5.
column 1 of Table 4, but with the difference that three individual-level characteristics have been controlled. In 1990, the estimate of the sex effect is −0.0460. This means that after controlling for educational level, age, and part-time status, women on average earn roughly 4.50% less than men working in the same occupation-establishment unit, about the same gap as without control variables.16 The results are similar for the other years. Only in 1970 did the gap decrease by several percentage points by taking into account the individual-level characteristics, from 10.15% to 7.63%. As in Table 4, the gap was lowest in 1978, of 3.89%, down from 4.03% when no control was made for the individual-level characteristics. The results say nothing about gender differences between men and women working in different occupation-establishment units.
338
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
Taking into account the three individual-level characteristics education, age, and part-time status, does reduce the gender wage gap at the occupation-establishment by about a quarter to half a percentage point, in 1990 from 5.00% to 4.50%. This is a reduction, but not a big one. The gap at the occupation-establishment level is thus not much further reduced by introducing the three individual-level characteristics.
Wage Gap by Rank within Occupation One could object that the wage gap we report is small because most of the occupation-establishment units pertain to relatively low-level jobs. In these there typically is little wage variation and the opportunities for paying men and women differentially are limited, with wages often set by collective bargaining. But as employees climb occupational ladders, wage dispersion within a rank increases, and the higher on the ladder the better are the opportunities for paying men and women unequally. This objection is relevant primarily for white-collar employees, for whom clear career ladders exist. For these employees we can in fact explore the conjecture. Recall that in Table 4 we first computed the relative wages separately for each sex-integrated occupation-establishment unit, where an occupation was defined as a given rank within each of the 51 broad occupational groups, in 1990 yielding 280 occupations. Then we took an average of the relative wages across all sexintegrated occupation-by-rank-establishment units. The resulting gap should thus really have been referred to as the occupation-by-rank-establishment level gap, which we now will do. In Table 6 we report the average of the relative wages at the occupation-by-rank-establishment level separately by rank for each of the 51 broad occupational groups. We do so only for the year 1990, since a single year fills up a whole table alone. In 1990, the cross-classification of 51 occupational groups and 7 ranks yields 280 occupation-by-rank positions, less than 357 since not all occupations span the entire rank system (see Petersen & Meyersson, 1999, Table 2). Four of these 280 positions have only incumbent, while another 25 are totally sex segregated. Among the 251 sex integrated occupation-by-rank units, 229 are sex integrated also at the establishment level. The 77 empty cells in the table correspond to the cases where the specific combination of occupation-by-rank does not exist. For example, occupation 14 has all ranks 1–7, whereas occupation 18 only has ranks 2–5. The 51 cells with a dash, –, cover the cases where all the occupation-by-rankestablishment units, at that rank for that occupation, are totally sex segregated. Three of the 51 broad occupational groups are completely sex segregated at each
The Within-Job Gender Wage Gap, Sweden 1970–1990
339
Table 6. Women’s Wages Relative to Men’s (in Percent) at the Occupation-by-Rank-Establishment Level, by Occupation and by Rank within Occupation, for White-Collar Employees, Sweden 1990. Occupation
By rank Overall
Administrative work 01 General analytical work 02 Secretarial work 03 Administrative efficiency improvement and development 04 Applied data processing, systems anal. and progr. 05 Applied data processing operation 06 Key punching
92.59 102.23 94.65
Production management 07 Administration of local plants and branches 08 Management of production, transp. and maintenance 09 Work supervision within production, repairs, etc. 10 Work supervision within building and construction 11 Administration, prod., superv. In forestry etc.
– 96.87
94.03
2
3
4
5
6
7
104.29
93.55
92.89
90.19
92.15
115.56
105.19 90.50
103.01 94.24
105.11 93.89
90.96 97.72
89.74 94.15
–
100.15
96.10
93.82
91.55
94.86
90.42
85.49
94.49
97.73
96.47
97.38
99.70
98.15
99.57
103.59
87.43
–
92.38
92.32
97.65
95.83
95.20
93.60
94.86
96.07
92.23
94.55
89.44
90.92
–
94.76
89.07
–
100.80
96.61
98.18
91.20
99.28
96.34
98.01
99.20
97.80
91.14
–
91.62
–
– 69.23
93.82
105.58
87.50
93.08
98.94
–
–
96.33
–
90.97
80.35
–
96.28
93.31
97.65
93.43
98.14
93.63
93.52
92.38
100.08
103.08
97.36
95.10
97.70
93.81
–
98.08 93.25
102.19 95.33
102.73 95.36
96.77 90.67
96.40 90.22
94.92 –
–
95.49 –
100.74 –
96.66 –
90.46 –
92.49
97.24
–
93.09 102.37
88.93 –
–
93.13 93.98 98.16
94.56 90.70 –
– – –
Research and development 12 Mathematical work 95.83 and calculation methodology 13 Laboratory work 95.92 Construction and design 14 Mechanical and electrical design engineering 15 Construction and construction programming 16 Architectural work 17 Design, drawing and decoration 18 Photography 19 Sound technology
1
95.83
–
Technical methodology, planning, control, service and industrial preventive health care 20 Production 94.63 95.64 95.56 93.74 engineering 21 Production planning 94.78 90.80 95.95 95.23 94.16 22 Traffic and 95.81 89.96 96.31 95.61 93.97 transportation planning 23 Quality control 94.24 103.96 96.07 92.66 95.50 24 Technical service 91.45 90.38 91.85 90.58 25 Industrial preventive 95.44 95.35 99.06 91.59 health care
– –
340
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
Table 6. (Continued ) Occupation
By rank Overall
1
2
3
4
5
6
83.20
98.97 101.45
98.29 98.60
90.34 95.51
97.93 98.46
99.30
93.88
93.67
101.66
99.49
100.37
105.49
88.66
98.09 100.10
98.11 98.97
95.54 96.57
93.48 98.51
90.40 106.52
100.98 –
221.54
95.68
94.82
101.82
96.64
82.55 94.62
84.21 92.67 95.41
98.85 89.71 92.91
91.09 90.63
95.16 105.23
98.77
97.94 93.70
94.34 97.90
93.11 106.43
98.31
93.54
93.71 90.95 96.13
75.48 93.45 95.06 –
97.27 89.16 95.08 –
– 94.23 88.51 –
– – 101.32 –
Communication, library, and archival work 26 Information work 95.44 27 Editorial work, 98.10 publishing 28 Editorial work, 94.76 technical information 29 Library, archives and 100.95 99.37 documentation Personnel work 30 Personnel service 31 Planning of education, training and teaching 32 Medical care within industries General services 33 Restaurant work 34 Marketing and sales 35 Sales within stores and department stores 36 Travel agency work 37 Sales at exhibitions, spare part depots, etc 38 Customer service 39 Tender calculation 40 Order processing 41 Internal processing of customer requests 42 Advertising 43 Buying 44 Management of inventory and sales 45 Shipping and freight services
96.01 98.34
110.00
96.27
87.67 92.14 93.88
95.51 94.34 87.13 91.82 95.13 –
–
93.44 93.88 94.39
– 98.48
91.52 97.49 94.34
97.19 94.62 93.87
91.82 91.93 92.15
93.81 93.17 89.81
95.58 85.89 –
95.75
106.10
99.25
94.28
93.38
87.92
80.93
99.41
97.95
94.70
93.46
90.79
110.36
94.63
92.15
76.51
105.72 96.76 96.02 –
99.47 103.33 91.83
92.54 – 96.98
93.42
97.15
95.52
93.60
93.22
Financial work and office services 46 Financial 96.61 100.98 administration 47 Management of 95.69 – housing and real estate 48 Auditing 98.39 49 Telephone work 98.98 90.78 50 Office services 95.75 97.75 51 Chauffeuring – All
–
95.00
98.30
97.76
7 – – –
– –
85.00
–
93.22
–
– 94.22
94.76
Note: For description of data see Section 3. For description of procedures see Sections 4 and 6. The results were obtained by the same procedures as the results in column 1, line 5, of Table 4. The columns 1–7 give the wage gaps at the occupation-by-rank-establishment level, separately by rank within each occupation. The rank variable goes from a low of 1 to a high of 7, indicating roughly the level of difficulty of the position within the broader occupational group. The results in the column ‘Overall’ give the average of the gaps at the occupation-by-rank-establishment level, separately for each occupation. There are 280 combinations of the 51 occupations and 7 ranks, less than 357 (= 51 × 7), since not all occupations span the entire rank system. The 77 empty cells in the table correspond to cases where the specific combination of occupation-by-rank does not exist. For the 280 existing combinations of occupation and rank, the 51 cells with a dash, –, cover the cases where all the occupation-byrank-establishment units, at that rank for that occupation, are totally segregated by sex; 4 of them are so because they have only one incumbent. This leaves 229 (= 280 − 51) occupation-by-rank units that are sex integrated at the occupation-by-rank-establishment level. The last line, in the column ‘Overall’, gives the average of the occupation-by-rank-establishment level gap, taken from Table 4, line 5, for year 1990. The last line, in the columns 1–7, gives the average of the occupation-by-rank-establishment level gap separately by rank.
The Within-Job Gender Wage Gap, Sweden 1970–1990
341
rank once one goes down to the occupation-by-rank-establishment level (Nos. 19, 41, and 51). For other occupations, there is complete sex segregation at some but not all ranks at the occupation-by-rank-establishment level. For example, occupation 12 has all ranks, but ranks 1 and 7 are totally sex segregated. There are clear variations in the wage gap between the 51 broad occupational groups. There are also variations across the ranks within an occupational group. There is however no systematic pattern to this variation, with the wage gap going up or down with the rank. The last line in the table partially documents this. It gives, separately for each rank, the average of the relative wages at the occupationby-rank-establishment level. The wage gap increases from rank 1 to 5, but is then smaller in ranks 6 and 7. This we have also verified in a regression analysis of the wage gap on dummy variables for occupation and rank, where the dependent variable then is the cell values in Table 6. The relatively small gap of about 5% we report for white-collar employees is thus not due to lower-level jobs being most highly represented in the data. The largest wage gaps, of 6–7%, are quite close to average gap in Table 4.
DISCUSSION Summary of Results from Sweden We have reported a large-scale empirical investigation of wage differences between men and women working in the same occupation for the same employer in Sweden in the period 1970–1990, providing comparative results to an earlier U.S. (Petersen & Morgan, 1995) as well as Norwegian study (Petersen et al., 1997). We now summarize the Swedish results, then compare them to similar results for the U.S. and Norway, and finally discuss the implications. We report three striking results, focusing on the year 1990. The first and central finding is that the wage gap is small when we compare men and women working in the same occupation for the same employer: Within-job wage discrimination is not a central cause of the wage gap and has not been so for 20 years. At the occupation-establishment level, the gap is lowest among blue-collar workers, about 1.4%, while it is larger among white-collar workers, about 5%. This larger gap probably reflects the cruder occupational codes we have access to among white-collar employees. For the latter we also performed a so-called fixed-effects regression analysis of sex-integrated occupation-establishment units, taking into account three individual-level characteristics in addition to sex: educational level, age, and part-time status. This reduced the sex effect somewhat, in 1990 from 5.00 to 4.50%, with similar results in other years.
342
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
This first and central finding confirms the conjecture in the literature. Wage differences between men and women are no longer an issue primarily of differential pay in the same occupation and establishment (Lazear, 1991; Treiman & Hartmann, 1981). We underline that this result is novel. More specifically, as we now shall elaborate, it could not be obtained from standard regression analysis of data from national probability or similar samples. In the latter, the real mechanisms are obscured. One compares men and women working in broadly similar occupations but invariably in different firms. A substantial wage gap is usually found and the researcher often alludes or even concludes that there is wage discrimination, implicating that men and women doing the same work for the same employer are paid differentially. But no such conclusion can be made and the analysis is inconclusive. All they show is that men and women doing broadly similar kinds of work with about the same amount of education and experience tend to earn different wages. But this does not imply that the differential pay is the outcome of any given employer treating men and women differentially. Such results may as well reflect that men and women tend to work for different employers. The occupational classifications used are usually too broad to allow comparison of men and women doing similar work. The second result is that among white-collar employees, establishment segregation is dramatically less important for the gender wage gap than occupational segregation, each explaining 9% and 82% of the gap respectively, meaning that differential allocation of men and women on establishments does not account for a large portion of the wage gap. Among blue-collar workers, in contrast, establishment segregation is quite important for the wage gap: Once one takes into account the distribution of employees on establishments, the wage gap drops to about 4%. Men thus tend to work in establishments with somewhat higher wage levels. But note that this may also come about when high-paying establishments mostly employ people in high-paying occupations, so that after all, the observed establishment effect may be an occupational effect. The third result is that the wage gaps at the occupation level are small, less than 4% and 7% among blue-collar and white-collar employees respectively. This reflects that within an occupation, wage levels are rather uniform across establishments. Since the 1950s it has been an objective of the central labor union to make wages in a given line of work or occupation independent of who you are and where you work. With respect to the gender wage gap, it is thus less a matter of where women work than what they do. The low within-occupation wage gaps clearly reflect the effects of solidaristic wage bargaining: equal pay for equal work. We also compared conditions over a 20-year period, using a consistent timeseries for every five years from 1970 through 1990 plus for 1978, the year prior to the passing of the Swedish equal pay act.
The Within-Job Gender Wage Gap, Sweden 1970–1990
343
Focusing first on blue-collar workers, it is striking that the major changes in the occupation-establishment level wage gap occurred between 1970 and 1978, when it dropped from about 5% to its current level of about 1.5%, with a small but steady decline from 1978 to 1990. Much the same is the case for white-collar employees except that the occupation-establishment gap never became as small as among blue-collar workers. As argued above this is most likely an artifact of the data; the occupational classification is too crude. The results on changes over time in the various wage gaps allow one to speculate on the role of legal changes and labor market institutions. What is most striking is the stability since 1978 in the gaps at all levels: at the overall, industry, occupation, establishment, and occupation-establishment levels. The major changes in these gaps occurred between 1970 and 1975 or 1978, for white- and blue-collar workers respectively. This occurred several years prior to the passing of the gender equality law in 1979 and almost a decade before passing of additional family legislation advantageous to female careers. Legislation probably played a role. But as important may have been the drives from the labor unions and the employers’ associations in addressing gender inequality, starting already in 1960 for blue-collar workers and strengthening in 1974 for white-collar employees (Svensson, 1995, pp. 127–128). Perhaps equally surprising is that the widening of wage inequality that occurred in the second half of the 1980s, after the weakening of solidaristic wage bargaining in 1983, appears to have had little or no impact on the gender wage gap. The overall gap among blue-collar workers increased with 1.5 percentage points from 1980 to 1990, while the overall gap among white-collar employees decreased with the same amount. But at the other levels, and most distinctively at the occupation level, the gaps have been more or less unaffected by the increases in inequality. Thus, wage inequality as such need not be a threat against women’s relative economic position as long as it is accompanied by less segregation of men and women on occupations and industries. In Sweden that countereffect seems to have fully compensated for the increase in inequality. In the U.S. it even more than compensated in the same period (Blau & Kahn, 1997). Without such countereffects, increased wage inequality clearly will have effects on the overall gender wage gap.
Comparison to the U.S. and Norway Compared to the U.S., at the occupation-establishment level we found in Sweden about the same gender wage gaps among blue-collar workers but somewhat larger gaps among white-collar employees (see Petersen & Morgan, 1995). Compared to Norway, the gaps in Sweden are somewhat smaller but otherwise very similar (see Petersen et al., 1997). The similarity to the U.S. is surprising given the higher
344
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
flexibility of pay there than in Sweden, Norway and Europe elsewhere. One should expect that greater flexibility in pay would lead to larger within-job wage gaps. But that is not what we found, rather the opposite. This difference between the U.S. on the one hand and Sweden and Norway on the other, albeit small, may reflect several factors. One may be dissimilarities in the occupational classifications used in the countries. The Swedish and also Norwegian classifications may be based on broader categories, which in turn will translate into a larger gap. But the Swedish and Norwegian data also comprise a larger spectrum of the occupations than the U.S. data, where the gap may be higher in more managerial, administrative, and professional occupations. These issues are not easy to settle because detailed occupational classifications are difficult to compare across countries. A second reason may be that an Equal Pay Act has been in operation for a longer time in the U.S. than Sweden and Norway, so there has been more time to deal with this type of inequality. A third reason may be differences in the legal systems and legal cultures in the countries. The U.S. is more litigious putting employers at higher risk of being sued and hence possibly more on guard. This creates a legal climate where the costs of litigation and of subsequent conviction can be high for employers and hence the deterrents against discrimination are stronger. Between 1980 and 1991, Sweden had only one equal pay case concerning within-job wage discrimination tried in the Work Courts (see SOU, 1993, p. 49). It is worthwhile to speculate on the relative roles of occupations and establishments for the wage gap in the three countries. Doing so is not entirely straightforward because one needs a concept of importance. Ours is as follows. We want to assess the impact on the wage gap of a redistribution of employees on occupations and establishments. The computations and comparisons we make hold under two conditions. In the case of occupation, but with similar conditions in the case of establishment, we assume that: (a) men and women get to be equally distributed on occupations; and (b) the wage gap within each occupation is the same across all occupations, equal to the average wage gap across occupations.17 Occupation matters more for the gender wage gap in Sweden and Norway than in the U.S. This is perhaps surprising given the higher amount of inter-occupational wage inequality in the U.S. But there are at least two countervailing effects. One is that the amount of occupational sex segregation is lower in the U.S. (see, e.g. Blau & Kahn, 1996). In the U.S., women gained on men in the 1980s even though the amount of wage inequality increased in the period (Blau & Kahn, 1997). This occurred because women improved their position relative to men in the occupational wage hierarchy, gaining access to highly paid jobs in the professions and elsewhere. The lower degree of occupational sex segregation in the U.S. reduces the role of occupation in explaining the gender wage gap. Another countervailing
The Within-Job Gender Wage Gap, Sweden 1970–1990
345
effect is that wage inequality within occupations likely is higher in the U.S., which lessens the effect of inter-occupational inequality. This may also partially account for the considerably larger within-occupation gender wage gap in the U.S. than in Sweden. Regarding the relative importance of establishments versus occupations in explaining the wage gap, some informative juxtaposition between Sweden, Norway and the U.S. can be made, even though the data are not entirely comparable. The sector of blue-collar workers in Sweden and Norway is quite similar to the eleven U.S. manufacturing industries used in part of the analysis in Petersen and Morgan (1995, Table 2), where the latter mostly cover blue-collar workers. The percentages of the raw wage gap explained by occupation and establishment respectively are 73.6% and 67.6% in Sweden, 45.8% and 22.1% in Norway, and 47.4% and 27.0% in the U.S. (using an unweighted average across the 11 industries). In all three countries segregation on occupations is more important than on establishments, but both are more important in Sweden than in the two other countries. For white-collar employees, the situation is quite similar across the three countries. Occupation explains a major part of the wage gap and establishment very little, again with establishment segregation appearing to be somewhat more important in Sweden and Norway than in the U.S., when data on five white-collar industries are used for the U.S. (Petersen & Morgan, 1995, Table 2). In all three countries, a redistribution of men and women on occupations will have a larger impact on the overall wage gap than a redistribution on establishments. Among white-collar workers in Sweden in 1990, the overall wage gap would drop from 27.0% to 6.6% with a redistribution on occupations, while only to 24.6% with a redistribution on establishments. We were somewhat perplexed by the similarities between the countries in this respect, that establishment had about the same or even larger impact on the wage gap in Sweden and Norway than in the U.S. It may be instructive to reflect briefly on the source of our surprise here, though without being able to offer a resolution.18 Sweden and Norway are highly egalitarian societies: Wage differences between firms and sectors in the economy are quite restrained, where at times even the government interferes when some industry or set of firms moves too far out of alignment with other industries or firms, which can be done through the annual wage negotiations (for Norway, see Høgsnes, 1989; for Sweden, see Lindbeck, 1997). Under such institutional arrangements one would expect establishment segregation to be considerably less important for the gender wage gap in Sweden and Norway than the U.S., because similar structures are absent in the U.S. But that is not what we found. So although firms or establishments matter for the wage gap in the three countries, it really is occupations and the matching of occupations
346
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
and establishments that drive the gap. Surprisingly, with respect to the relative importance of occupations versus establishments for the gender wage gap, the same outcome is achieved in the three countries: In Sweden and Norway in part through concerted coordination attempting to minimize wage differences between establishments, in the U.S. through market competition. This clearly warrants further speculation, but not in the present article. Implications With respect to research and policy, the implications of our findings are straightforward. Both should focus less on within-job wage discrimination and more on addressing differential access to occupations and establishments, differential rates of promotion, and differential rates of pay for lines of work done primarily by women. The analysis of differential access requires addressing the hiring process, in terms of procedures for recruitment, for who receives offers and who does not, and for conditions offered among those who receive offers of employment, a process that hardly has been studied (see, e.g. Collinson et al., 1990; Fernandez et al., 2000; Fernandez & Weinberg, 1997; Granovetter, 1995; Petersen et al., 2000). The analysis of promotion processes is more developed (e.g. Rosenfeld, 1992; Spilerman, 1986), but has not been extensively studied, while the analysis of valuative discrimination, the differential pay in occupations held primarily by women, has been carefully addressed in a large number of studies in many countries (see, e.g. England, 1992). These two other forms of discrimination, allocative and valuative, are obviously harder to deal with, but that is where the highest payoff can be realized, which clearly is of importance for policy, but with relevance for research as well.
NOTES 1. We follow the convention that discrimination occurs when men and women with equal qualifications and productivities are treated unequally. 2. One careful and extensive study claims this is an underestimate, reporting instead wage gaps at the occupation-establishment level of about 12% (Bayard et al., 1999). The study does however have inexact measurements of central variables. Occupation and establishment are measured for 1990, with some but likely minor error. Hourly wages are measured for 1989, imputed from annual earnings divided by imputed annual hours worked, the latter imputed as weeks worked times usual hours worked per week; an imputation from an imputation. The 1989 earnings data may come from several jobs held at different and same times during the year. They need not reflect the pay rate of a given employer, but rather what a person earned from several different employers. They may also pertain to a job or
The Within-Job Gender Wage Gap, Sweden 1970–1990
347
jobs different from the job held in 1990. The measure of imputed annual hours worked is imprecise, from a question on number of weeks worked the previous year and usual hours worked per week. 3. An early U.S. study (Blau, 1977), close to the current in design, as well as to Petersen and Morgan (1995) and Petersen et al. (1997), found that in 11 clerical occupations, differences in men’s and women’s wages were larger between than within establishments. Of later studies the one most resembling our’s is Groshen (1991a). Unlike the current study, it does not present the amount of within-job wage differences between men and women (i.e. in same occupation and establishment). 4. There is a substantial literature addressing the gender wage gap in Sweden and Scandinavia (e.g. for Sweden, Gustafsson, 1988; le Grand, 1991; Rosenfeld & Kalleberg, 1991; see also SOU, 1993; for Scandinavia generally, Westerg˚ard-Nielsen, 1994; and for Norway, Birkelund, 1992). But no study for Sweden addresses the gap at the level done here: the occupation-establishment level. 5. The law was expanded in 1991 and 1994, but those are years not covered in our quantitative data analysis below. 6. In his comparative study of Denmark, Norway, and Sweden, Esping-Andersen (1985, pp. 174, 176, 323) makes clear that Norway has had an “exceptionally aggressive drive for equality” and provides evidence showing considerably less income inequality between occupational groups in Norway than in Denmark or Sweden (see also OECD, 1995a). 7. A comparative study of nine OECD countries finds that Sweden has the most egalitarian distribution of disposable income, followed by Norway, with the U.S. surpassed only by Switzerland in the amount of inequality (Fritzell, 1991, pp. 143–148, 174, Table 5.5). 8. The impact in the political sphere is also strong, with high levels of participation of women in government and political leadership. In 1990 the percentage of women in legislative assemblies was 38%, higher than in other countries (Phillips, 1995, p. 59). 9. Sweden saw several changes in parental-leave policies and child-care provisions over the period 1970–1990. For example, the total leave period after childbirth was seven months at 90% pay in 1975, increased to nine months in 1978, with fathers being able to share leave periods since 1974. In 1973, 11% of preschoolers had access to public child care, 38% in 1983, and 49% in 1988, at a subsidized rate (see Rønsen & Sundstr¨om, 1996). Fathers accounted for 7% of leaves taken in 1988 (OECD, 1995b; see also Haas, 1991), with much lower numbers in other European countries. 10. In the data the code went from 2 (high) to 8 (low), which we recoded to 1 (low) to 7 (high). 11. For more details on the white-collar occupations see Petersen and Meyersson (1999). 12. This is computed by first by taking the difference between the raw relative wages and the relative wages at the level in question, then dividing this difference by 100 minus the raw relative wages, and finally multiplying the ratio with 100, as in Petersen and Morgan (1995). See Eq. (6) in the appendix. 13. We have tried several alternative decomposition weights, with only negligible changes in the qualitative pattern of our results. For example, at the occupationestablishment level, we computed the average relative wages weighting the relative wages in each sex-integrated occupation-establishment unit with respectively the proportion of workers, of male workers, or of female workers that is employed in the integrated occupation-establishment unit. The basis for the proportion is the number of workers, of males, or of females employed in integrated occupation-establishment units.
348
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
14. In the regression analysis, each individual working in a sex-integrated occupationestablishment unit gets assigned the same weight in computing the sex effect, unlike the more straightforward descriptive results explained above where each sex-integrated occupationestablishment unit gets assigned equal weight. 15. We have made the same type of computations for the standard deviation and the coefficient of variation, with no differences with respect to the point made here. 16. The exact interpretation of the coefficient is that the average of the logarithm of wages is −0.0460 lower for females than males, which in the unlogged wages implies that the geometric mean of wages is 4.5% lower for females, not the average wages as incorrectly stated above. 17. For how to compute the overall wage gap after a redistribution of men and women on occupations but allowing for a wage gap that varies by occupation, see Eq. (8) in the appendix. 18. Groshen (1991b) found establishment generally to be more important than occupation in explaining wage variation in six U.S. industries, but she did not investigate variations in the gender wage gap.
ACKNOWLEDGMENTS This article is based on individual-level wage data made available by the main employer’s association in Sweden, the Swedish Employers’ Confederation (Svensk Arbetsgivarforening or SAF). We are grateful to Ari Hietasalo and Marianne Lindahl at SAF for their extensive and exceptionally expert cooperation in preparing these data for analysis. We thank the editors of Research in Sociology of Organizations for extensive comments. We also thank Karen Modesta Olsen, who helped greatly in this study and who participated and was a coauthor of the earlier Norwegian study (Petersen et al., 1997). An earlier version of the chapter was presented at SOFI, Stockholm University, Department of Sociology, Stockholm University, The Industrial Institute for Economic and Social Research, Stockholm, FIEF, Stockholm, Department of Economics, Uppsala University, and at the Swedish Employer’s Confederation, Stockholm. We are grateful for financial support from the Swedish Council for Research in the Humanities and Social Sciences (HSFR). This chapter is an extended version of Meyerson and Petersen (1997) and Meyersson Milgrom, Petersen, and Snartland (2001). It includes, with some changes and rewriting, many of the materials that appeared in the two earlier chapters.
REFERENCES Bayard, K., Hellerstein, J., Neumark, D., & Troske, K. (1999). New evidence on sex segregation and sex differences in wages from matched employee-employer data. Working Paper 7003, March 1999. Cambridge, MA: National Bureau of Economic Research.
The Within-Job Gender Wage Gap, Sweden 1970–1990
349
Birkelund, G. (1992). Stratification and wages. Acta Sociologica, 35, 46–62. Blau, F. D. (1977). Equal pay in the office. Lexington, MA: Lexington Books. Blau, F. D., & Kahn, L. (1996). Wage structure and gender earnings differentials: an international comparison. Economica, 63(supplement), S29–S62. Blau, F. D., & Kahn, L. (1997). Swimming upstream: Trends in the gender wage differential in the 1980s. Journal of Labor Economics, 15, 1–42. Chen, P., & Edin, P.-A. (1994). Gender wage differentials, discrimination and work effort across methods of pay. In: B. Holmlund (Ed.), Pay, Productivity, and Policy (chapter 3). Stockholm: Trade Union Institute for Economic Research. Collinson, D. L., Knights, D., & Collinson, M. (1990). Managing to discriminate. London: Routledge. Edin, P.-A., & Richardsson, K. (2002). Swimming against the tide: Solidarity wage policy and the gender earnings gap. Scandinavian Journal of Economics, 104, 49–67. Ellis, E. (1991). European Community sex equality law. Oxford: Oxford University Press. England, P. (1992). Comparable worth: Theories and evidence. Hawthorne, NY: Aldine de Gruyter. Esping-Andersen, G. (1985). Politics against markets. The social democratic road to power. Princeton, NJ: Princeton University Press. Fernandez, R. M., Castilla, E., & Moore, P. (2000). Social capital at work: Networks and employment at a phone center. American Journal of Sociology, 105, 1288–1356. Fernandez, R. M., & Weinberg, N. (1997). Sifting and sorting: Personal contacts and hiring in a retail bank. American Sociological Review, 62, 883–902. Fritzell, J. (1991). Icke av marknaden allena: Inkomstf¨ordelningen i Sverige. Stockholm, Sweden: Almqvist & Wiksell International. Granovetter, M. (1995). Getting a job. A study of contacts and careers (2nd ed.). Chicago: University of Chicago Press. Groshen, E. L. (1991a). The structure of the female/male wage differential: Is it who you are, what you do, or where you work? Journal of Human Resources, 26, 457–472. Groshen, E. L. (1991b). Sources of intra-industry wage dispersion: How much do employers matter? Quarterly Journal of Economics, 106, 869–884. Gustafsson, S. (1988). L¨oneskillnader mellan kvinnor och m¨an – gapet o¨ kar igen. Ekonomisk Debatt, 16, 209–215. Gustafsson, S., & Lantz, P. (1985). Arbete och l¨oner. Ekonomiska teorier och fakta kring skillnader mellan kvinnor och m¨an. Stockholm, Sweden: The Research Institute for Industrial Economics and ALC. Haas, L. (1991). Equal parenthood and social policy: Lessons from a study of parental leave in Sweden. In: J. S. Hyde & M. J. Essex (Eds), Parental Leave and Child Care: Setting a Research and Policy Agenda (pp. 375–405). Philadelphia, PA: Temple University Press. Hoem, B. (1993). The compatibility of employment and childbearing in contemporary Sweden. Acta Sociologica, 36, 101–120. Hoem, B. (1995). The way to the gender-segregated Swedish labor market. In: K. O. Mason & A.-M. Jensen (Eds), Gender and Family Change in Industrialized Countries (pp. 279–296). Oxford, UK: Oxford University Press. Hoem, J. M. (1993). Public policy as the fuel of fertility: Effects of policy reform on the pace of childbearing in Sweden in the 1980s. Acta Sociologica, 36, 19–31. Høgsnes, G. (1989). Wage bargaining and norms of fairness – A theoretical framework for analysing the Norwegian wage formation. Acta Sociologica, 32, 339–357. Judge, G. G., Griffiths, W. E., Carter Hill, R., L¨utkepohl, H., & Lee, T.-C. (1985). The theory and practice of econometrics (2nd ed.). New York: John Wiley and Sons.
350
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
Kamerman, S. B. (1988). Maternity and parenting benefits: An international overview. In: E. F. Zigler & M. Frank (Eds), The Parental Leave Crises: Toward a National Policy (pp. 235–244). New Haven, CT: Yale University Press. Kamerman, S. B. (1991). Parental leave and infant care: U.S. and international trends and issues, 1978–1988. In: J. S. Hyde & M. J. Essex (Eds), Parental Leave and Child Care: Setting a Research and Policy Agenda (pp. 11–23). Philadelphia, PA: Temple University Press. Lazear, E. P. (1991). Discrimination in labor markets. In: E. Hoffman (Ed.), Essays on the Economics of Discrimination (pp. 9–24). Kalamazoo, MI: Upjohn Institute. le Grand, C. (1991). Explaining the male-female wage gap: Job segregation and solidarity wage bargaining in Sweden. Acta Sociologica, 34, 261–278. Lindbeck, A. (1997). The Swedish experiment. Stockholm: SNS F¨orlag. ˚ (1989). Diskriminering p˚a svensk arbetsmarknad. En analys av l¨oneskillnader mellan L¨ofstr¨om, A. kvinnor och m¨an. Doctoral dissertation. Department of Economics. Ume˚a, Sweden: Ume˚a University. ˚ (1991). Kvinnors l¨oner i industrien. Den solidariska l¨onepolitikens effekter. ArbetarhisL¨ofstr¨om, A. torie, 15, 38–41. Meyerson, E. M., & Petersen, T. (1997). Lika l¨on f¨or lika arbete. In: I. Persson & E. Wadensj¨o (Eds), Kvinnors och m¨ans l¨oner–Varfor s˚a olika? (pp. 104–125). Statens offentliga utredningar (SOU), No. 136. Stockolm, Sweden: Fritzes. Meyersson Milgrom, E. M., Petersen, T., & Snartland, V. (2001). Equal pay for equal work? Evidence from Sweden and a comparison with Norway and the U.S. Scandinavian Journal of Economics, 103, 559–583. NOU. (1997). Stillingsvurdering for a˚ oppn˚a likelønn. Norges Offentlige Utredninger 1997, 10. Oslo: Statens Trykning. OECD. (1995a). Income distribution in OECD countries: Evidence from Luxembourg income study. OECD Social Policy Studies No. 18. Prepared by A. B. Atkinson, L. Rainwater, & T. M. Smeeding. Paris, France: Organisation for Economic Co-operation and Development. OECD. (1995b). Long-term leave for parents in OECD countries. Employment Outlook (July 1995), 171–202. Paris, France: Organisation for Economic Co-operation and Development, Department of Economics and Statistics. Petersen, T., & Meyersson, E. M. (1999). More glory and less injustice: The glass ceiling in Sweden 1970–1990. Research in Social Stratification and Mobility, 17, 199–261. Petersen, T., & Morgan, L. A. (1995). Separate and unequal: Occupation-establishment sex segregation and the gender wage gap. American Journal of Sociology, 101, 329–365. Petersen, T., & Saporta, I. (2003). The opportunity structure for discrimination. American Journal of Sociology (forthcoming). Petersen, T., Saporta, I., & Seidel, M.-D. (2000). Offering a job: Meritocracy and social networks. American Journal of Sociology, 106, 763–817. Petersen, T., Snartland, V., Becken, L.-E., & Olsen, K. M. (1997). Within-job wage discrimination and the gender wage gap, the case of Norway. European Sociological Review, 13, 199–213. Phillips, A. (1995). The politics of presence. Oxford, UK: Oxford University Press. Rønsen, M., & Sundstr¨om, M. (1996). Maternal employment in Scandinavia. A comparison of the after-birth employment activity of Norwegian and Swedish women. Journal of Population Economics, 9, 267–285. Rosenfeld, R. A. (1992). Job mobility and career processes. Annual Review of Sociology, 18, 39–61. Rosenfeld, R. A., & Kalleberg, A. L. (1991). Explaining the male-female wage gap: Job segregation and solidarity wage bargaining in Sweden. Acta Sociologica, 34, 207–225.
The Within-Job Gender Wage Gap, Sweden 1970–1990
351
SOU. (1993). L¨oneskillnader och l¨onediskriminering. Om kvinnor och m¨an i arbetsmarknaden. Statens offentliga utredningar (SOU), No. 7. Stockholm, Sweden: Allm¨anna F¨orlaget. Spilerman, S. (1986). Organizational rules and the features of work careers. Research in Social Stratification and Mobility, 5, 41–102. Stoiber, S. A. (1990). Family leave entitlements in Europe: Lesssons for the United States. Compensation & Benefits Management, 6, 111–116. Svensson, L. (1992). Vad betyder l¨onepolitiken f¨or kvinnol¨onerna? Ekonomisk Debatt, 7, 535–545. Svensson, L. (1995). Closing the gender gap. Determinants of change in the female-to-male bluecollar wage rates in Swedish manufacturing 1913–1990. Lund, Sweden: Ekonomiska Historiska Foreningen (Vol. 121). Treiman, D. J., & Hartmann, H. I. (Eds) (1981). Women, work, and wages: Equal pay for jobs of equal value. Washington, D.C.: National Academy Press. Verba, S., Kelman, S., Orren, G. R., Miyake, I., Watanuki, J., Kabashima, I., & Ferree, G. D., Jr. (1987). Elites and the idea of equality. A comparison of Japan, Sweden, and the United States. Cambridge, MA: Harvard University Press. Westerg˚ard-Nielsen, N. (Ed.) (1994). The Nordic labor markets in the 1990s. Wage differentials in the Nordic countries. Amsterdam: North Holland.
APPENDIX We give the equations used for computing the decompositions in column 1 of Tables 3 and 4. The average wages for women, for men, the relative wages, and the number of sex-integrated units (only in (2)–(5)) are given by: (1) (2) (3) (4) (5)
overall: w ¯ f, w ¯ m , and w ¯ r,r = w ¯ f /w ¯ m; in industry b: w ¯ b,f , w ¯ b,m , w ¯ b,r = w ¯ b,f /w ¯ b,m , and N b(I) ; in occupation o: w ¯ o,f , w ¯ o,m , w ¯ o,r = w ¯ o,f /w ¯ o,m , and N o(I) ; in establishment e: w ¯ e,f , w ¯ e,m , w ¯ e,r = w ¯ e,f /w ¯ e,m , and N e(I) ; in occupation-establishment unit oe: w ¯ oe,f , w ¯ oe,m , w ¯ oe,r = w ¯ oe,f /w ¯ oe,m , and N oe(I) ;
where the notation Nb(I) with the subscript (I) indicates that this is the number of sex-integrated units, in this case at the industry level, with corresponding meaning for No(I) , Ne(I) , and Noe(I) . The raw relative wages between women and men is given as the ratio of average women’s to average men’s wages (multiplied by 100): w ¯f × 100. (1) w(r,r) = w ¯m The relative wages controlling for industry obtains as w(b,r) =
Nb(I ) 1
N b(I)
b=1
wb,r × 100 =
Nb(I ) ¯ b,f 1 w × 100. N b(I) w ¯ b,m b=1
(2)
352
T. PETERSEN, E. M. M. MILGROM AND V. SNARTLAND
Here, and similarly below, the notation w(b,r) with the parenthesis (b, r) indicates that this is an average of the observations at the relevant level, in this case an average of the industry-specific gaps. The relative wages controlling for occupation obtains as w(o,r) =
No(I ) 1
N o(I)
wo,r × 100 =
o=1
No(I ) ¯ o,f 1 w × 100. N o(I) w ¯ o,m
(3)
o=1
The relative wages controlling for establishment obtains as w(e,r) =
Ne(I ) 1
N e(I)
we,r
e=1
Ne(I ) ¯ e,f 1 w × 100 = × 100. N e(I) w ¯ e,m
(4)
e=1
The relative wages controlling for occupation-establishment obtains as w(oe,r) =
Noe(I )
1 N oe(I)
woe,r × 100 =
oe=1
Noe(I )
1 N oe(I)
w ¯ oe,f × 100. w ¯ oe,m
(5)
oe=1
The percentage of the raw wage gap – that is, 100 minus w(r,r) , where w(r,r) comes from (1) – due to occupation-establishment segregation alone is given by %w(oe,r) =
w(oe,r) − w(r,r) × 100. 100 − w(r,r)
(6)
The percentage due to industry, occupation, or establishment alone, obtains by replacing w(oe,r) in (6) with w(b,r) , w(o,r) , or w(e,r) . One interpretation that can be given to the measures in (3)–(5) is this. In the case of occupation, with similar interpretations for establishment and occupationestablishment, equation (3) gives the overall wage gap one would observe if (a) men and women were equally distributed on occupations and (b) the wage gap within each occupation is the same across all occupations, equal to the average gap across occupations, so that wo,r equals w(o,r) from (3). The assumption made in (b) above, that the wage gap within each occupation is the same across all occupations, equal to the average gap across occupations, namely w(o,r) from (3), amounts to the relationship w ¯ o,r × 100 =
w ¯ o,f × 100 = w(o,r) , w ¯ o,m
for all occupations o.
(7)
If instead one wants to estimate the overall wage gap with an equal distribution of men and women on occupations, but where the gap within an occupation varies across occupations, one would have to proceed as follows. Let o,m be the proportion of the men who are in occupation o. With women having the same distribution
The Within-Job Gender Wage Gap, Sweden 1970–1990
353
∗ , on occupations as men, we get that the overall wage gap, which we can call w(r,r) would be N o(I) o,m · w ¯ o,f ∗ w(r,r) = No=1 × 100. (8) o(I) ¯ o,m o=1 o,m · w
Under the assumption in (7), replacing w ¯ o,f in (8) with the right-hand side of ∗ (7), after solving for w ¯ o,f , yields that w(r,r) then equals w(o,r) . In (8), one could use other distributions on occupations than the male distribution, for example, the marginal distribution on occupations. In the regression analysis of white-collar employees we regress the logarithm of the hourly wage on occupation-establishment and employee-level characteristics. The semilogarithmic form of the wage equation is chosen for its ease of interpretation: a coefficient is interpretable (roughly), after having been multiplied by 100, as the percentage change in the dependent variable resulting from a unit increase in the independent variable. Below, the subscripts used are as follows: i for individuals, o for occupations, and e for establishments. So, the subscript ioe denotes individual i in occupation o in establishment e. The generic equation in focus is: lnwioe = x ioe + ␥S ioe + oe + ioe
(9)
Here, xioe are the covariates for individual i in occupation o and establishment e and Sioe is the employee’s sex, equal to 1 for women, 0 for men. The individual-level characteristics included in xioe are educational level, age, and part-time status, while oe is a dummy variable for occupation-establishment unit oe, and ioe is a random error term. In the analysis we include only occupation-establishment units that are sex integrated, so that the results are based on the same set of observations as those in Table 4, the last line in each year. In analysis of covariance terminology (e.g. Judge et al., 1985: chapter 13.3), equation (9) corresponds to the within estimator. In (9), oe is treated as a fixed effect, entered as a set of dummy variables, one for each occupation-establishment unit. Equation (9) allows us to determine whether there still is a gender wage gap at the occupation-establishment level, captured by the coefficient for sex, after having taken into account the individual-level characteristics in xioe .
FRAME DECAY, INFORMAL POWER, AND THE ESCALATION OF SOCIAL CONTROL IN A MANAGEMENT TEAM: A RELATIONAL SIGNALING PERSPECTIVE Rafael Wittek, Marijtje A. J. van Duijn and Tom A. B. Snijders ABSTRACT In a study of conflict in organizations, Lindenberg’s relational signaling theory is used to develop hypotheses on the impact of relationship strength, network embeddedness, and organizational change on social escalation. Social escalation is defined as the involvement of one or more third parties in a conflict. An empirical test is conducted with data on 67 conflicts involving 22 managers, gathered during three years of ethnographic fieldwork and a longitudinal network study in a management team of a German paper factory. Multilevel analysis indicates that strong ties between conflicting parties decrease the level of social escalation, whereas informal power advantage of one party increases the chances for social escalation. Both effects disappear over time. It is argued that the dissolving impact of relationships and networks is due to the disappearance of so-called
The Governance of Relations in Markets and Organizations Research in the Sociology of Organizations, Volume 20, 355–380 Copyright © 2003 by Elsevier Science Ltd. All rights of reproduction in any form reserved ISSN: 0733-558X/PII: S0733558X02200130
355
356
RAFAEL WITTEK, MARIJTJE A. J. VAN DUIJN AND TOM A. B. SNIJDERS
solidarity frame-stabilizing activities in the firm. The results highlight the context-dependence of network effects and escalation processes.
INTRODUCTION It has often been argued that during the past two decades “flat” organizational structures and self-organizing teams have acquired a more important role in the governance of organizations (Cappelli et al., 1997, pp. 94–99; Osterman, 2000). This development also stimulated an increasing theoretical interest in factors that facilitate or inhibit the functioning of these new organizational forms (see, e.g. Podolny & Page, 1998). Informal conflict management plays a crucial role in this context (Wittek, 1999). Three implicit assumptions seem to underlie writings on conflict management and social control behavior. First, teams are successful because they contribute to the solution of the second-order free-rider problem. This means that, when confronted with a problem caused by a colleague, team members opt to actively sanction their peer rather than remain passive and wait for an intervention by a superior. Second, workgroups help reduce transaction costs because they encourage quick bilateral conflict resolution among those involved, thus avoiding the time-consuming activation of third parties as mediators or allies and minimizing the use of formal grievance procedures. The number of individuals and institutions involved in the resolution of intraorganizational conflicts is far less than in bureaucratic organizations. Third, in comparison with bureaucratic organizations, conflicts are resolved informally rather than through interventions by formal agents of control. The theoretical construct that encapsulates all three dimensions – passive versus active, direct versus indirect, informal versus formal – is called social escalation (Morrill & King Thomas, 1992). The level of social escalation is defined by the number of actors who participate in the process of sanctioning. In other words, the higher the number of third parties that are involved, the higher the level of escalation. The different levels of social escalation can be distinguished as: unilateral, direct, indirect, formal, and public strategies. Examples of these types of behavior are, respectively: hidden retaliation, talk to the other conflict party, gossip with colleagues about the other person, complain to your boss about your colleague, and discuss the issue during a department meeting. Thus, social escalation is lowest where the “instigator” acts on his own, and highest where sanctions are allocated to another actor in the presence of at least one colleague. The activation of a superior is considered as ranking higher in the escalation hierarchy than the incorporation of a peer. The degree to which this theoretical construct is in fact congruent with our cognitive representation of conflict management strategies is of course a question
Frame Decay, Informal Power, and the Escalation of Social Control
357
that needs further empirical investigation (but see Ellickson, 1991; Rooks, 2002 for supporting evidence). In the discussion section, we will tackle some of the problems this conceptualization of the theoretical construct entails. So far, relatively little systematic empirical research has been done to address the determinants of social escalation in organizations. Most social-psychological research conceives of escalation as increasing levels of assertiveness or emotional reactions within a dyad (see, e.g. Nauta & Sanders, 2000; Pruitt et al., 1997) while the management and organization literature usually emphasizes the role of supervisors as arbitrators in conflicts. The result is that these literatures focus on the effectiveness of different negotiation strategies, leadership styles, and the use of grievance procedures (see, e.g. Sheppard, 1984; Staw & Ross, 1987). Both research traditions have largely neglected the question: under which conditions does someone involve a third party in a conflict? This article will attempt to address this question. We want to find out how choices that are made in a conflict can explain variations in the level of social escalation. By doing so, we want to show the usefulness of a specifically sociological approach which emphasizes structural aspects of conflict resolution and the social and institutional embeddedness of actors. The article is structured as follows. The following section gives a brief overview of current research on social escalation, particularly the impact of three factors: relationship strength, network embeddedness, and organizational context. The theoretical framework and five hypotheses will be sketched and empirically tested with data on informal control and social networks between 22 members of a management team in a German paper factory. We conclude with a discussion of the implications of the research findings for organization theory and conflict research.
RELATIONSHIP STRENGTH, NETWORK EMBEDDEDNESS, AND ORGANIZATIONAL CONTEXT AS DETERMINANTS OF ESCALATION OF CONFLICTS Studies on the determinants of informal control strategy choice and conflict management behavior usually focus on one of four different levels of analysis with regard to the independent variables: the organizational context, the pattern and quality of relationships, the grievance event, or actor attributes and personality characteristics (for a systematization and overview of this literature, see Wittek, 1999, ch. 2). In this article, we limit ourselves to the organizational and relational dimension. Conflict researchers studying how the quality of social relations between two conflicting parties can influence social escalation tend to disagree about the direction of this effect. Empirical research often produces contradictory results (Dillard & Burgoon, 1985, p. 292). Some researchers (Cody et al., 1981;
358
RAFAEL WITTEK, MARIJTJE A. J. VAN DUIJN AND TOM A. B. SNIJDERS
Fitzpatrick & Winke, 1979; Miller et al., 1994; Roloff & Barnicott, 1978) show that relational closeness between both parties favors escalation. Moreover, they argue that strong ties tend to lead to the use of a broader range of conflict management strategies (Sillars, 1980). Other authors (Cody et al., 1986; Michener & Schwertfeger, 1975; Pruitt & Carnevale, 1993, p. 136; Raven & Kruglanski, 1970) report effects in the opposite direction. They say that a high degree of intimacy is more likely to mean that direct and conciliatory strategies are used because those involved do not want to jeopardize their relationship (Cody et al., 1986). In other words, friends prefer not to resolve their conflicts via third parties. These contradictory findings led us to conclude that the closeness of relationships as such does not explain escalation, and that more attention should be paid to the mediating role of context conditions: “intimacy alone is of limited consequence in determining message selection, but [that] it interacts with other situational variables” (Dillard & Burgoon, 1985, p. 292). In this article, two context conditions will be examined more closely: network embeddedness of the conflicting parties and the organizational context. Since escalation implies a process in time, we will also take into account how conflict management behavior between two persons changes over time. Research on the effect of network embeddedness on social escalation shows that indirect conflict management strategies tend to occur when the “relational distance” between the two conflicting parties is great and when both parties have contact with third parties whom they can involve either as allies or mediators (Barley, 1991; Burt & Knez, 1996; Gargiulo, 1993; Kapferer, 1969; Lazega & Vari, 1992; Thurman, 1979; Wittek, 1997; Wittek & Wielers, 1998). In both cases, it is assumed that the likelihood of social escalation between conflicting parties who have no or a poor relationship increases with the strength of their network embeddedness. Organizational contexts as determinants of conflict management behavior are still among the least explored areas in intraorganizational conflict research (Barry & Watson, 1996, p. 289; Nauta & Sanders, 2000). However, there is evidence that both formal structure and organizational strategy play a role. Direct (bilateral) conflict management strategies are more likely to be found in “flat” and matrix organizations, whereas supervisors are more important for conflict resolution in more “traditional” bureaucratic organizations (Morrill, 1995). It seems that, particularly, in flat organizations conflict management acquires a more hostile character than in traditional bureaucratic settings (Barker, 1993; Nauta, 1996). Though the factor time is inextricably linked with the phenomenon of escalation, we are not aware of any empirical work – apart from case studies (Ellickson, 1991; Kapferer, 1969; Thurman, 1979) – that systematically investigates how conflict management behavior between organizational members changes over time.
Frame Decay, Informal Power, and the Escalation of Social Control
359
Experimental research (Pruitt et al., 1997) and the case studies just mentioned show that both emotional and social escalation are the result of unsuccessful previous attempts to resolve the conflict. An additional complication for organization studies results from the fact that changes in organizational context are often among the reasons for the escalation of conflicts (Tebbutt & Marchington, 1997). This brief sketch of previous research leads to the following conclusions. Although there is a lack of systematic empirical research, the strength of the relationship between the contestants, their network embeddedness, the organizational context, and the history of the conflict itself seem to be important determinants of social escalation. The question is how these factors interact with each other. Previous research has not given a theoretical explanation for the contradictory effects of relationship strength. Therefore, it is still unclear under which context conditions a strong relationship between the contestants will favor or impede social escalation. The following section sketches a theoretical framework capable to resolve this problem.
RELATIONAL SIGNALS AND ESCALATION OF CONFLICTS The interaction between relationship strength, network embeddedness, context conditions and cooperation is a central element of Lindenberg’s (1997, 1998, and Lindenberg’s, 2003 contribution in this volume; see also M¨uhlau, 2000; Wielers, 1997) relational signaling theory, which has recently been used to explain conflict management behavior (Wittek, 1999). Relational signaling theory analyzes relationship strength in terms of the contribution of social relationships to individual social well-being. The latter can be achieved either through the exchange of affect, the recognition of prestige, or through social rewards as they follow from compliance to solidarity norms. In solidarity relationships, the realization of individual gain is tempered by the desire to conform to solidarity norms. The stronger the relationship, the more one is willing to take into consideration the interests of the other actor, even if doing so entails considerable cost. Such solidarity considerations are absent in opportunistic relationships. Here, both actors try to maximize their personal gains and are willing to cheat. How do solidarity norms influence conflict management behavior? In order to answer this question it is necessary to address the content of solidarity norms. In stable solidarity relationships both parties conform to normative expectations. This implies that they take into consideration how their actions will affect the other’s well-being, as well as the relationship to this person (see also Pruitt & Carnevale, 1993, p. 137). Each party avoids actions that could be interpreted as
360
RAFAEL WITTEK, MARIJTJE A. J. VAN DUIJN AND TOM A. B. SNIJDERS
hostile and expects signals, which show that the other is still interested in the relationship. Both actors are attentive to cues (relational signals) in the behavior of the other person that may reveal something about his or her intentions with regard to the maintenance of the relationship. A positive relational signal means that the other person wants to continue the relationship with the receiver of the signal. In solidarity relationships, conflict management behavior will also be evaluated according to its relational signaling content. Direct strategies have some advantages in this respect. First, compared to indirect strategies they reduce the likelihood of communication biases because information that is spread via third parties is more likely to become distorted than direct communication. Involving a third party lengthens the chain of communication and multiplies the potential sources of misunderstandings. Third parties might misrepresent the original information either deliberately (because they pursue their own goals) or by accident (because they forget relevant pieces of information or emphasize different aspects). Second, direct strategies enable the receiver to judge the reliability of the signal better. Direct strategies imply that the sender and the receiver interact with each other verbally or non-verbally. While communicating, both actors have the opportunity to visually screen each other’s behavior for possible cues of relational (dis)interest. Goffman’s (1959, p. 2) distinction between expressions given vs. expressions given off illustrates this process. He showed that relational signals in the first category – like a gift – enable the sender to hide his or her potential opportunistic intentions (an expression given), whereas signals in the second category are far more difficult to manipulate because they are physiological signals (blushing, showing nervousness, etc.) which are difficult to control (they are given off ). If a person sends a message to another person via a third party rather than in a face-to-face encounter, the receiver of the message does not get the opportunity to observe expressions given off. From this perspective, the content of the message (Kellermann & Cole, 1994) becomes less important than the question whether the conflict party chooses a direct or an indirect strategy (Morrill & King Thomas, 1992). Face-to-face communication between both conflicting parties means that they enable the other person to receive relational signals as expressions given off. These advantages, and the fact that direct strategies are themselves an expression of a direct relation, imply that, in case of a conflict, the use of a direct strategy can often be seen as a positive relational signal. The question is: when will one of the conflicting parties choose a direct or an indirect strategy to resolve the conflict? The choice of an indirect strategy implies that the other party does not get the opportunity to receive expressions given off. Therefore, in a functioning solidarity relationship indirect strategies are more likely to be interpreted as negative relational signals than direct strategies. Assuming that conflict parties anticipate on the relational signaling character of their sanctions
Frame Decay, Informal Power, and the Escalation of Social Control
361
it follows that the stronger the solidarity relationship between the two parties, the more they will prefer direct strategies of conflict resolution. However, individuals in solidarity relationships will not immediately suspect opportunistic motives if one of the others in the relationship violates a solidarityrule. To a certain extent, they will tolerate such infractions – if they are not too costly, or if the other party makes credible efforts to restore the status quo by apologizing or actively attempting to repair any damage (Lindenberg, 1998; see also Ellickson, 1991). As a result, particularly in the case of minor problems, individuals in solidarity relationships are less likely to actively sanction each other than those in opportunistic relationships (Wittek, forthcoming). In relationships where solidarity is absent or opportunistic behavior dominates, conflicting parties do not take the impact of their behavior on the other’s well-being into consideration and they do not expect positive relational signals. Consequently, their choice of a conflict management strategy is not constrained by the relational signaling character of their sanctions. In an opportunistic relationship, where ego is the instigating party and alter is the other party in a relationship, it is likely that alter will put up more resistance to direct control than alter in a solidarity relationship would. Therefore, ego in an opportunistic relationship will try to choose a level of escalation at which alter’s resistance is likely to be lower or the benefits of conflict management are expected to be higher. Alter’s resistance will be lower if the direct control attempt is made by somebody (an ego) with whom he or she has a solidarity relationship. Empirical research on the expected effectiveness of attempts to wield influence in different relational contexts shows that expected effectiveness is higher in solidarity relationships than in opportunistic relationships (Cody et al., 1986; Dillard & Burgoon, 1985). Assuming a continuum of social escalation that ranges from unilateral through direct, indirect, formal, and public forms of conflict resolution, it can be hypothesized that conflicts within solidarity relationships will be resolved unilaterally or directly: Solidarity Hypothesis (H1). The stronger the solidarity relationship between ego and alter, the lower the level of escalation that ego chooses to resolve conflicts with alter. How does network embeddedness influence the level of social escalation? From a relational signaling perspective, social relationships are an important means to acquire social well-being through the production of social approval in the form of behavioral confirmation, affect, or status recognition. For situations in which an actor (ego) is confronted with negative externalities generated by another actor (alter), this assumption has the following implications. First, if the relationship between the two is defined in terms of (weak or strong) solidarity, ego may see behavior that results in negative externalities as a potential
362
RAFAEL WITTEK, MARIJTJE A. J. VAN DUIJN AND TOM A. B. SNIJDERS
threat to the stability of the solidarity relationship. Alter’s action may be an indicator for the decreasing salience of alter’s solidarity frame. In other words, ego’s estimation of the likelihood of upsetting alter (i.e. losing social approval) will increase. If ego has no credible evidence that alter misbehaved unintentionally (e.g. alter offers credible apologies, the misbehavior turns out to be a mistake) ego has an incentive to compensate the temporary loss of social approval. The threat to ego’s social production function will stimulate him to look for alternative sources of social approval (Lindenberg, 1998). In network structures, there will be potential, substitute, sources of social approval in the form of third-parties who are related to ego in a solidarity relationship, but who are themselves not linked to alter, and may even have a negative relationship with alter. This specific structural situation increases the likelihood that they will not defend alter, but express sympathy for ego’s statements about alter, and even actively defend ego in discussions with alter. Burt (2001) refers to this process as the echo-effect. Note that this effect does not necessarily exclude the possibility that ego may try to resolve the conflict with alter bilaterally. Due to the solidarity relationship, ego will estimate alter’s expected resistance to a direct control effort as being lower than the expected resistance of somebody with whom he has no ties. Second, if the relationship between ego and alter is not a solidarity relationship, the negative externality caused by alter is not a threat to ego’s production function for social well-being, since the relationship was not necessary as a source of social approval. Consequently, there is no regulatory interest originating from the fear of loss of social approval. However, in organizational settings, functional interdependencies often force people to work together who have a negative relationship. Where employees depend on each other to carry out their jobs properly, a regulatory interest emerges from the functional interdependencies between them. It is to ego’s advantage if alter stops producing negative externalities in the future. Since they do not have a solidarity relationship, ego will estimate alter’s resistance to a direct control effort as being higher than the expected resistance of somebody with whom he as a solidarity relationship. This decreases the expected probability that bilateral control efforts will be successful. The primary purpose of mobilizing allies in this case is to increase the potential indirect pressure on alter. Allies can exert indirect pressure either verbally by arguing with alter or supporting ego’s standpoint, or by withholding resources necessary for alter to carry out his work. Hence, the first condition for networks to have an effect on social escalation consists of the presence of potential allies defined as third parties who have no, or a negative, tie to alter, but to whom ego has a solidarity relationship. Such allies are either alternative sources of social approval for ego or partisans who can exert indirect pressure on alter on ego’s behalf.
Frame Decay, Informal Power, and the Escalation of Social Control
363
The presence of allies is a necessary, but not a sufficient condition to increase the level of escalation. The reason is that alter also has the opportunity to mobilize allies, and can therefore threaten ego’s work situation. Allies will only be mobilized if the expected benefits in terms of social approval or indirect pressure outweigh the expected costs from alter’s counter-mobilization efforts. It can be assumed that the expected benefits of mobilization of an ally will be higher if one has more allies than the other party. In an opportunistic relationship, if ego has more allies than alter, ego can exert more indirect pressure on alter than alter can on ego. Ego has more supporters and will be less vulnerable to alter’s efforts to mobilize his own allies against him. In a solidarity relationship, if ego has more allies than alter, mobilizing these allies increases the opportunity to produce social approval. Hence, the second factor in the way a network effects social escalation consists of the relative informal power advantage of ego over alter. If both factors are present, that is to say if ego has potential allies, and the number of allies constitutes a relative informal power advantage over alter, the likelihood of social escalation increases. Informal Power Advantage Hypothesis (H2). The larger the relative informal power advantage that ego has over alter, the higher the level of social escalation that ego will choose to resolve conflicts with alter. A crucial requirement for the effects specified in Hypotheses 1 and 2 is the stability of the underlying solidarity frames. Only if both actors have an interest in maintaining their relationship and if both are convinced that the other is not driven by opportunistic motives, will repeated conflicts between two persons not result in escalating conflict management behavior. However, solidarity frames are inherently unstable and vulnerable (Lindenberg, 1998, pp. 78–92). The major reason is that conforming to solidarity norms will not always attract the attention of other persons. As a result, norm-conformity is often not rewarded, while the person conforming to norms will be well-aware of the related costs. This leads to a gradual erosion of the solidarity frame: When there are fairly high costs involved in executing solidary behavior and when the situations are repetitive, then we are likely to observe, ceteris paribus, a decay in the overall salience of the solidarity frame (Lindenberg, 1998, p. 80).
This process will not occur if there are supplementary arrangements to stabilize the solidarity frame, such as activities or institutions in the group context. Lindenberg mentions ritual confirmation of group membership and identification of common goals as the most important factors:
364
RAFAEL WITTEK, MARIJTJE A. J. VAN DUIJN AND TOM A. B. SNIJDERS
The most obvious and well recognized focus of a frame-stabilizing ritual in a sharing group is the identification of the group as group, making membership easily recognizable, and the celebration of a common goal that is abstract enough to cover all joint lower-level goals (Lindenberg, 1998, p. 83).
Thus, solidarity frames will erode only if these types of stabilizing activities are lacking or decreasing in frequency or intensity. If the strength of the solidarity frame influences conflict management behavior – as is claimed in the Solidarity Hypothesis – then the level of escalation between two persons who repeatedly have a conflict will increase through time in settings where little frame-stabilizing activities occur: Frame Decay Hypothesis (H3). The fewer solidarity frame-stabilizing activities occur in the dyad or the group context, the higher the level of escalation that ego will choose to resolve (repeated) conflicts with alter. The effect specified in the Frame Decay Hypothesis in fact predicts the increase of indirect conflict management strategies over time independently of the strength of the relationship between the conflicting parties and their informal power advantage. Put differently: social networks lose their impact as a determinant of conflict management behavior if solidarity frame-stabilizing activities in the organizational context disappear. There are two major reasons for this process. First, somebody who has repeatedly been annoyed by the same person will be less and less inclined to interpret this behavior in a solidarity frame. Even though solidarity considerations (and the focus on the costs associated with the conflict) will not immediately disappear, ego will become increasingly aware of the costs resulting from alter’s behavior. Therefore, the de-escalating effect of a solidarity relationship between the two parties will decrease over time if no other factors contribute to the stabilization of the solidarity frame. This means that the solidarity norm to resolve conflicts directly will lose its strength as the principle that guides ego’s conflict management efforts. Second, a reduction of solidarity frame-stabilizing activities in the organizational context will increase the likelihood that individuals misinterpret each other’s behavior as ambiguous relational signals. Behavior that would be perceived as innocent within a salient solidarity frame, will raise suspicion as potentially being driven by opportunistic motives when stabilizing efforts are lacking. The reason for this has to be sought in a spill-over effect. The individuals in a dyad are exposed to information about grievances and sanctioning behavior taking place in the rest of the group. Solidarity stabilizing activities facilitate the interpretation of these actions in terms of their relational signaling character because they create opportunities for direct interaction and thus mutual monitoring of group members. They also help to create clear distinctions between actions that constitute either a negative or a positive relational signal. The reduction of frame-stabilizing activities
Frame Decay, Informal Power, and the Escalation of Social Control
365
will increase the uncertainty about the real intentions behind other’s actions. It becomes more likely that norm violations as well as sanctions are seen as ambiguous relational signals. The result is a destabilization of the solidarity frame in social relationships. Where the solidarity frame between two conflicting parties loses its salience, their ties are also less likely to stimulate the use of direct conflict management strategies. Similarly, the fading solidarity frame of allies and mediators will render the power base of a potential controller increasingly insecure and therefore reduce the tendency to use higher levels of social escalation. As a result, the use of indirect strategies will also decrease in dyads characterized by opportunistic relationships. The following hypothesis summarizes these interaction effects between decreasing frame-stabilizing activities, relationship strength, and third-party ties on social escalation: Destabilization Hypothesis (H4). The less solidarity frame-stabilizing activities occur in the dyad or group context, the weaker will become the effect of: (a) relationship strength and (b) informal power advantage on social escalation over time. Figure 1 presents a graphical overview of all hypotheses.
Fig. 1. Graphical Representation of the Hypotheses.
366
RAFAEL WITTEK, MARIJTJE A. J. VAN DUIJN AND TOM A. B. SNIJDERS
RESEARCH DESIGN A test of the hypotheses requires detailed information on five dimensions: (a) the conflict management strategy that ego has used with regard to alter in a conflict, (b) the quality of the relationship between ego and alter, (c) the network embeddedness of ego and alter (i.e. how many potential mediators and allies are in ego’s network), (d) repeated measurements of conflict management behavior in a group, taken over an extended time period, and (e) type and frequency of frame-stabilizing activities in the group context. In order to collect this information, a multi-method design was used to study conflict resolution practices in the management team of a German paper factory (Wittek, 1999). The three most important sources of data are questionnaire based longitudinal social network research, semi-structured “trouble case” interviews, and participant observation. In what follows, we will first describe the organizational context. A brief description of the network panel and the trouble case methodology follows. The Organization. The factory is situated in a village with 800 inhabitants in southern Germany. When fieldwork started in 1995, the organization had 170 employees and two paper machines. After being declared bankrupt in 1993, the company was taken over by a German multi-national that decided to invest 40 million German Marks on enlarging the site with a new production hall and a third paper machine. The latter was scheduled to be operative on September 1, 1995. During the observation period (February 1995 to July 1997), these activities before the deadline of September 1 1995 were the main focus at the factory. The formal structure of the factory was substantially changed twice during the observation period. This means it is necessary to distinguish three phases (for a detailed description of these processes see Wittek, 1999, pp. 67–152). During the first phase, the managers had to cope with a double workload. Besides their normal job in the daily production process, they were now also responsible for the successful realization of the common project. Mutual interdependence between them and the necessity to coordinate and cooperate reached previously unknown heights. During this phase (1995) a clear group goal was present. Daily meetings of the whole group contributed to the stability of a strong solidarity frame. With the successful completion of the project at the end of 1995, the common group goal disappeared, although the production department still formed a single entity. The allocation of responsibilities concerning the new paper machine was highly ambiguous. In the beginning of 1996, solving the new machine’s implementation problems was, on the whole, considered to be a joint task. But during the same
Frame Decay, Informal Power, and the Escalation of Social Control
367
period, the number of times the whole group met was reduced from regular daily gatherings to irregular meetings every two to three weeks. Finally, in 1997 the production department was split up into three semi-autonomous units. The frequency of meetings of the whole management team was further reduced to once a month. To sum up, formal activities, rituals, and categorizations, which could contribute to the stabilization of a solidarity frame of the whole group were gradually reduced in the period from 1995 to 1997. Network Panel. The questionnaire-based survey is based on a panel study with three measurements, spread over 18 months, with six months in between each measurement. Apart from items measuring attitude, the questionnaire contained a number of socio-metric questions that dealt with, for example, the intensity of personal relations. Twenty-two of the twenty-five members of the management team responded. This allowed the reconstruction of the network of personal relations between 22 managers at three points in time. In 1995, mean age of the team members was 41 (SD 10.9), with a mean tenure of 13.2 years (SD 11.8). Almost 80% had been working for the factory for more than 5 years. Two thirds of the managers had a degree in engineering. There were seven departments: production, the chemical laboratory, maintenance, logistics, personnel, technical customer service, and a project department. The majority of the managers did not live in the village where the factory is situated, and had little personal contact with each other outside the work context (Wittek, 1999, p. 243). Participant Observation. Approximately 7 months, spread over a period of three years (1995–1997), were spent doing ethnographic research at the site. The key element of fieldwork consisted of eliciting trouble cases through semi-structured interviews and direct observation (Morrill, 1995). A trouble case is defined as a critical incident (a) that was caused by a colleague, (b) that irritated the respondent, and (c) for which the reaction of the respondent was known (the so-called control event). Content analysis of the transcribed interviews and field notes resulted in a total of 171 control events, of which 125 involved the 22 respondents for whom there was also information available about the social network.
DATA AND METHOD OF ANALYSIS Dependent Variable. For each control event it was determined who was ego (the person carrying out the control effort), who was alter (the person towards whom
368
RAFAEL WITTEK, MARIJTJE A. J. VAN DUIJN AND TOM A. B. SNIJDERS
the control effort was directed), and whether it was a unilateral, direct, indirect, formal, or public reaction. Unilateral strategies are reactions to cooperation problems where ego does not talk to anybody about the incident. This includes resignation, but also tactics like avoidance of interaction or hidden retaliation. Direct strategies are control attempts during which ego privately talked to alter about the problem. Indirect strategies are situations when ego talks to one or more other colleagues about the problem without alter being present. If the third person is a superior of ego or alter, the control attempt is coded as a formal strategy. Finally, public control refers to the situation when ego talks to at least one other third person about the problem in the presence of alter. The dependent variable is the level of social escalation between two managers, increasing in strength from 1 to 5 in the following order: (1) (2) (3) (4) (5)
unilateral, direct, indirect, formal, public.
Thus, rather than using explorative scaling techniques, a theoretical ordering of the categories of the dependent variable was preferred.1 In 11 cases ego carried out a repeated effort (and in one case two repeated efforts) during one of the three phases in order to resolve a conflict with alter. The dependent variable is the strategy that was chosen during a control attempt between two actors at a particular point in time. If more than one conflict occurred between two persons during the observation period, the strongest strategy was coded. Since the selection of the events was purely based on the dependent variable, this selection entails no bias for the test procedure. A total of 67 complete observations were available for the 22 actors (see Table 1 ). Six of the events were unilateral, 30 direct, 1 indirect, 10 formal and 20 public. Independent Variables. Five independent variables were operationalized. Relationship strength between the managers was determined at three points in time using socio-metric questions. Managers indicated the intensity of their relationship with each colleague by ranking colleagues as those “whom you would certainly not take into your confidence” to “colleagues with whom you discuss personal problems related either to work or private life.” Answers were coded accordingly on a five-point scale ranging from: distant, absent, neutral, trusting to very trusting. The results were as follows (frequencies in brackets): distant (1), absent (5), neutral (26), trustful (29), and very trustful (6). Informal power advantage – a measure of the degree of network embeddedness – was determined by carrying out a triad census. For this purpose, relationships were
Frame Decay, Informal Power, and the Escalation of Social Control
369
Table 1. Frequency of Conflict Management Strategies per Phase. Strategy
Time 1995
1996
1997
Total
Unilateral Direct Indirect Formal Public
2 8 0 4 11
4 8 0 4 6
0 14 1 2 3
6 30 1 10 20
Total
25
22
20
67
trichotomized. “Trustful” and “very trustful” ties were coded as “strong,” and the remaining ties were coded as “absent” or “negative.” Figure 2 shows all the possible combinations of relationship strength and network embeddedness for the smallest possible network, the triad. Informal power advantage of ego over alter was calculated taking the difference in frequency between the power advantage of ego (see Fig. 2 for conditions: 4, 7, 8, 13, 16, 17, 22, 25, 26) and the power advantage of alter (see Fig. 2 for conditions: 2, 3, 6, 11, 12, 15, 21, 24). In a triad in which ego has an informal power advantage, ego has either: (a) a positive, a negative, or no tie to alter, and (b) a positive tie to a third party who has no, or a negative, tie to alter. The mean informal power is 3.04 (SD 5.68). The variable frame-stabilizing activities indicates whether a conflict took place in 1995 (“−1”), 1996 (“0”), or 1997 (“1”). Since frame-stabilizing activities decreased through time, the variable is reverse coded. As was indicated in the description of the organizational context, activities to stabilize a strong solidarity frame gradually decreased from 1995 to 1997. Since each of the three phases coincides with a change in the organizational context, the indicators for time and stabilizing activities in the organizational context are identical in this study. Two interaction terms were created by multiplying the centered variable framestabilizing activities with the (centered) variables relationship strength and informal power advantage. Control Variable. Previous research has shown that superiors tend to prefer the direct resolution of conflicts, and therefore choose a low level of escalation (Cody & McLaughlin, 1985; Kipnis et al., 1980; Miller et al., 1994, p. 183; Putnam & Wilson, 1982). Therefore, the formal power relationship between ego and alter was included as a control variable. A control effort of a superior towards a subordinate was coded “1,” control attempts between equals were coded “0,” and control efforts from a subordinate towards a superior were coded “−1.”
RAFAEL WITTEK, MARIJTJE A. J. VAN DUIJN AND TOM A. B. SNIJDERS
Fig. 2. Classification of Triads.
370
Frame Decay, Informal Power, and the Escalation of Social Control
371
Forty control attempts were lateral (i.e. between managers in the same hierarchical position). Nineteen control efforts were from a superior directed towards a subordinate, and 8 control efforts originated from a subordinate towards a superior. Method. The complex structure of the data – repeated measurements between actors in a social network – requires a non-standard method of analysis. There are various sources of variance: the two actors in a dyad, the directed relationship between them, and the repeated measurement. The various sources of variance result in dependence between the observations, so that methods like linear regression – which assumes that residuals are independent – are not suitable. An appropriate model for the present data structure is the Social Relations Model (Snijders & Bosker, 1999, ch. 13; Snijders & Kenny, 1999). This is a special multilevel model with crossed random effects. In this model, the dependence between relations from and to the same actor is modeled with (correlated) random effects of the “sending” and the “receiving” actor. Due to the data structure in our case, the Social Relations model needs to be enlarged by means of an additional (lowest) level, representing the repeated measurements. Since there are only 67 observations (i.e. only 5% of 22 × 21 × 3 = 1386 possible observations for 22 actors and 3 points in time), the data are highly imbalanced. This is not a problem for the definition of the model, but leads to a lower power for testing various variance parameters (which represent the strength of the sources of variance as they were sketched above), and it may result in numerical instability of the estimation algorithm. The estimation of the Social Relations Model showed that some variances were estimated 0, and in some specifications could not be estimated at all due to non-convergence. In the models using various explanatory variables reported here, the variance for the “sending” and the “receiving” actors was estimated to be 0, and therefore negligible compared to other variances. Therefore, given these data, it is not necessary to use the Social Relations Model, whose cross-nested structure can be simplified into a standard multilevel model with three levels: the observation (level 1), nested in a directed relationship (level 2), nested in a dyad (level 3). The dyad, formed by actors i (ego) and j (alter), comprises of two directed relationships: from i to j and from j to i (which, as explained above, do not need to be realized both in this data set). Thus, this model assumes that the dyads (i.e. the pairs of actors) are independent, which is based empirically on the zero estimates for the variances in the Social Relations Model. In the model, both a linear and a random effect of frame-stabilizing activities were specified at the dyad level, in order to investigate whether or not the level of escalation varies through time. Interaction-effects of all centered variables with frame-stabilizing activities were specified, in order to find out whether or not the effect of the variables varies through time. In order to model the effect of time, a random slope (for a discussion of this term see Snijders & Bosker, 1999, ch. 5)
372
RAFAEL WITTEK, MARIJTJE A. J. VAN DUIJN AND TOM A. B. SNIJDERS
for frame-stabilizing activities (coded as “−1,” “0,” and “+1”) was incorporated into the model. This effect represents the possibility that, in some dyads, social escalation – as far as it is not modeled through explanatory variables – can increase or decrease more strongly than in other dyads.
RESULTS Table 2 summarizes the parameter estimates. On the level of the dyad (i.e. between dyads) the variance turns out to be negligible. On the relationship level (i.e. between relations within a dyad) the variance is small. Variance is mainly found between the three points in time. When testing at the 0.05 level, significant main effects of relationship strength and frame-stabilizing activities, as well as
Table 2. Parameter Estimates for Complete Multilevel Models for the Effects of Relationship Strength, Network Embeddedness, and Time on Social Escalation. Effect
Fixed effects Relationship strength Informal power advantage Stabilizing activities (decrease)b Stabilizing activities × Relationship strength Stabilizing activities × Informal power advantage Formal power advantagec Random effects Between-dyad variance Within-dyad variance Between-time variance Constant Deviance
Hypothesis
Parameter estimate
Standard error
p-valuea
− + + +
−0.42 0.045 −0.48 0.42
(0.22) (0.031) (0.21) (0.31)
0.026 0.073 0.020 0.088
−
−0.057
(0.035)
0.052
−
−0.36
(0.28)
0.102
0 0.14 1.62 3.10 227.5
0 (0.49) (0.55) (0.18)
Note: N = 67 conflicts. The dependent variable is coded as follows: 1 = unilateral, 2 = direct, 3 = indirect, 4 = formal, 5 = public. a For frame-stabilizing activities a two-sided p-value is reported, since the direction of the effect differs from the predicted direction. All other p-values are based on one-sided tests. b Coded as: “−1” (1995, frequent); “0” (1996, intermediate); “1” (1997, rare). c Coded as: “1” control effort from a superior towards a subordinate; “0” control effort between equals; “−1” control effort from a subordinate towards a superior.
Frame Decay, Informal Power, and the Escalation of Social Control
373
significant interactions between frame-stabilizing activities and informal power advantage (the effect decreases through time) were found. The main effects of formal power and informal power advantage are significant at the 0.10 level, as is the interaction effect between frame-stabilizing activities and relationship strength (the effect becomes weaker through time). The Solidarity Hypothesis (H1) predicts a negative effect of relationship strength on social escalation: the stronger the relationship between ego and alter, the lower the level of social escalation. The effect of relationship strength is negative ( p = 0.026). This means that a strong tie between two conflicting parties in fact reduces the likelihood of escalation of a conflict. Our results support the Solidarity Hypothesis. Informal power advantage has a positive effect on social escalation ( p = 0.073), as specified in the Informal Power Advantage Hypothesis (H2). Thus, individuals who have more allies than their counterparts also are more likely to induce higher levels of social escalation. The Frame Decay Hypothesis (H3) predicts that the level of escalation increases over time if no, or few, frame-stabilizing activities take place or their intensity decreases. Since the frequency of frame-stabilizing activities decreased from 1995 to 1997, one would expect a gradual increase in the level of escalation. The opposite is the case: the effect of the variable frame-stabilizing activities is negative ( p = 0.020). This means that the level of social escalation decreases rather than increases over time. This trend is shown in Table 1. At the beginning of the observation period, the managers preferred public strategies, whereas later they prefer direct ones. However, the lowest level of social escalation (i.e. unilateral strategies) is used most frequently during the second phase. We will address the implications of this result in the discussion. The Destabilization Hypothesis (H4) predicts that the effects of: (a) relationship strength and (b) informal power advantage on social escalation decrease over time if frame-stabilizing activities are absent or decreasing over time. Both parts of this hypothesis are supported by the data. First, the interaction effect of framestabilizing activities with relationship strength is positive ( p = 0.088). The almost equal but negative main effect of relationship strength implies that during the early phases of the observation period a strong relationship between the conflicting parties favors de-escalation, whereas during the last phase a strong relationship does not contribute to the escalation of the conflict. Second, also the interaction effect between frame-stabilizing activities and informal power advantage is supported by the data: the effect is negative ( p = 0.052). Given the fact that the main effect of informal power advantage is positive, and more or less the same in size, it can be concluded that, the escalating impact of the relative informal power advantage between conflicting parties decreases through time and disappears in the last phase.
374
RAFAEL WITTEK, MARIJTJE A. J. VAN DUIJN AND TOM A. B. SNIJDERS
The effect of the control variable formal power is negative – as expected – but fails to reach significance at the 0.10 level ( p = 0.102). Those with formal authority over their targets have a slight tendency to choose lower levels of escalation than those without formal authority. Thus, the formal relationship between the conflicting parties still affects conflict management behavior after controlling for the variables relationship strength, informal power advantage, and framestabilizing activities. The positive effect supports earlier claims that subordinates use more non-confrontational and indirect strategies because they can compensate for their lack of a formal power base by building coalitions (Cody & McLaughlin, 1985, p. 291).
DISCUSSION AND CONCLUSION Based on relational signaling theory, four hypotheses were developed to explain how social escalation is effected by: relationship strength, informal power advantage, and changes in the frequency of frame-stabilizing activities. These hypotheses were tested empirically using data collected from 67 conflicts in a German paper factory, in particular the social networks between 22 members of the management team. Three of the four hypotheses are supported by multilevel analyses. These are: the level of social escalation decreases the stronger the relation between the conflicting parties; the de-escalating impact of social relationship decreases through time; the escalating effect of having a larger number of allies than one’s adversaries decreases through time. The effect of frame-stabilizing activities did not occur as expected. This variable ( frame-stabilizing activity) was found to have a mainly negative effect. This suggests that the level of social escalation decreases through time in contrast to an increase as predicted by the escalation hypothesis. A closer look at the frame-stabilizing activities (changes) and social escalation in the organizational context clarifies this result. Frame-stabilizing activities were coded on the basis of the indicator “frequency of meetings,” which decreased linearly from 1995 to 1997. However, as became evident from the ethnographic research, the most problems occurred in the organization in the second phase when the managers ceased to respond to cooperation problems, or tried to resolve them via their superior. During this phase of the observation period (1996) the system of informal social control had partially broken down, with visible negative consequence for the production process (Wittek, 1999, pp. 141–147). Social control
Frame Decay, Informal Power, and the Escalation of Social Control
375
activities, particularly direct strategies, increased again in the final phase. Based on the ethnographic evidence, it was shown that ambiguity of task responsibilities and the disappearance of a clear group goal had caused this discontinuous pattern. Two conclusions can be drawn from this evidence. Firstly, the number of meetings alone is not sufficient to stabilize a solidarity frame if there is no salient group goal and responsibilities are not clearly demarcated. Secondly, public control strategies were frequently used and proved extremely effective in sanctioning misbehavior in highly solidarity teams even though our theoretical construct predicted social escalation. Our conclusions seem to apply to situations characterized by the absence of relational signaling problems. In other words, a high solidarity setting with strong functional interdependencies, frequent frame-stabilizing activities, a salient group goal and clearly demarcated individual responsibilities. The lack of a group goal, combined with a decrease in frame-stabilizing activities and a blurring of responsibilities, resulted in an opportunistic frame being more common in phase two. Behavior that was perceived as unproblematic during the first phase increasingly acquired an ambiguous relational signaling character. The managers became increasingly wary of the opportunistic motives behind their colleagues’ behavior. The ambiguity of relational signals had a direct effect on conflict management behavior, and solidarity relationships between the conflicting parties were increasingly less able to prevent escalation. An increase in the use of formal and unilateral strategies was the result. Apparently, these ways of dealing with conflict are indicators of relational signaling problems. Finally, with the restructuring in phase three, weak solidarity became the dominant frame. Frame-stabilizing activities further decreased, nevertheless, clearly demarcated responsibilities, group goals and interdependencies were re-established at sub-unit level, which then led to the use of direct strategies. Thus, direct strategies can be said to flourish in settings where weak solidarity is the dominant relational frame. These results indicate that future research should pay closer attention to the role of public control strategies, a dimension that up to now has been largely neglected by researchers on conflict management and informal control in organizations. Our results also support a conclusion from previous research that argues that how social relationships and networks affect conflict management behavior depends on the specific organizational context. Hence, network research on organizational conflict management will never be complete as long as the broader organizational context and frame-stabilizing activities are neglected. In order to assess to what degree the theoretically motivated coding of the dependent variable social escalation influenced our results, we repeated several analyses after recoding. First, we dichotomized the dependent variable that distinguished between public and direct strategies, on the one hand, and indirect/formal and unilateral strategies, on the other. This produced quite different (and non-significant)
376
RAFAEL WITTEK, MARIJTJE A. J. VAN DUIJN AND TOM A. B. SNIJDERS
results in the parameter estimates. This finding supports our conclusion that public and direct strategies have different meanings as relational signals. Public strategies are more appropriate in strong solidarity settings while direct strategies are favored in weak solidarity contexts. Second, we dichotomized the dependent variable by distinguishing direct versus indirect, formal and public strategies. This produced comparable results to the ones presented here. Third, the position of the unilateral strategy in the hierarchy of control strategies (i.e. coding it as “high” or “low”) did not influence our results. This supports our proposed theoretical ordering of escalation strategies.2 Our results emphasize the importance of a process and contextual approach to social escalation. A strong tie reduces the likelihood that a conflict will escalate, whereas informal power advantage favors escalation. Both effects become weaker over time because of a gradual decrease in solidarity stabilizing activities (meetings etc.). Thus, the contradictory findings of earlier studies can be explained by the fact that they did not take into account the history of the conflict and changes in organizational contexts. Solidarity relations are fragile. If frame-stabilizing institutions and activities do not continuously support them, solidarity will gradually erode. The results also lead to a more general question. To what degree is the relational signaling character of different conflict management strategies context dependent? It is possible that in some organizations the mobilization of a superior is less problematic from a relational signaling point of view than the attempt to resolve a conflict in public during a meeting. The question is, of course, under which circumstances this will be the case. Identifying these circumstances is an interesting topic for future research, as is the relational signaling character of formal grievance procedures. The results show the advantages of using a relational signaling perspective to study informal control and conflict management in organizations. It provides a theoretical explanation for the contradictory findings in previous research. It also offers a specification of the interaction between relational, structural, and organizational factors and their effects on conflict management behavior. By doing so, the relational signaling approach can be a valuable supplement to perspectives, such as Dual Concern Theory (Pruitt & Carnevale, 1993; van de Vliert, 1997) or Conflicting Loyalties Theory (Flap, 1988), which either limit their attention to only one type of factor or neglect the dynamic aspect of conflict management processes. In this study, we did not look into all the factors that may affect social escalation such as: the type of the grievance (how strong are ego’s stakes in the issue?), the simultaneous use of various strategies (does ego attempt to influence alter directly and indirectly?), functional interdependencies between the actors (i.e. whose input does ego need to carry out to do his tasks properly?), the presence or absence of grievance procedures, as well as actor attributes like age or the amount of
Frame Decay, Informal Power, and the Escalation of Social Control
377
self-confidence. Nor did we examine the effect of different conflict management strategies on organizational outcomes like performance or cooperation (de Dreu, 1997; Gupta et al., 1994; Jehn, 1995; Murnighan & Conlon, 1991; Wittek et al., 2000). Rather than assessing the use of alternative theoretical approaches (for an overview, compare Wittek, 1999, ch. 2) or deriving competing hypotheses, we opted to elaborate on the relational signaling theory. This theory is well suited to transcend the “narrow” conceptualization of rationality because it can incorporate framing effects, assumptions about social goals, and network embeddedness systematically. Future research would benefit from increasing the level of cognitive complexity in comparison to less intricate models of social escalation (Wittek, forthcoming) that are limited to the cost aspects of grievances (Rooks & Snijders, 2001). The current study focuses on the impact of social networks on conflict management behavior and social escalation in organizations by analyzing empirical data of conflict over a period of three years in a German paper factory. At least in the present case study, informal social networks turned out to be far less powerful in providing a structural basis for the self-regulation of a work team than much of the literature on social networks would suggest. Informal networks will not be able to “replace” hierarchies as instruments to govern the firm. Our results emphasize that increasingly “horizontal” work relations require new organizational contexts that will produce less ambiguous relational signals and thus contribute to the stability of solidarity frames. Social networks will only have a substantial impact on de-escalation where intraorganizational rules and institutions work properly.
NOTES 1. Theoretically, unilateral strategies can represent both extremes of a continuum of social escalation. On the one hand they can reflect the fact that the conflict has a low priority for ego (ego tolerates the behavior). On the other hand, they can be an indicator for the fact that ego wants to dissolve the relationship with alter, because the conflict has escalated. The implications of this ambiguity will be addressed in the discussion. 2. Other dichotomizations (like take action vs. take no action, direct vs. indirect, public vs. indirect) were not feasible due to highly skewed distributions and lack of data (see Table 2).
ACKNOWLEDGMENTS We would like to thank the editors Vincent Buskens, Werner Raub, and Chris Snijders for their elaborate and constructive critique on earlier versions of this chapter. The chapter was written while the first author stayed at the Netherlands
378
RAFAEL WITTEK, MARIJTJE A. J. VAN DUIJN AND TOM A. B. SNIJDERS
Institute of Advanced Study in the Humanities and Social Sciences (NIAS). Research for this article was subsidized by a grant (Vernieuwingsimpuls, 2000) from the Dutch Organization for Scientific Research (NWO). The generous support of both institutions is gratefully acknowledged.
REFERENCES Barker, J. R. (1993). Tightening the iron cage: Concertive control in self-managing teams. Administrative Science Quarterly, 38, 408–437. Barley, S. R. (1991). Contextualizing conflict. Notes on the anthropology of disputes and negotiations. Research on Negotiation in Organizations, 3, 165–199. Barry, B., & Watson, M. R. (1996). Communication aspects of dyadic influence in organizations: A review and integration of conceptual and empirical developments. Communication Yearbook, 19, 269–317. Burt, R. S. (2001). Bandwith and echo: Trust, information, and gossip in social networks. In: A. Casella & J. E. Rauch (Eds), Networks and Markets (pp. 30–74). New York: Russell Sage. Burt, R. S., & Knez, M. (1996). Trust and third party gossip. In: R. M. Kramer & T. R. Tyler (Eds), Trust in Organizations. Frontiers of Theory and Research (pp. 68–89). Thousand Oaks, CA: Sage. Cappelli, P., Bassi, L., Katz, H., Knoke, D., Osterman, P., & Useem, M. (1997). Change at work. Oxford: Oxford University Press. Cody, M. J., Greene, J. O., Marston, P. J., O’Hair, H. D., Baaske, K. T., & Schneider, M. J. (1986). Situation perception and message strategy selection. Communication Yearbook, 9, 390–420. Cody, M. J., & McLaughlin, M. L. (1985). The situation as a construct in interpersonal communication research. In: M. L. Knapp & G. R. Miller (Eds), Handbook of Interpersonal Communication (pp. 263–312). Beverly Hills, CA: Sage. Cody, M. J., McLaughlin, M. L., & Schneider, M. J. (1981). The impact of relational consequences and intimacy on the selection of interpersonal persuasion tactics: A reanalysis. Communication Quarterly, 29, 91–106. Dillard, J. P., & Burgoon, M. (1985). Situational influences on the selection of compliance-gaining messages: Two tests of the predictive utility of the Cody-McLaughlin typology. Communication Monographs, 52, 289–304. de Dreu, C. K. W. (1997). Productive conflict: The importance of conflict management and conflict issue. In: C. K. W. de Dreu & E. van de Vliert (Eds), Using Conflict in Organizations (pp. 9–22). London: Sage. Ellickson, R. (1991). Order without law. How neighbors settle disputes. Cambridge, MA: Harvard University Press. Fitzpatrick, M. A., & Winke, J. (1979). You always hurt the one you love: Strategies and tactics in interpersonal conflict. Communication Quarterly, 27, 3–11. Flap, H. D. (1988). Conflict, loyalty, and violence. The effects of social networks on behaviour. Frankfurt am Main: Peter Lang. Gargiulo, M. (1993). Two-step leverage: Managing constraint in organizational politics. Administrative Science Quarterly, 38, 1–19. Goffman, E. (1959). The presentation of self in everyday life. Garden City, KS: Doubleday.
Frame Decay, Informal Power, and the Escalation of Social Control
379
Gupta, P. P., Dirsmith, M. W., & Fogarty, T. J. (1994). Coordination and control in a government agency: Contingency and institutional theory perspectives on GAO audits. Administrative Science Quarterly, 39, 264–284. Jehn, K. A. (1995). A multimethod examination of the benefits and detriments of intragroup conflict. Administrative Science Quarterly, 40, 256–282. Kapferer, B. (1969). Norms and the manipulation of relationships in a work context. In: C. Mitchell (Ed.), Social Networks in Urban Situations (pp. 181–245). Manchester: Manchester University Press. Kellermann, K., & Cole, T. (1994). Classifying compliance gaining messages: Taxonomic disorder and strategic confusion. Communication Theory, 4, 3–60. Kipnis, D., Schmidt, S. M., & Wilkinson, I. (1980). Intraorganizational influence tactics: Explorations in getting one’s way. Journal of Applied Psychology, 65, 440–452. Lazega, E., & Vari, S. (1992). Acteurs, cibles et leviers: Analyse factorielle des relations de controle indirect dans une firme americaine d’avocats d’affaires. Bulletin de M´ethodologie Sociologique, 37, 41–51. Lindenberg, S. M. (1997). Grounding groups in theory: Functional, structural and cognitive interdependencies. Advances in Group Processes, 14, 281–331. Lindenberg, S. M. (1998). Solidarity: Its microfoundations and macrodependence. A framing approach. In: P. Doreian & T. Fararo (Eds), The Problem of Solidarity. Theories and Models (pp. 61–112). London: Gordon and Breach. Lindenberg, S. M. (2003). The cognitive side of governance. In: V. Buskens, W. Raub & C. Snijders (Eds), Research in the Sociology of Organizations (Vol. 20, pp. 47–76). Oxford: Elsevier Science. Michener, H. A., & Schwertfeger, M. (1975). Liking as a determinant of power tactic preference. Sociometry, 35, 190–202. Miller, L. C., Cody, M. J., & McLaughlin, M. L. (1994). Situations and goals as fundamental constructs in interpersonal communication research. In: M. L. Knapp & G. R. Miller (Eds), Handbook of Interpersonal Communication (2nd ed., pp. 162–198). Thousand Oaks, CA: Sage. Morrill, C. (1995). The executive way. Conflict management in corporations. Chicago, IL: Chicago University Press. Morrill, C., & King Thomas, C. (1992). Organizational conflict management as disputing process. The problem of social escalation. Human Communication Research, 18, 400–428. M¨uhlau, P. (2000). The governance of the employment relation. A relational signaling perspective. Amsterdam: Thela Thesis. Murnighan, J. K., & Conlon, D. J. (1991). The dynamics of intense work groups: A study of British string quartets. Administrative Science Quarterly, 36, 165–186. Nauta, A. (1996). Oog om oog en baas boven baas. Interactiepatronen bij interpersoonlijk conflict op bureaucratische en organische organisatieafdelingen. Dissertation, University of Groningen. Nauta, A., & Sanders, K. (2000). Interdepartmental negotiation behavior in manufacturing organizations. The International Journal of Conflict Management, 11, 107–133. Osterman, P. (2000). Work reorganization in an era of restructuring: Trends in diffusion and effects on employee welfare. Industrial and Labor Relations Review, 53, 179–196. Podolny, J. M., & Page, K. L. (1998). Network forms of organization. Annual Review of Sociology, 24, 57–76. Pruitt, D. G., & Carnevale, P. J. (1993). Negotiation in social conflict. Buckingham: Open University Press. Pruitt, D. G., Parker, J. C., & Mikolic, J. M. (1997). Escalation as a reaction to persistent annoyance. International Journal of Conflict Management, 8, 252–270.
380
RAFAEL WITTEK, MARIJTJE A. J. VAN DUIJN AND TOM A. B. SNIJDERS
Putnam, L. L., & Wilson, C. E. (1982). Communicative strategies in organizational conflicts: Reliability and validity of a measurement scale. Communication Yearbook, 6, 629–652. Raven, B. H., & Kruglanski, A. W. (1970). Conflict and power. In: P. Swingle (Ed.), The Structure of Conflict (pp. 69–109). New York: Academic Press. Roloff, M. E., & Barnicott, E. F., Jr. (1978). The situational use of pro- and antisocial compliancegaining strategies by high and low Machiavellians. Communication Yearbook, 2, 193–205. Rooks, G. (2002). Contact and contract: Strategisch management van inkooptransacties. Amsterdam: Thela Thesis. Rooks, G., & Snijders, C. (2001). The purchase of IT-products by Dutch SMEs: A quantitative study. Journal of Supply Chain Management, 37, 34–42. Sheppard, B. H. (1984). Third party conflict intervention: A procedural framework. Research in Organizational Behavior, 6, 141–190. Sillars, A. L. (1980). The stranger and the spouse as target persons for compliance-gaining strategies: A subjective expected utility model. Human Communication Research, 6, 265–279. Snijders, T. A. B., & Bosker, R. J. (1999). Multilevel analysis. An introduction to basic and advanced multilevel modeling. London: Sage. Snijders, T. A. B., & Kenny, D. (1999). Multilevel models for relational data. Personal Relationships, 6, 471–486. Staw, B. M., & Ross, J. (1987). Behavior in escalation situations: Antecedents, prototypes, and solutions. Research in Organizational Behavior, 9, 39–78. Tebbutt, M., & Marchington, M. (1997). ‘Look before you speak’: Gossip and the insecure workplace. Work, Employment and Society, 11, 713–735. Thurman, B. (1979). In the office: Networks and coalitions. Social Networks, 2, 47–63. van de Vliert, E. (1997). Complex interpersonal conflict management behavior. Theoretical frontiers. Hove: Psychology Press. Wielers, R. J. J. (1997). The wages of trust: The case of child minders. Rationality and Society, 9, 351–371. Wittek, R. (1997). Sociaal kapitaal and machtstrategie¨en in organisaties. Gedrag en Organisatie, 6, 357–371. Wittek, R. (1999). Interdependence and informal control in organizations. Amsterdam: Thela Thesis. Wittek, R. (forthcoming). Norm violations and informal control in organisations. A relational signalling perspective. In: B. Nooteboom & F. E. Six (Eds), The Trust Process within Organizations: Empirical Studies of the Determinants and the Process of Trust Development. Cheltenham: Edward Elgar. Wittek, R., Hangyi, H., van Duijn, M., & Carroll, C. (2000). Social capital, third party gossip, and cooperation in organizations. In: J. Weesie & W. Raub (Eds), The Management of Durable Relations: Theoretical and Empirical Models for Organizations and Households (pp. 100–101). Amsterdam: Thela Thesis. Wittek, R., & Wielers, R. (1998). Gossip in organizations. Computational and Mathematical Organization Theory, 4, 189–204.
