Intertemporal decision making studies on the working of myopia

Intertemporal Decision Making Studies on the working of myopia

Rijksuniversiteit Groningen

Intertemporal Decision Making Studies on the working of myopia

Proefschrift

ter verkrijging van het doctoraat in de Psychologische, Pedagogische en Sociologische Wetenschappen aan de Rijksuniversiteit Groningen op gezag van de Rector Magnificus, dr. D.F.J. Boscher, in het openbaar te verdedigen op donderdag 25 april 2002 om 16.00 uur

door

Alexander Ludwig Wolfgang Gattig

geboren op 23 october 1969 te Bremen, Duitsland

Promotores: Prof. Siegwart Lindenberg Prof. Jacques Siegers Prof. Charles Vlek

Manuscript Committee: Prof. George Loewenstein (Pittsburgh) Prof. Tom Snijders (Groningen) Prof. Frans van Winden (Amsterdam)

ISBN: 90 5170 603 0

Acknowledgements

This book summarizes research carried out when I was a Ph.D. student at the Interuniversity Center for Social Science Theory and Methodology (ICS) in Groningen. Though it has been written by a single author many people contributed to its completion and here I want to express my gratitude to them. First, I would like to thank my supervisors for the time and effort they took to improve this book but also for their patience and for the freedom they gave me to pursue research topics that I was interested in. Instead of beginning each chapter with the acknowledgement that their comments improved the quality of the text I would like to thank them here in the preface for their support. More specifically, I would like to thank Sigi Lindenberg for the numerous stimulating discussions with respect to theoretical issues. Further, he had a big impact on the structuring of the whole text. Jacques Siegers' quick and precise comments improved the accessibility of the text a lot. Although Charles Vlek was not involved in the initial stages of this project his impact on this book was profound. Not only are Chapters 6 and 7 results of our joint efforts, but his extensive and thorough comments on the other parts of the book made the text much more clear and substantially improved the reasoning. I also would like to thank the members of the Dissertation Committee for taking their time to read this thesis and for allowing me to defend it. Further, I would like to thank George Loewenstein, Drazen Prelec, Matthew Rabin, and Itamar Simonson for the time they took during my stay at Stanford to discuss several ideas concerning my Ph.D. thesis. Some others helped to improve this book. Maarten Reinink's support in designing the experiment reported in Chapter 7 and his help to collect the data for it were invaluable. Laurie Hendrickx provided competent comments on several parts of the manuscript. Jeroen Weesie helped to design the questionnaire used for Chapter 5 and he also improved the reasoning in this Chapter. However, the stimulating conversations with him also influenced many other parts of this book. I would like to thank all three of them not only for their help but also for the friendly atmosphere in which our conversations took place. I benefited a lot from my fellow Ph.D. students (past and present) at the ICS. Thus, instead of trying to single out individuals who contributed to the completion of this book (and forgetting others) I would like to thank them collectively not only for their professional support but also for sharing other activities with me than work, e.g., soccer, chess, or board games in general. Finally, though their contribution to this book was only indirect, I would like to thank my family for their support and patience.

Contents: Intertemporal Decision Making Chapter 1: Intertemporal choices 1. Introduction: intertemporal choices 1.1. The practical importance of intertemporal choices 1.2. The importance of intertemporal choices for rational choice theory 1.3. The Discounted-Utility model 1.4. Anomalies 1.4.1. Negative discount rates 1.4.2. Extra-ordinary high discount rates 1.4.3. Discount rates varying over time 1.4.4. Magnitude and sign effects 1.4.5. Different discount rates for different objects 1.4.6. Different discount rates due to framing effects 1.5. How to deal with these anomalies? 1.5.1. Denying the relevance of experimental results for actual behavior 1.5.2. Integration of the anomalies into neoclassical economic theory 1.5.3. Existing models to explain certain essential anomalies 1.5.4. Analogies between intertemporal choice and choice under risk 1.6. Problems treated in this book 1.7. Theory, general: framing and overview of arguments 1.7.1. Previous notions of framing in decision theory 1.7.2. Framing in sociology 1.7.3. Framing as used in this book 1.8. Application to intertemporal choice 1.8.1. Framing theory, the discrimination model and salience in intertemporal choice 1.8.2. Good-specific influences of patience: hedonic and utilitarian goods 1.8.3. Situation specific influences of patience: vividness 1.8.4. Individual determinants of patience 1.8.5. Outcome distance and discount rates 1.9. Outline of the book Chapter 2: Salience and the discrimination model in intertemporal choice 2.1. Introduction 2.2. A framing interpretation of ‘standard anomalies’ in intertemporal choice 2.3. Effects of salience on preference 2.4. Effects of discriminatory power on salience 2.5. Context effects in consumer choice 2.6. Linking context effects and similarity judgments 2.7. Overall conclusion and discussion

1 2 2 4 6 8 8 10 10 11 12 12 12 13 14 16 19 20 21 21 23 23 25 26 27 28 28 29 30 31 32 34 35 39 42 45 50

Chapter 3: Intertemporal choices for hedonic and utilitarian goods 3.1. Introduction 3.2. Theory 3.3. Method 3.4. Experiments and results 3.4.1 Conclusion 3.5. The influence of measurement on framing 3.6. Overall conclusion and discussion

53 54 55 56 57 59 60 64

Chapter 4: Vividness in intertemporal choices between hedonic and utilitarian goods 4.1. Introduction 4.2. Theory 4.3. Experiments and results 4.4. Overall conclusion and discussion

67 68 68 73 84

Chapter 5: Causes and consequences of patience 5.1. Introduction 5.2. Review and theory 5.2.1. Effects of parental characteristics on patience 5.2.2. Effects of family characteristics on patience 5.2.3. Effects of family environment on patience 5.2.4. Control variables 5.3. Tests of the hypotheses: subjects, procedure, and measurement 5.3.1. Subjects 5.3.2. Procedure 5.3.3. Measurement 5.4. Causes of patience: results for the cfc-measure 5.5. Consequences of patience 5.5.1. Measurement of the dependent variables 5.5.2. Consequences of patience: results 5.5.3. Discussion: patience as a general trait? 5.6. Patience for monetary outcomes (discount rates) 5.6.1. Measurement of patience by monetary discount rates 5.6.2. Corrections on the data 5.6.3. Causes of discount rates: results 5.6.4. Consequences of discount rates: results 5.7. Summary and discussion Appendix 5A: overview of causes of patience Appendix 5B: overview of consequences of patience Appendix 5C: consideration of future consequences scale Appendix 5D: money-scale

87 88 88 90 92 94 94 95 95 95 96 100 103 104 108 110 111 112 113 113 116 117 119 122 124 125

Chapter 6: Parallels in ‘anomalies’ in choice under risk and intertemporal choice 6.1. Introduction 6.2. Axiomatic models for EU, SEU and DU 6.3. Behavioral regularities in choice experiments

127 128 129 134

6.3.1. Anomalies in risky choice 6.3.2. Anomalies in intertemporal choice 6.4. Similarities between regularities in risky and intertemporal choice 6.5. Conclusions

134 140 144 146

Chapter 7: Generalizing ‘anomalies’ in choice under risk and intertemporal choice to other preference dimensions 7.1. Introduction 7.2. Spatial and social discounting 7.3. Effects investigated 7.4. Subjects and procedure 7.5. Stimuli and method 7.5.1 The interpersonal stimuli 7.6. Data analysis and control variables 7.7. Results 7.7.1. Consistency check for individual answers 7.7.2. Manipulation check for social distance 7.7.3. Common difference effects 7.7.4. Common ratio effects 7.7.5. Sign effects 7.8. Overall summary and discussion Appendix A: Preference distributions over choice problems

149 150 150 153 155 155 160 165 169 170 170 171 176 179 181 185

Chapter 8: Summary and discussion 8.1. Introduction 8.2. When do people act myopically 8.3. Who acts myopically 8.4. Generalization of effects 8.5. Measuring impatience 8.6. Suggestions for further research

187 188 189 191 192 193 193

References Appendix A: Questionnaire presented to the subjects in the first experimental session Appendix B: The logistic regression model Summary in Dutch

197 209 228 231

Chapter 1

Intertemporal choices

Summary In this chapter the importance of intertemporal choices in everyday life is underlined and the necessity for rational choice theory for modeling them appropriately is emphasized. Then the standard economic model and its anomalies are sketched, followed by a short review of responses to these empirical challenges. The research questions and a short overview of theoretical concepts used throughout this book conclude this chapter.

2

Chapter 1

1. Introduction: Intertemporal choices Do you prefer to receive $1000 now or $1200 in a year? Does your preference hold when the amount of money is substantively enlarged or reduced? Does preference change when the choice is between two consumer goods of corresponding value? Does it matter for your decision whether you yourself receive both goods or a friend gets both? Does it matter whether receipt of both alternatives is delayed the same amount of time? Why do some people invest in education or save substantial amounts of money while others do not and become drug addicts? Why are these questions important for social scientists? Except for the last one, these questions are central to decision theory in so far as it is devoted to intertemporal choice. Though already in the early 19th century similar questions as the ones posed above received scientific attention, until recently the three disciplines sociology, economics, and psychology approached these problems from quite different perspectives. If sociologists have realized the importance of this topic at all it is usually assumed, following Parsons (1937), that weighting of future costs and benefits is learned via socialization. By contrast, economics has - beginning in the 19th century with the investigation why there are positive interest rates (Rae 1834) - given more attention this question. In the beginning of this century, economic research on intertemporal choice culminated in Fishers (1930) theory of interest and in Samuelsons (1937) Discounted Utility model (in the following DUmodel) described in detail in section 1.3 below. Psychology, as the third social science interested in intertemporal choices, has headed by Mischel's work on delay of gratification in the 1960's explored intertemporal choice in more detail.1 Recently, interest in the behavioral mechanisms underlying intertemporal choices has increased in all three disciplines and there has been some convergence - at least for economics and psychology - in approaches towards these phenomena. In the following, how people balance current and future outcomes will be investigated by using arguments from all three disciplines. 1.1. The practical importance of intertemporal choices Decisions occur frequently, where costs and benefits spread over different time points have to be compared, e.g., in the choice between spending time for leisure vs. studying, whether or not to engage in healthy behavior, whether to have children now or later, or selecting appropriate health insurance and pension plans. Given the abundance and importance of intertemporal choices, neglect of future consequences potentially has profound impact on a variety of settings. Examples include investments in human capital and social resources such as networks or cultural resources, household relations; refusing to make relationspecific investments such as children, professional relations; jeopardizing the long term gains from a relation by shirking, or health; smoking or behave in otherwise unhealthy ways. Furthermore, stable individual differences in such investments may cumulate across domains and over time: people who continuously (relatively) neglect the future reach lower levels of wealth and a relatively worse health state.2 1

Loewenstein (1992) provides a historical overview and an elaboration how economical and psychological research on intertemporal choice related to each other. 2 As to evidence concerning the stability of such individual differences across settings Mischel (1968) concluded that though situational factors are more important in determining whether people are short-term

Intertemporal choices

3

Not only are intertemporal preferences potentially important in a variety of settings but there is also evidence for individual differences in the willingness to delay gratification, as well as for long-term effects of such differences. In a financial setting, Gately (1980) investigated consumer purchases, for example of air conditioners, where people had to make intertemporal trade-offs between the immediate purchase prices of a good and it’s long run energy costs. Discount rates for the long run energy costs were extremely high (up to 300% per year) and varied by 200% between individuals. Lawrence (1991) derives discount rates from expenses for consumption and reports differences in discount rates between races. As to consequences of differences in patience, Donkers and Van Soest (1997), using a nation-wide sample of Dutch households, demonstrated that, after controlling for income and family size, people more willing to wait for hypothetical rewards are more likely to own houses than less patient people. For educational achievements, Mischel, Shoda and Rodriguez (1992) report that children at age four, who were better able to delay gratification than their pre-school comrades, did significantly better at school more than a decade later. Moreover, these children turned out to be rated higher by their parents on such different abilities as social responsibility, achievement striving, verbal abilities, coping with stress and frustration, and intelligence.3 With respect to household behavior, impatient couples report more problems within and shorter duration of their relation (Snijders & Weesie 1998). Wunderink (1995) suggests effects of impatience on desired timing of births. As to deviant behavior, Hirschi and Gottfredson (1994) report correlations between several otherwise unrelated forms of deviant behavior such as car accidents, criminal behavior and drug abuse, and attribute this finding to a lack of self-control, which includes the ability to delay gratification. In addition, there exists a vast stream of research on intertemporal choices and addiction (Becker & Murphy 1988, Braun & Vanini 1998, Orphanides & Zervos 1998) or organizational behavior (Lindenberg 1995, O'Donoghue and Rabin 1999). Finally, differences in discount rates have been offered by Benartzi and Thaler (1995) as an explanation for the observation that people are much more willing to invest in bonds instead of stocks despite the latter offering much higher return in the long run (the 'equity premium puzzle'). They argued that when investing in stocks people sometimes experience (then) immediate losses. A strong focus on immediate outcomes in combination with loss aversion ('myopic loss aversion') thus might cause a preference for bonds. This reasoning was experimentally confirmed by Gneezy and Potters (1997). In the context of this research a study on consequences of impatience has been carried out. There, 165 students stated their concern about future consequences of their actions on a scale consisting of 12 items. Subjects relatively neglecting future consequences consumed more alcohol, more drugs, and were less likely to engage in sporting activities. In addition, these subjects did worse in a variety of topics related to education. For example, lower scores on the scale indicating less concern for future outcomes raised the probability of partying before exams, led to worse preparation of exams, reduced homework at high

or long-term oriented there are stable intraindividual differences. That is, subjects being more impatient than other in one situation are more likely to be more impatient in other situations. 3 Mischel et al. suggest (1992: 158), ‘that an early family environment in which self-imposed delay is encouraged and modeled also may nurture other types of behavior that facilitate the acquisition of social and cognitive skills’. Hence, more patient people should also be less aggressive and should engage in risky behavior less often. This view is consistent with the reasoning presented in this text and with results presented by Hirschi and Gottfredson (1994).

4

Chapter 1

school, and reduced the likelihood of following topics not required in a study. A more detailed account of the study and its results is provided in Chapter 5. 1.2. The importance of intertemporal choices for rational choice theory Besides the relevance of intertemporal choices for several otherwise unrelated streams of research, it is also important to study intertemporal choices for theoretical reasons. Rational choice theory has always been dominant in economics, but has recently also become prominent in sociology and political science.4 But despite its sophisticated theoretical elaborations, political scientists Green and Shapiro (1994: 6) criticize empirical applications of rational choice theory in political sciences as ‘either failed’ or ‘banal: they do little more than restating existing knowledge in rational choice terminology.’ Further ‘rational choice hypotheses are too often formulated in ways that are inherently resistant to genuine empirical testing' (Green & Shapiro 1994: 9). According to these authors, this is due to models that are too abstract. These models generally either lack statements about the relevant behavioral constraints in a situation or they model these constraints wrongly. Blossfeld (1996) provides a related critique from sociologists and Herrnstein (1990) is one of the numerous examples for a similar critique from psychologists. Even within economics recently there has been some concern with the strong concept of rationality as employed by mainstream economists, resulting in a growing interest to incorporate findings from psychology and behavior decision making (e.g., Hogarth & Reder 1987; Rabin 1998). Two reactions to such a critique are possible: First, abolish the approach (suggested by Herrnstein 1990), and second, extend simple (in terms of abstractness of the underlying assumptions) microeconomic theory by stepwise adding more complex assumptions about preferences, goals and cognitive imperfections. Not surprisingly, proponents of rational choice theory have chosen the second possibility. Examples include Lindenberg (1992) with his ‘method of decreasing abstraction’ for sociology, and Rabin (1998) or Laibson and Zeckhauser (1998) for economics. The economic theory of households and the family provides a nice example of this strategy and will therefore be briefly reviewed. The economic theory of households was developed in the 1970s (see Becker 1981, 1991 for an overview of method and theoretical insights) and allows for a consistent explanation of many of the associated phenomena by means of standard economic theory with more complex constraints and goals. In these models households allocate their resources (money and time) to maximize a household utility function given their constraints. This analysis was extended by game-theoretical modeling of household behavior (Manser & Brown 1980; McElroy & Horney 1981; Ott 1992). Here the joint household utility function is the result of a bargaining process of household members having different interests. Frey and Eichenberger (1996) introduced some complexities at the individual level (taken from findings in experimental economics and behavioral decision theory) in order to explain some puzzles in the household formation process. They argue that judgement biases in the assessment of probabilities are responsible for

4

Hechter and Kanazawa (1997) provide an overview of recent successfully empirical applications of rational choice theory to sociology.

Intertemporal choices

5

the underestimation of the probabilities of divorce and endowment effects5 result in too little search for and a biased evaluation of potentially better marriage partners. This present research aims at adding new constraints for intertemporal decision making. This is necessary for two reasons. First, given the evidence in section 1.4 below, in many situations current constraints are a poor description of reality and may lead to false predictions. Second, standard economic models by definition exclude many problems from study. Institutions, for example, serve as major means to overcome individual deviations from long-term rationality (e.g., Frey & Eichenberger 1996), an aspect of institutions usually neglected by economic theory. Lindenberg (1992) argued that often informal institutions (norms) guide behavior towards long-term goals while operating on the basis of short-term costs or benefits, such as social approval or disapproval. In addition, many individuals undertake action to limit their own possibilities for myopic behavior (Schelling 1992), for example. Standard economic theory has problems to explain these phenomena and has not given much attention to them (see Elster 1979, Schelling 1984, Lindenberg 1995). Assuming ‘rationality’ and the elimination of so-called ‘irrational’ behavior by the market, rules out all questions as to how and how efficient institutions and individuals promote rational (intertemporal) behavior. Consequently, Lindenberg (1998) explicitly stressed the necessity to develop an adequate theory of intertemporal choices: ‘At the present time, it looks like the most relevant foundation (for micro-sociology – AG) is research on short-term/long-term tendencies in decision making.’ Fortunately, in the last decades there has been an increasing interest in the positive validity of economic and decision-theoretic postulates. Anomalies6 to the axioms of subjective expected utility theory were found already in the 1950’s (Allais 1953) and several competing descriptive theories were developed (Kahneman and Tversky 1979, Luce and Fishburn 1991, Tversky and Kahneman 1992). Anomalies to other areas of standard economic theory received widespread attention in the last two decades, too. Experimental economics and behavioral decision making, both focussing on individual choice behavior, have become major streams in economics with their own journals (‘Journal of Experimental Economics, ‘Journal of Behavioral Decision Making’, ‘Journal of Risk and Uncertainty’) or regular sections on anomalous behavior (e.g., Thaler’s ‘Anomalies’ in the respected ‘Journal of Economic Perspectives’). Recently, also anomalies to the DU-model received considerable attention (e.g., Loewenstein & Elster 1992; special edition of the ‘Journal of Behavioral Decision Making [2000] on ‘Time and Decision’). Reviewing some of the anomalies Loewenstein and Prelec (1992: 574) conclude that the empirical evidence ‘present(s) a challenge to normative theory that is as least as serious as that posed by the more familiar EU (expected utility – A.G.) anomalies ... unlike the EU violations .... the counterexamples to DU are simple, robust and bear directly on central aspects of economic behavior.’

5

The endowment effect states that persons demand a higher selling price for objects which they already own than they are willing to pay in order to acquire the same object (in case of non-possession). This discrepancy grows with endowment time (Thaler 1980). 6 Anomalies, of course, are defined only against the axioms of standard economic theory. If a decision maker accepts these axioms as reasonable action principles, he should act in a certain way qualified as ‘rational’ – he may well behave differently, in case he does not feel that these axioms describe his preferences correctly. Economists usually regard these axioms as reasonable action principles for all people. Behavior departing from models based on these axioms then is called anomalous.

6

Chapter 1

Still there exist different viewpoints about how damaging these anomalies are to economic theory. Becker (1996: 22), for example states that ‘cognitive imperfections are sometimes important, but in recent years they may have received excessive attention at the expense of significant models of rational choice’. On the other hand, psychologists such as Herrnstein (1990: 356) concluded ‘rational choice theory.... accounts only poorly for actual behavior’ and should thus be abolished as a descriptive theory. In this dissertation, a mixed position is taken. Rational choice theory is seen as a useful means to analyze behavior but some assumptions will be investigated more closely and when being descriptively wrong they should be replaced with findings from behavioral decision theory (see Hogarth & Einhorn 1986; Rabin 1998, or Laibson & Zeckhauser 1998 for similar positions). By doing so, two main shortcomings of the many competing descriptive models in intertemporal choice presented in section 1.5 below can be avoided. The first shortcoming of these models is their limited range of application. For example, the prominent hyperbolic discounting model, accounting for procrastination of unpleasant tasks, is, despite of recent incorporation into axiomatic systems (Loewenstein & Prelec 1992; Ahlbrecht & Weber 1995), generally not related to other phenomena in intertemporal choice not to speak of phenomena outside of intertemporal choice. It would be more satisfying, however, if this hyperbolic discounting model could be related to other parts of economic and behavioral decision theory, e.g., S-shaped probability weighting functions or S-shaped utility functions. Second, competing descriptive models often use explanations unrelated to other economic assumptions. Akerlof (1991), for example, when investigating procrastination introduces the psychological term ‘salience’. While there is no general problem with explaining the delay of unpleasant tasks with changes in the salience of current and future costs and benefits, allowing for changing salience seems to be incompatible with the common assumption of stable preferences. The remainder of this chapter is structured as follows. First, the DU-model and predictions resulting from it will be shortly presented. Despite its descriptive shortcomings, the DU-model serves as a cornerstone against which competing models are tested. Second, existing anomalies to this model will be reviewed, and thirdly, current attempts to explain these anomalies and their shortcomings are analyzed. Finally, the central questions of this dissertation and general theoretical considerations will be presented. 1.3. The Discounted-Utility model While philosophers, e.g. Rawls (1971), have frequently stated that future utility should not be discounted at all, economic theory assumes that people have a time preference (e.g., Olson and Bailey 1981). That is, they have a preference with respect to when outcomes should be received. Generally it is assumed that time preference is positive, i.e., people prefer to receive positive outcomes, such as monetary amounts, as soon as possible. Negative outcomes should accordingly be delayed as much as possible.7 The degree of time preference is expressed by a discount rate, describing the percentage of increase or decrease that is necessary to make a person indifferent between immediate and future rewards or

7

There exist different justifications for this proposition. Sometimes it is seen as a general psychological mechanism sometimes as a rational reaction upon the uncertainties of life and sometimes it is a theorem derived from more elementary propositions (e.g., Koopmans 1960, Fishburn and Rubinstein 1982).

Intertemporal choices

7

losses.8 This concept was introduced by Fisher (1930) and further developed by Samuelson (1937). Later, axiomatic formulations have been developed, presented in some detail in Chapter 6, by, e.g., Koopmans (1960), Lancaster (1963), and, more recently, Fishburn and Rubinstein (1982). According to the DU-model introduced by Samuelson (1937), a decision maker should prefer a temporal sequence of consumption levels (c0, ......, cT) to another sequence (c’0, ......, c’T) iff: T

∑ δ t u(ct) > t =0

T

∑ δ t u(c’t),

(1)

t =0

with u being a utility function and δ being the constant 9discount factor per time period t.10 ct and c't are sequences of consumption levels beginning at t = 0 and ending at t = T. There is some disagreement about the relation between the discount factor δ and the market interest rate r. While in principle every δ is consistent with the model, Loewenstein and Thaler (1989: 182) state that due to the existence of capital markets, ‘consumers should make intertemporal trade-offs so that their marginal rate of time preference equals the interest rate’ for money. They argue that if time preference was higher than r, consumers could borrow to satisfy their preferences more adequately. Otherwise, consumers would be better of by saving. Consequently, time preference should also be equal for all goods and situations. Olson and Bailey (1981) conclude that in equilibrium 0 < δ < r, because capital can also be used for production and not just for consumption. An initial stock of capital x yields thus in a future period x plus the gains from productivity. The interest rate r has to account for this gain in productivity as well as for time preference δ. Hence, r will be larger than δ. Nevertheless, also Olson and Baily (1981: 9) state that: ‘The marginal rate of substitution between future and present consumption in long-run equilibrium will of course always equal the rate of interest’ for reasons elaborated by Loewenstein and Thaler. Lea et al. (1982) argue that discount rates should also depend on expectations about the future state of the economy, as well as on the expected future utility level of the decision maker. In this line of reasoning discount rates only have to respond to market interest rates. Practically, all these positions result in the prediction that discount rates at least for a larger sample of subjects should be closely related to the market interest rate. This prediction also underlies the familiar ‘present value’ concept and it is frequently used in economic modeling. For example, when modeling investments in 8

Discount rates may reflect other things than time preference. Suppose a person prefers $100 now over $110 in one year. He thus reveals a discount rate of more than 10% per year. This discount rate reflects his time preference, the market interest rate, the utility difference between the outcomes ($110 may provide the same utility as $100), uncertainty with respect to actually obtaining the outcome and the utility derived from the $110 in one year, and rational considerations (he may have an investment opportunity which in his perception yields more than 10% per year). 9 Constancy of the discount factor is not a requirement of the model but an additional assumption standardly made on normative grounds to prevent time-inconsistent behavior. If the discount factor varies over time an individual could depart from his or her initial long-term consumption plans without changes in prices, income, or information (see section 1.4.3 for evidence and discussion). Likewise δ usually is taken to be bounded by 0 and 1. 10 This actually describes sequences of outcomes. However, it follows from the additively separable form of this formula that sequences are simple aggregations of the single outcomes to which the axioms of Fishburn and Rubinstein apply.

8

Chapter 1

human capital, it is assumed that individuals take into account and discount returns on investment. The same holds for losses due to dropping out of the labor force. This discounting should equal the market interest rate (Mincer & Ofek 1982, Becker 1991 [1981]: 26). However, in the following section we will see that not only this strong implication, but also virtually every other prediction of the models as well as its axioms have been challenged empirically. The anomalies11 listed below were observed either in experiments using actual or hypothetical monetary payoffs or in field studies. The general pattern of the anomalies remained the same no matter which method has been chosen. The following anomalies have been demonstrated. 1.4. Anomalies 1.4.1. Negative discount rates Perhaps the most striking anomaly, given a positive market interest rate, the impatience axiom and given that ‘the case for positive time preference is absolutely compelling... both in the positive and normative senses’ (Olson and Bailey 1981: 1), is that sometimes people exhibit negative discount rates. That is, people prefer to delay positive outcomes or to receive negative outcomes as soon as possible. This even has been observed for decisions dealing with monetary outcomes. In discussing negative discount rates, two distinctions are useful: sequences versus single-events and monetary/non-monetary decisions. Negative discount rates have been observed in three of these four domains. a) Sequences of monetary decisions Loewenstein and Thaler (1989) and Lindenberg (1993) reported instances where people prefer to get their annual income paid in twelve monthly installments instead of receiving the same yearly income within nine months (with a higher payment per month in the latter case). Obviously, these people would have been better off financially, if they had chosen the latter mode of payment and had put the difference of the monthly payments in the bank. Further, Loewenstein and Sicherman (1991) reported that people have a ‘taste' for increasing wage profiles. In their experiment, subjects were given the choice between equal life earnings that could be received in either increasing or decreasing monthly payments. Most subjects preferred the increasing sequence and - although the percentage declined somewhat - most subjects still did so even after having been made familiar with the economic reasoning that, given a positive interest rate, present value maximization would imply preference for an (otherwise equal) decreasing wage profile over an increasing one. Related to this phenomenon is Frank’s (1992) observation that wages rise also in professions in which virtually no changes in productivity can be reasonably assumed, e.g. for bus drivers and airplane pilots. This rising profile can also only insufficiently be explained by the deterrence of employee shirking, because it occurs independently of the costs associated with shirking. Frank argues that employers, by offering jobs with rising wage profiles, react to preferences of potential employees.

11

Again, note that these findings only constitute anomalies only insofar as one is willing to subscribe to the axioms above.

Intertemporal choices

9

b) Sequences of non-monetary decisions According to Loewenstein and Prelec (1991: 349) ‘negative time preference is applied selectively, to those events that are seen as a part of a meaningful sequence, having a welldefined starting and ending point.' The consumption level of the end point of such a sequence exceeds the level of the starting point due to a wish of improvement. In an experiment, subjects expressed preference of a French dinner to a Greek dinner and also to get the former as soon as possible. Nevertheless they chose to receive the French dinner after the Greek dinner, if the interval between these two is short enough (one week) to make it appear as a sequence of events (Loewenstein and Prelec 1991, 1993). However, preference reversed as the interval between the two events was prolonged (half a year) such that both dinners were no longer perceived as a sequence. The explanation of a wish for improving sequences also applies to the Loewenstein and Sicherman observation. In addition, in a series of experiments Kahneman and his co-workers demonstrated that this wish for improving sequences possibly results in preference for sequences including more pain over sequences including less pain (Kahneman & Varey 1992, Kahneman, Wakker & Sarin 1997)). For example, people being exposed to the unpleasant experience of putting a hand in cold water for a short period of time rated this experience as more harmful than people putting their hand in cold water for the same amount of time plus a short period where the water was slightly warmed. Note, that the period where the water was slightly warmed still created an unpleasant experience for the subjects but led to an improvement (less pain). This improvement outweighed the additional length of exposure to unpleasant experience. Hence, preference for improving sequences appears to be fairly general.12 c) Single decisions for non-monetary events Loewenstein (1987) reports experiments where subjects were willing to delay favorable events such as ‘a kiss from your favorite movie star' a certain time. On the other hand, subjects preferred to be exposed to an unpleasant event (‘a non-lethal electric shock of a hundred and ten volt') immediately, instead of delaying it as long as possible. Loewenstein argued that the delay of positive events is due to ‘savouring’ while getting over with things quickly is caused by ‘dread’. Displaying negative time preference then actually maximizes utility over the whole sequence. In this dissertation not much attention will be given to negative discount rates. The major focus will be on single events, for which negative discount rates seem to be somewhat exceptional. Note, that according to Loewenstein’s models, procrastination (delay of unpleasant experiences) should occur when people do not ‘dread’ this experience. That is, appointments at the dentist are kept when people are afraid of the treatment and cancelled and delayed otherwise. This counterintuitive result might be explained by the fact that subjects in the Loewenstein experiment stated that they would like to be exposed to negative events soon but not now. When actually facing the negative event they might, however, want to delay it further. Here further research might be needed. 12

Chapman (1996) demonstrated that this does not hold for sequences of health outcomes. In a series of experiments subjects preferred to be healthy when they are young, i.e., sooner as predicted by the DUmodel, and delay sickness until old age. Chapman attributes this result to norms as well as subjects expectations: people usually are and expect to be healthy when they are young.

10

Chapter 1

1.4.2. Extraordinary high discount rates As stated above, revealed discount rates should approach, and in a perfect market even equalize the market interest rate. Discount rates much higher than the usual market interest rates thus challenge this model13. Unfortunately for the DU model such behavior exists. Holocomb and Nelson (1992) report an experiment with actual payments ranging from to ‘high' (17 $) to ‘low' (5 $). Subjects chose between immediate and delayed payments with an offered interest rate of either 1.5 or 3 percent per day. Preferring the delayed payoff, including interest, thus reveals a personal discount rate below the offered interest rate and vice versa. In spite of the extraordinarily high offered interest rates, the largest percentage choosing the delayed payment (tomorrow instead of today) was 54 percent (for the 17$ and the 3% interest rate). For all other combinations only a minority chose the delayed payoff, the lowest value being 18% for 1.5% interest rate of 5$. In other words, a majority (82%!) revealed discount rates above 1.5% per day and 46% revealed a rate higher than 3% per day. Somewhat surprisingly, Holocomb and Nelson concluded that ‘this was not a startling result'. The Holocomb and Nelson experiments replicated the results of Benzion et al. (1989) and Thaler (1981). Thaler (in a less careful study) detected even higher discount rates for hypothetical choices. Also in field settings higher discount rates than implied by the market interest rate occurred. Hausman (1979) reports an estimated average consumer discount rate of 25% derived from actual purchases of air conditioners. Discount rates were calculated by taking into account buying prices and delayed savings on the energy use of the conditioners. Gately (1980) reports even more extreme results. Since myopic behavior occurs frequently and poses a strong challenge to normative concepts of rationality, in this dissertation I will look more closely at when it is likely. 1.4.3. Discount rates vary over time This is maybe the best-documented and the most well-known anomaly concerning choice over time. According to the DU-model, to be called rational the discount rate should be independent of the time delay and thus result in an exponential discount function. Instead, data from experiments with both animals and humans shows that the discount rate varies with time delay (e.g., Ainslie 1992, Kirby 1995). Stevenson (1986, 1993) demonstrated that payoffs, as predicted by the economic models, are discounted at a ratio-rate. That is, delayed outcomes are devalued by a percentage of the amount and not by a constant premium for delay. However, usually, contrary to the DU-model, payoffs closer to the present are discounted at a much higher rate than those in the distant future. One important (and quite unpleasant) property of such behavior is that it gives rise to a socalled ‘dynamic inconsistency’ (Strotz 1956) or ‘preference reversal' (Ainslie 1975)14. Suppose, an individual compares two future payoffs, one smaller and sooner, the other 13

Throughout the text ‘high’ or ‘extremely high’ discount rates refers to discount rates much higher than the market interest rate. 14 The Ainslie term might appear strange, because ‘preference reversal’ usually describes varying buying and selling price for bets and lotteries (e.g., in Lichtenstein and Slovic 1973). When comparing two gambles of equal expected value, subjects bid more for the gamble with the higher winning probability, but charge a higher selling price for the gamble offering the larger amount. Thus, in the following time-varying discount rates will be described by hyperbolic discounting, common difference effect or dynamic inconsistency.

Intertemporal choices

11

one larger but later. If he initially prefers the latter, he might well prefer the sooner payoff as both payoffs approach the present. This behavior implies a tendency to deviate from one's own (well-understood) long-term interests and might thus be exploited (see Strotz for further elaboration). The rates may well decline over time when the individual has no possibility to reverse his initial choice, i.e., if his initial decision is binding. Related to this overweighing of immediate rewards is the phenomenon of procrastination (Akerlof 1991, O’Donoghue & Rabin 1999). Because of the ‘undue salience' of present costs relative to future costs, a series of decisions - each resulting in not undertaking an unpleasant task now, but to delay it by one time unit - might lead to disastrous consequences in the long run without being fully anticipated at the moment of each decision. For the relevance of this ‘myopic' behavior for a classical topic of sociology (organizational sociology) see also Lindenberg (1993, 1995). In this book we will have a closer look when discount rates vary more or less strongly. 1.4.4. Magnitude and sign effects There is convincing evidence that people discount small payoffs much more strongly than large payoffs - independent of the payoff sign (Thaler 1980, Benzion et al. 1989). This partly explains the high discount rates obtained by Holocomb and Nelson (1989). In addition, this implies that discount rates derived from experiments with actual payoffs using small amounts overestimates time preference. Further, the discount rate for gains and losses is different. Because losses loom larger than gains (Kahneman &Tversky 1979, Tversky & Kahneman 1991), they are discounted less strongly (Thaler 1981, Loewenstein 1988, Benzion et al. 1989).15 Since the gain-loss asymmetry is strongest for small outcomes, it can not be due to decreasing marginal utility, since for small outcomes the utility function is (almost) linear. For larger outcomes, observed discount rates might be higher than the pure rate of time preference, because the difference in utility between an immediate small and a delayed larger reward is smaller than the numerical difference between the two. Hence, time preference might differ from observed discount rates but this difference should be wider for large outcomes and should even disappear for very small outcomes, when the utility function is almost linear (Loewenstein & Prelec 1992). Note, that magnitude and sign effects do not necessarily constitute ‘anomalies’ to the DU-model or other normative models, since these model refer only to utility. Individuals are free to derive whatever utility they want from varying amounts. Because the magnitude effect in intertemporal choice is satisfactorily explained with an appropriate utility function (e.g., Loewenstein & Prelec 1992) it will not be further investigated in this book. The same holds for the gain/loss asymmetry. However, different value functions for gains and losses will be assumed in several of the following chapters. In addition, it will be investigated whether magnitude and sign effects can be generalized to other preference domains.

15

Shelley (1993) argued that the gain/loss asymmetry results from an interaction of outcome sign and time frame. Differing discount rates are observed when subjects have to delay either gains or losses which are otherwise due immediately. Subjects thus give up something now for the later, and thus discounted, receipt. In a ‘neutral’ question format, not involving delays, the gain/loss asymmetry disappeared in Shelley (1993) as well as in Ahlbrecht and Weber (1997).

12

Chapter 1

1.4.5. Different discount rates for different objects According to economics, people maximize utility given their constraints, such as time and money. Different goods as well as different forms of money, e.g., lottery wins, income, bonuses, inheritances, are aggregated into final states of wealth. They are ‘fungible’. Consequently, the proportion of money spent immediately for consumption and money saved should be identical for a rise in income or a lottery win of equal value. However, Thaler (1990) and Shefrin and Thaler (1992) report that different forms of money had different marginal propensities to consume. A rise in income is much more likely to be spent for consumption than a rise in the savings for retirement. Akers et al. (1994) found that unanticipated money, ‘windfall gains’ was more likely to be spent immediately than money that was anticipated. The speed with which money is converted into consumption can be interpreted as being determined by discounting, thus resulting in different discount rates for equal amounts of fungible goods. Chapman (1996) found that discount rates for health, money and vacation trips differed even when controlling for utility. Empirical results on impulse buying (unplanned purchases which are regretted afterwards) also show differences in the likelihood of different goods to be purchased impulsively (Bellenger 1978, Dittmar et al. 1995). To sum up, there is ample evidence that discount rates differ between commodities of equal value. Despite the examples above, this anomaly has been relatively neglected in research on intertemporal choice. Therefore in this dissertation it will be investigated which goods are most likely to be discounted strongly. 1.4.6. Different discount rates due to framing effects In the DU-model only absolute time differences matter for the evaluation of outcomes. Therefore, jointly shifting the receipt of two different outcomes available at two different time points should not alter preference between them. As seen above, this runs counter to empirical evidence. The DU-model also implies that speeding up the receipt of an outcome and delaying its receipt by the same amount of time have a numerically identical effect on the evaluation of this outcome. Suppose a decision maker has a discount rate of 5% per year. Then, speeding up-the receipt of a good by one year results in a 5% higher value, delay of one year to 5% devaluation. However, postulating different value functions for outcomes that are framed as gains and losses (Kahneman & Tversky 1979, Tversky & Kahneman 1991) leads to a different prediction. When benefits are delayed subjects give up (lose) something now in order to gain it later, they compare an immediate loss to a future gain. This gain is discounted. In a speed-up situation, they gain something now and suffer from a later discounted loss. In the delay case, they require compensation whereas in the speed-up case they are willing to pay a premium. Loewenstein (1988) demonstrated that, due to different value functions for gains and losses, subjects charge more to delay video recorders, restaurant dinners, or money receipts than they are willing to pay for accelerating their receipt (‘delay-speedup asymmetry’). This finding has been replicated in studies by Shelley (1993) and Shelley and Omer (1996). 1.5. How to deal with these existing anomalies? The following overview summarizes several attempts to explain these empirical results and focuses on current streams of research. It contains the following sections: a) questions concerning the relevance of experimental research for actual behavior,

Intertemporal choices

13

b) integration of the anomalies into neoclassical economic theory, c) relaxing assumptions of neoclassical economic theory in order to build competing models explaining certain anomalies, d) analogies to peculiarities to human choice under risk. 1.5.1 Denying the relevance of experimental results for actual behavior Economists have up to now followed two strategies in order to generally reject the importance of findings demonstrating the existence of 'anomalies'. First, it has been argued that economics only deals with rational, i.e., consistent and utility maximizing, behavior - the above listed examples are thus excluded by definition. However, now there have been various attempts to demonstrate that even love, addiction, etc. though appearingly irrational can be well handled within economic approaches (for example in Becker’s work over the past decades) and hence, rational choice theory is of use for sociologists and political scientists, usually interested in appearingly 'irrational' behavior, as well. But if the economic approach is used as the leading paradigm, its anomalies and limitations are of central importance for a variety of disciplines. Second, anomalies may be taken seriously at the individual level, therefore being a very interesting topic for psychologists, but irrelevant on an aggregate level, thus being of no importance for economists. According to Frey and Eichenberger (1989a, 1989b) it might be claimed that: - anomalies at the individual level are irrelevant, if one assumes that deviations from the economic model follow a random distribution; - anomalies are only observed in experiments and, due to this ‘unnatural’ setting, have nothing to do with actual, i.e. real world, decisions; - the observed anomalies are due to insufficient incentives for serious behavior; - anomalies may exist, but in reality people learn from and thus avoid errors; - anomalies are eliminated by competitive markets - irrational actors score less well and thus disappear in the long run. As to the first argument – budget constraints drive out randomly distributed deviations from rationality at an aggregate level (Becker 1976 [1962]) - it can be stated that the above mentioned anomalies are far from being random. Instead, they reveal systematic behavioral tendencies, so this argument does not apply here. The same holds for the argument that the anomalies are caused by insufficient incentives for rational behavior. High incentives tend to reduce careless errors, but in general subjects behave similarly in case of high incentives. Camerer (1995) reviews evidence on this matter observed in a variety of contexts such as ultimatum games, experimental auctions or choice under risk and uncertainty and concludes that though it is likely that there are situations where incentives affect results, such effects seem to be the exception rather than the rule. Perhaps the strongest evidence against a general disappearance of anomalous effects in decision making when payments are involved comes from a study done by Kachelmeier and Shehata (1992) who investigated behavior under risk with Chinese subjects (Masters students with a background in economics). Although these subjects in some experiments could earn three times their monthly income, similar departures from neoclassical economic theory as in previous experiments occurred (Beattie & Loomes 1997 reach similar conclusions on the general effect of payments on choices under risk).

14

Chapter 1

Further, these anomalies are not restricted to experiments but were also found in ‘real life situations' (e.g., Lichtenstein & Slovic 1973, Hausman 1979), in situations where people are given the possibility to learn (e.g., Herrnstein, Loewenstein, Prelec & Vaugham 1991, Herrnstein & Prelec 1992) or in situations where experts (with sufficient possibilities to learn) make decisions (again Lichtenstein & Slovic 1973). Thaler (1987b, 1987c) and De Bondt and Thaler (1985, 1987, 1990) even report cases of anomalies in financial markets where one can quite safely assume expert actors, high incentives, and almost perfect competition. Anomalous behavior seems to do quite well under certain conditions even in highly competitive markets. Summarizing the above arguments Frey and Eichenberger (1987: 15) conclude: ‘serious theoretical reasons and empirical observations exist for the belief that in economic markets as they exist in reality -....- individual anomalies are not eliminated at the aggregate level'. They should therefore be taken seriously in economics as well as in sociology. This statement underlines that for certain problems a more complex behavioral theory is needed. 1.5.3. Integration of the anomalies into neoclassical economic theory Attempts to integrate the observed anomalies (for intertemporal choice) into neoclassical economic theory up to now only deal with certain anomalies like procrastination or varying discount rates. To Anderson and Block (1995) the problem of procrastination is due mainly to a misrepresentation of the individuals ‘true' preferences and not to a failure of their ‘true' utility maximization.16 They argue, that if people procrastinate or give in to a temptation, they do not fail to maximize utility, but have a couple of reasons to do so: They could want to make themselves or others believe they would have liked to make a different choice, they also could have other goals in a situation beside resisting this certain temptation (e.g. impress others) and finally, their ordering of preferences could have changed. For example, imagine a person fails to stick to his diet or to abstain from taking alcohol. This failure can, according to Anderson and Block (1995), be due to an only temporary wish to impress a person (maybe the spouse) which becomes less important as another goal appears. The failure might also be due to a strategic misrepresentation of the dieter’s preferences (i.e. the dieter did never really want to engage in this diet, but stated this wish only to keep the peace) and finally it can be due to a change in preferences, because ‘[t]he standard economic model assumes that individual preferences are consistent at any moment, yet not necessarily over time' (Anderson & Block 1995: 205). Therefore, people procrastinate or give in, because, given their set of preferences, it actually is the best they can do. So there clearly exists no problem for economic theory at all. Although anomalous phenomena may be due to a strategic misrepresentation of preferences, given the serious attempts that are sometimes made and the often quite high costs that are involved with these attempts, it seems rather implausible that people value, say, the next alcoholic drink more than the costs for a three-week therapy or undergo such a therapy for strategic reasons. The growing literature on self-command and commitment (e.g., Elster 1979, Schelling 1992, Ainslie & Haslam 1992) indicates that people are willing to pay quite significant costs in order to raise incentives not to deviate from their long-term 16

One could argue that they deny the relevance of anomalies and thus should be placed in the section above. However, that section reviews arguments accepting experimental anomalies but questioning their importance for aggregates in real world settings. Thus, an approach trying to prove that these results can be explained in standard economic terms is somewhat distinct.

Intertemporal choices

15

interest in the light of short-term temptations.17 Further, it is strange to explain violations of the standard micro-economic model by temporary changes in preferences. For good reasons preferences should be regarded as ‘similar among persons and stable over time' (Stigler & Becker 1977: 76). Abolishing this assumption would lead to the possibility of ‘explaining' almost every action by a sudden - and maybe even temporary - change in preferences - as it is done by Anderson and Block, thus rendering a systematic testing of predictions impossible. In this case the remedy seems to be much worse than the disease. Another, more sophisticated, attempt to integrate part of the observed phenomena is via endogenous changes in preferences (Becker & Murphy 1988, Becker 1996, Becker & Mulligan 1997; Wathieu 1997 offers a similar approach). Here present consumption of a good influences the utility derived from its future consumption. For example, drug use now raises current utility levels but in addition alters (lowers) the utility derived from consumption of the same amount of drugs later. Becker and his co-workers introduce a good-specific stock of consumption capital, which, as in human capital theory, depreciates when not used. Different goods have different rates of depreciation as well as different influences on the utility derived from future consumption. For example, immediate drug use influences future utility from drugs stronger than immediate consumption of chocolate influences future utility from chocolate. Note, that this influence of present on future consumption need not be harmful. Exercising (jogging, doing piano lessons) potentially raise utility of future consumption and thus are positive addictions.18 Next to the initial rate of good-specific consumption capital, the depreciation rate of this capital, and the influence of current consumption on future consumption, intertemporal utility then depends on discount rates. People discounting strongly will presently consume more of harmful goods despite the negative future consequences. Therefore, subjects may maximize intertemporal utility even in case of addiction. According to Becker and Murphy (1988) utility maximization does not imply that these persons achieve very high levels of utility, i.e., that they are very happy, but only that given their set of preferences and their consumption capital they behave optimally. In addition, the initial choice of experimenting with drugs or alcohol might very well be made under uncertainty (Becker 1992). When deciding to experiment with the use of drugs or not, it may simply not be clear whether one will become an addict or not. Thus, people might become addicted and later regret that fact and also their previous choices, but they still acted rational. Therefore people may undertake considerable pain in order to stop addictions.19 This model raises interesting research questions, e.g., which people have a high time preference and are at risk of becoming addicted and where such high time preferences stem from.

17

Note, that the prime example for the use of commitments – Ulysses letting himself bind to the ship’s mast in order not to fall prey to the singing of the sirens - used commitments much more efficiently than his modern descendants. Contrary to their attempts to remove chocolate bars in order to stick to their diets, Ulysses is able to reap the benefits of giving in to a temptation, without having to pay the price for it. 18 Because investment in beneficial addictions is no potential anomaly, the focus in this chapter lies on the more familiar harmful addictions. 19 Note, that this model incorporates addictive behavior but no longer assumes that intertemporal utility function is time-separable. However, with regard to the violations of the separability axiom, such a violation is not necessarily a defect of the model. Ironically it also runs counter to Becker’s own theory of the family where he assumes that people discount future benefits at the market interest rate.

Chapter 1

16

The problem with this model, despite its correct observation that addicts sometimes regret their choices, is its neglect of the ongoing shifting preferences of addicted people and their sometimes even repeated tremendous efforts and expenses to stop their addiction. If subjects have already been addicted, they should be able to judge the risks associated with drug much better than when making their initial choices. In addition, it remains unclear why people at one time prefer to stay away from taking harmful substances and next time to succumb to the need for these substances. 1.5.4. Existing models to explain certain essential anomalies Due to a plethora of competing models this section could be easily enlarged to a whole chapter. It will not, because generally these models do not cover a broader range of anomalies but are restricted to the anomaly under study (e.g., Akerlof's [1991] model of procrastination). As a result, they sometimes look like searching for an appropriate utility function. In this section only more general models will be reviewed. The most prominent model and a major descriptive advance over the DU-model is the well-confirmed hyperbolic discount curve (e.g., Ainslie 1975, 1991; Mazur 1987): U (V 0) , (2) (1 + αt ) U(V0) being the immediate value of an outcome, U(Vt) its value at time delay t, α being a constant determining the steepness of the discount function, and t the number of time periods the reward is to be delayed. 20 U (Vt ) =

Contrary to an exponential discount function, hyperbolic discount rates imply time-varying discount rates where the decision maker is relatively strongly aware of short delays. Due to the prominence of this ‘anomaly’ this hyperbolic discount function has been incorporated in several models as well as axiomatic systems. For example, Loewenstein and Prelec (1992), Harvey (1986) and Ahlbrecht and Weber (1995) developed axiomatic systems where the stationarity axiom is replaced with an axiom allowing hyperbolic discounting. Ahlbrecht and Weber (1995) further argue that hyperbolic discounting is normatively justified, i.e., rational, in cases where decisions are binding, that is, if current decisions are not reevaluated in the future. Certainly any successful theory of intertemporal choice must incorporate hyperbolic discounting. Yet, it remains quite unclear why certain things (clothes, alcohol or sweets) are usually related with ‘impulsive behavior' (Ainslie's term for the 20

There exist a variety of competing models for such a hyperbolic discount function, e.g., Harvey 1986, Mazur 1987; Loewenstein and Prelec 1992; Laibson 1997. The most general one, having been successfully tested also with survey data (Cairns & van der Pol 2000), is that of Loewenstein and Prelec (1992). They define decision weights (wt) for future time periods as follows:

(wt) =

1 . Depending on the parameter values for h and α this formulation encompasses standard (1 + αt ) h / α

exponential discounting, if α approaches 0, as well as Mazur's (1987) model above, if h / α equals 1. Note, that in Mazur's formula above as well as in the Loewenstein and Prelec model α may take negative values, indicating that a person values positive future outcomes more than immediate ones. If a person is indifferent between immediate and future outcomes, in the Loewenstein and Prelec model all parameters equal 1, while in the Mazur model α has to equal zero.

Intertemporal choices

17

overweighing of immediate gains in his seminal 1975 paper), while others, such as the decision which bank offer to accept in order to pay back one’s mortgage for the following twenty years, are not. Consequently, Ainslie (1991: 335) raises the question: ‘But then, which is the ”true” discount function - the matching law (implying hyperbolic discounting – AG) or the exponential discount rate available from the bank?’ Unfortunately, he leaves the problem unresolved by answering that this depends on the circumstances, and focuses instead on the means of arriving at a proper exponential rate. To specify such circumstances is indeed one of this book’s aims: who is most likely to behave myopic and when? While hyperbolic discounting accounts for time-varying discount rates the anomalies concerning the sign and magnitude effect are modeled via the value function for gains and losses. Kahneman and Tversky (1979) introduced a value function that is split by a reference point in two parts. Above the reference point the value function is concave, below the reference point it is convex.21 In addition, the function is steeper in the negative domain, resulting in an S-shaped utility function. Loewenstein and Prelec (1992) applied a similar value function in their model for intertemporal choices. Besides for the sign and magnitude effects, this value function also accounts for anomalies in sequences and the delay/speed-up asymmetry. With respect to sequences, Frank (1992) points out that people react to events and evaluate them in relation to some reference point, mostly the status quo. Past levels of consumption form reference points for future levels of consumption (Kapteyn & Wansbeek 1982), thus leading to a preference for sequences with growing consumption profiles as reported by Frank (1992) and Loewenstein and Sicherman (1991). As to the delay/speed-up asymmetry, Loewenstein (1988) found that charges to delay a good are higher than premiums people are willing to pay for speeding-up the receipt of the good the same amount of time. In delay situations people experience an immediate loss compensated by a later but discounted gain. In a speed-up situation they experience an immediate gain followed by a later and discounted loss. If losses weigh more heavily than gains, delay charges are higher than speed-up premiums. Unfortunately, the specification of the (quite important) reference point and the ensuing distinction between gains and losses remain unspecified. Hence, only in certain situations can predictions be derived about when an outcome is perceived as a gain and when as a loss. It seems unlikely, for example, that the average shopkeeper views his daily sales as a loss and that the amount of money he gained in exchange is unable to compensate him for this loss. Loss aversion seems to be of no importance in this situation. Fischhoff (1983) investigating the stability of perceived reference points found that subjects differed remarkably in the perception of the relevant frame in a situation. The missing specification of what constitutes the reference point is also responsible for the contradiction that Loewenstein (1988), in his work on delay and speedup, states that anticipated consumption serves as the reference point. By contrast Loewenstein and Sichermann (1991) as well as Frank (1992) in their work on increasing wage profiles pose that the current level of consumption serves as the reference point. While both explanations seem intuitively plausible they are certainly not derived from theory. 21

Such a utility function does not exhibit decreasing marginal utility in the negative domain, i.e., decreasing marginal disutility. Consequently, economists frequently question the S-shaped form of the utility function. For empirical evidence favoring S-shaped utility and a discussion see Fennema and Van Assen (1999).

18

Chapter 1

Worse, capturing the above-mentioned effects (sign, magnitude, delay/speed-up asymmetry and negative discount rates in sequences) by choosing an appropriate value function implies that objects valued equally are discounted equally, no matter whether it is drugs or garden furniture. Chapters 3 and 4 will provide evidence against this prediction, by assuming, among other things, that value functions in the negative domain differ between goods according to their hedonicness (Dhar & Wertenbroch 2000). As to the phenomenon that different objects (here: forms of money) are discounted in a different way, Thaler (1990) and Shefrin and Thaler (1992) argue that people ”act as if they used a system of mental accounts... Specifically some mental accounts, those that are considered ‘wealth’ are less tempting than others that are considered ‘income” (Shefrin & Thaler 1992: 290). They view mental accounting as part of framing (Kahneman & Tversky 1984) and predict the mental account ‘income’ to have a higher marginal propensity to be spent for consumption than ‘lump-sum payments’. These in turn have a higher marginal propensity than ‘pension increases’. They also predict a much higher marginal propensity for so-called ‘windfall gains' (i.e., unexpected gains), if these gains are small and a lower marginal propensity if these gains are large, because in the latter case they are framed as ‘wealth'. Although a precise distinction between ‘small' and ‘large' windfall gains might be quite debatable in practice, the prediction that windfall gains have a higher marginal propensity than expected gains has been successfully tested (Akers et al. 1994). Note, that in this section ‘framing’ has been used for a variety of different phenomena. Frank (1992) and Loewenstein and Sicherman (1991) assume that people adopt a reference point that influences the evaluation of outcomes depending on whether they are perceived as gains or losses. Loewenstein (1988) does the same, but additionally shows that the perception of the reference point can be manipulated. Shefrin and Thaler (1992) further add complexity by introducing their system of mental accounts for which discounting is not only depending on the reference point but also on relating an outcome to a specific mental account which in turn is related to a specific discount rate. While these models have a much higher positive validity than the DU-model, they still leave many questions open. I already dealt with the problems related to the specification of the reference point. As to the existence of mental accounts that influence the discount rate, other problems remain. If mental accounts are crucial, an explanation of how many of them exist for a certain object and which accounts are associated with which discount rates would be very desirable. Unfortunately, Shefrin and Thaler tell us nothing about this problem. Second, and even more important, their model assumes the ordering of mental accounts to be static. This implies, that if people reveal much higher discount rates for alcohol than for money, say, when celebrating in a pleasant environment at parties, the same ordering still holds if an alcoholic drink is offered at a conference. This implication does not seem very plausible.22 Another stream of research focuses on the relation between intertemporal choice and choice under risk and compares models between the two domains. Because this approach to intertemporal choice differs from building descriptively more valid models, it will be covered separately in the following section. 22

It is of course also possible to assume that events are linked to mental accounts, thus allowing for identical forms of consumption to be linked to different mental accounts and to be discounted differently. This makes the need to systemize mental accounts and their relation to temporal discounting even more urgent.

Intertemporal choices

19

1.5.5. Analogies between intertemporal choice and choice under risk

There is a striking similarity between most of the observed anomalies concerning choice behavior under risk and over time. That is, virtually the same violations of the standard models occur (see Chapter 6 for a demonstration). Based on this similarity, Prelec and Loewenstein (1991) as well as Quiggin and Horowitz (1995) argue that basically the same behavioral patterns underlie the deviations from the normative models. In both cases, alternatives are described by the notation (X, x), X being the payoff and x being either a time delay or a probability/degree of risk. Both choices (over time and under risk) are sensitive to the existence of reference points, i.e., it matters for the decision-maker whether the given alternatives include gains or losses. Further, in both cases the decisionmaker is sensitive to manipulations of the time/probability parameter that normatively should not affect preference. Thus, in both choice domains there is a preference for sure/immediate events and in both choices an initial indifference between two outcomes disappears, if the time/probability component is altered by the same ratio or the same difference for each outcome. For example, a person being indifferent between a smaller but less distant and a larger but more remote reward, prefers the latter one if both choices are delayed the same amount of time. Prelec and Loewenstein (1991) describe this as the common ratio/common difference effect. Finally, in both preference domains (under risk and over time) the same effects with regard to sign and magnitude of the reward can be observed. Prelec and Loewenstein (1991) conclude that these similarities reflect the intimate connection between intertemporal and risky choice: delayed objects are almost by definition uncertain. The observed regularities therefore do not necessarily hold for other multi-attribute choices. In contrast with this view Rachlin and Rainieri (1992) generalize the hyperbolic discount functions observed in intertemporal choice to the discounting of risk, social distance and other. Their models include important aspects of intertemporal choice, such as hyperbolic discounting or utility functions differing between gains and losses. But two important problems remain. First, as expressed in the Ainslie quotation above, discount rates are highly dependent on situational characteristics.23 Otherwise different goods of equal value should be discounted identically. Such situational aspects are completely left out. Consequently, they cannot explain different discount rates within an individual. Second, there exist also personal differences in discounting (Mischel, Shoda and Rodriguez 1992). Some people manage better than others to stick to their long-term consumption plans than others as every member of AA can tell. While in principle compatible with the models above, after being to some extent investigated in the 1950s and 60s such personal differences have been largely neglected in the past decades.

23

Even Loewenstein claims that ‘As a research agenda it may be more fruitful (than investigating whether time preference is positive or negative - AG) to adress the more nuanced question of why intertemporal choice behavior is so unstable and to discover the situational determinants of time preference’ (Loewenstein & Sicherman 1991: 81/82).

20

Chapter 1

1.6. Problems treated in this book

In the previous sections the importance of intertemporal choices was stressed, the DUmodel underlying most economic analysis and its accompanying anomalies were presented and some alternative models were reviewed. In this section the main problems treated in this book will be outlined briefly. Existing models in intertemporal choice were criticized for paying too little attention to varying discount rates within an individual.24 Hence, influences leading to high or low discount rates respectively will be specified. More specifically, it will argued that people act myopically when hedonic consumption is considered and that this short-term orientation is enhanced when consumption experience can be imagined vividly. By contrast people act according to their long-term interest when utilitarian considerations are prominent. Secondly, it was repeatedly underlined that existing models in intertemporal choice are agnostic with respect to individual differences in discount rates. Such differences usually are regarded as differences in tastes and hence, taken as given. In this book, we attempt to demonstrate that the ability to take future consequences of one’s actions into account is an asset produced by parents within their children. People are naturally shortsighted (see Ainslie 1992 for support) but are taught, mainly by parents, to take long-term consequences of their actions into account. Because parents differ with respect to abilities and restrictions to teach this ability and because people differ with respect to which they can learn this ability, people should differ systematically with respect to temporal discounting. Therefore, the first two problems treated in this book are: When do people act myopically and who behaves impatiently?

In addition, existing models in intertemporal choice were criticized for being unnecessarily restrictive. In this book it is argued that behavioral regularities observed in intertemporal choice cannot only be applied to choice under risk, as to some extent has already been demonstrated by Prelec and Loewenstein (1991), but can also be generalized to other preference dimensions, e.g., to interpersonal or interspatial choice. Thus, the third problem treated in this book is: How can the empirical results observed in intertemporal choice be generalized to other preference dimensions?

It will be argued that processes in temporal discounting are first much more specific than postulated by existing models, because these models do not take into account varying discount rates within individuals. It will also be argued that discounting is much more general than postulated in existing models: outcomes distant from a decision maker’s reference point are discounted in decisions concerning other preference dimensions than just temporal delay. 24

Recently, O'Donoghue and Rabin in a series of papers (O'Donoghue & Rabin 1999, 2000, 2001) as well as Laibson (1997, 1998) - all following suggestions from Schelling (1984) and Ainslie (1992) - used gametheoretical modeling to explain some of the empirical deviations of the DU-model. They argue that people consist of different selves that interact strategically to determine the behavior of a decision maker. From this assumption they also derive results as to when people are likely to behave impatient. The arguments developed in this book appear to be compatible with their reasoning.

Intertemporal choices

21

To tackle these problems, in the following sections a framing theory developed by Lindenberg will be presented which serves as an integrating framework to answer the two main research questions. It will also briefly be sketched how this theory is applied to answer the questions. 1.7. Theory, general: framing and overview of arguments

Because the research questions mentioned in the previous section will be tackled by using a framing theory, this section first briefly reviews other concepts of framing before elaborating the theory used throughout this dissertation. 1.7.1 Previous notions of framing in decision theory

Framing effects have been introduced into decision theory (for choice under risk) by the psychologists Kahneman and Tversky (1979). They developed a descriptive theory of choice under risk, Prospect Theory (PT), the central concepts of which have also been applied to intertemporal choice (e.g., Loewenstein 1988, Loewenstein & Prelec 1992). In Prospect Theory the value V of a prospect25 P is modeled by: V (Pi) = v(xi) δ(di)26 (3) v(xi) being a value function and δ(di) being a weighting function for probabilities Due to the existence of a reference point (usually the status quo) the value function implies that decision makers distinguish between gains and losses. Further, the function is concave above the reference point (for gains) and convex below it (for losses). It also is steeper for losses than for gains. These properties hold for risky as well as intertemporal choice. The probability weighting and intertemporal discounting function also show some similarities, viz. diminishing sensitivity with greater distance from the reference point. That is, changes on the risk or time attribute have less influence on the evaluation of prospects the more distant from the reference point they occur. For choice under risk this reference point can either be 'certainty' or 'impossibility', while for intertemporal choice it will generally be 'now'. Prospect Theory additionally assumes that there is an editing phase consisting ‘of a preliminary analysis of the offered prospects which often yields a simpler representation of these prospects’ (Kahneman & Tversky 1979: 274). In this phase the following four operations take place. Coding: Instead of integrating outcomes into final states of wealth, outcomes are usually perceived as gains or losses. When an outcome is perceived as a gain or a loss depends on a given (chosen or suggested) reference point.

25

Prospects are vectors either of the form (x1, p1; x2, p2; ...; xn, pn) for risky options and (x1, t1; x2, t2; ...; xn, tn) for intertemporal options. Xi describes the outcome attribute i in both prospects whereas pi and ti describe the corresponding risky or temporal attribute. A gamble offering a 50% chance to win $100 and a 50% chance to win nothing is thus described as (100, .5; 0, .5). For matters of convenience zero outcomes are usually omitted, hence the prospect reduces to (100, .5). The procedure for intertemporal options is analogous but note that in risky choice outcomes usually are mutually exclusive whereas in intertemporal choice this need not be the case. 26 This holds for prospects consisting of one outcome differing from zero and one risk or time attribute. For more outcomes the formula becomes more complex but the main properties remain unchanged.

22

Chapter 1

Combination: The probabilities of identical prospects can sometimes be combined. A prospect of (200, .25; 200, .25) will thus be combined and evaluated in the form (200, 50). Segregation: Some prospects contain a riskless component that is segregated from the risky component of the prospect. The prospect (100, 40; 300, 60) will thus be perceived as a prospect with a sure win of 100 and a 60% probability of winning an additional 200. Cancellation: Shared components of the offered prospects are cancelled. Thus a choice between (A, x; B, y; C, z) and (A, x; B', y'; C', z') will be reduced to a choice between (B, y; C, z) and (B', y'; C', z'), since (A, x) is a common component in each of these prospects. Elimination of stochastically dominated27 prospects also occurs during the editing phase. Kahneman and Tversky (1979) provide some empirical evidence for the existence of such an editing phase. However, no such effects up to now have been observed in intertemporal choice. Seidl and Traub (1998) report little evidence for the existence of an editing phase in choices with multiple attributes. To summarize, in Prospect Theory framing effects occur at several steps. They first influence the perception of the prospect via the editing operations. If these operations have an influence on the actual choice this constitutes a framing effect. Second, and most important, framing works via the setting of the reference point, i.e., via the distinction of outcomes as gains and losses. The evaluation of outcomes depends crucially on their perception as gain or a loss. In their famous Asian-disease problem28 Kahneman and Tversky demonstrate that the reference point can be altered by a different formulation and that such an alteration has considerable consequences for choice behavior. In a follow-up paper (1984) they also offer a third framing effect for multi-attribute choices. Kahneman and Tversky argue that 'in order to evaluate such a multi-attribute option, a person sets up a mental account that specifies the advantages associated with the option, relative to a multi-attribute reference state' (Kahneman and Tversky (1984: 346). The evaluation of such a prospect then depends on the balance of this mental account. Kahneman and Tversky (1984) introduce the following mental accounts: 1. A minimal account: only the differences between two options are included, shared features are of no relevance for the decision. 2) a topical account: the consequence of a choice is related to a reference level with respect to this choice, thus involving more dimensions. 3) A comprehensive account: the consequences of a choice are related to a reference level with respect to a wider range of choices. Framing effects here can be understood as a selection of relevant dimensions, which in turn influence the perceived costs and benefits of an option as well as the size of these costs and benefits. For example, options evaluated in a wider context as in the case of the comprehensive account appear much less important. Kahneman and Tversky (1984) suggest that decisions will generally be framed in terms of topical accounts. 27

A prospect (100, .1; 50, .5) stochastically dominates a prospect (100, .05; 50, .5) and will thus be preferred. 28 In the Asian-disease problem subjects have to choose between alternative programs to prevent a potentially deadly disease. Preference between alternatives depends on whether the consequences of the programs are described in terms of gains, ‘lives saved’ (which elicits risk-aversion’), or losses, ‘number of people dying’ (which provokes ‘risk-seeking’).

Intertemporal choices

23

In intertemporal choice, researchers have applied PT mainly with its distinction between gains and losses via reference points (Loewenstein 1988, Shelley 1993), resulting in different discount rates for such gains and losses, on the existence of different mental accounts which are related to different discount rates (Thaler & Shefrin 1992) and on reference point manipulation within the gain or loss domain making outcomes appear more or less immediate (Roelofsma 1994). 1.7.2 Framing in sociology

Contrary to behavioral decision theory framing, has a long tradition in sociology, but with a different notion. It is traditional sociological knowledge that actors define situations, i.e. they determine which aspects are relevant in a given situation. Depending on this definition they choose a proper behavior (see already Thomas & Znaniecki 1928; later this idea was developed further in the works of Mead and Schütz). Behavior is thus a product of the situation and the actor's definition of it. The term framing for the process of identifying and defining situations was introduced by Goffman (1974). For a long time the importance of a definition of the situation has been stressed mainly by proponents of interactionistic sociology (such as symbolic interactionists or ethnomethodologists) and has been neglected by theories that place little value on interactions, e.g., microeconomic consumer theory. Only in the 1980’s some attempts to combine the insights of traditional sociology and rational choice theory have been undertaken by Lindenberg in the belowsketched framing theory and by Esser (1990, 1991, 1996), both arguing that such an integration might contribute to our understanding of anomalous behavior.29 1.7.3 Framing as used in this book

The framing theory developed by Lindenberg, which will be used in this book, is based on the Discrimination Model, originally developed for stochastic repetitive choice (Lindenberg 1980) and successfully tested experimentally (Lindenberg 1981). Later the model was developed into a more elaborate framing theory (Lindenberg 1983, 1989) for which Braspenning (1992) provided some empirical support and which Ligthart (1995) successfully applied to bargaining. Recently, applications in the field of intertemporal choice have been suggested (Lindenberg 1993, 1995, 1996). The theory links two major findings of economics and sociology. From economics it borrows that goods are scarce and behavior is thus heavily influenced by the relative scarcities of these goods. More precisely, if goods become more scarce, prices rise and demand drops. From sociology it borrows that an actor's definition of a situation heavily 29

Esser even claims that the failure to look at the definition of the situation is responsible for almost all anomalies rational choice theory: 'Den Kern nahezu aller Anomalien der Theorie des rationalen Handelns bildet ein Vorgang, der von der Soziologie immer schon sehr betont worden ist: In jede Handlungsentscheidung wird eine besondere "Definition" der Situation geschaltet, die erst die Präferenzen und die Erwartungen strukturiert, von denen dann erst die Selektion eines Handelns ausgehen könne (Esser 1996: 2). Since the ‘definition of the situation’ structures preferences, it is by definition responsible for behavior irrespective of whether behavior is to be called ‘rational’ or ‘anomalous’. However, because this approach allows goal salience to change depending on what frame structures the situation, rationality is only possible within a definition of a situation, i.e. within a given frame. Hence, the distinction between rational and anomalous behavior is meaningless. Toda (1981) provides a similar argument from psychologists.

24

Chapter 1

influences his subsequent choice. In other words, the relevant goods in a situation might be very different for different types of actors and also the effect of a rise in the relative price on the demand of these goods might be very different depending on situation and actor (Lindenberg & Frey 1993). By doing so, it is possible to integrate the rich descriptiveness of classical sociology (especially the studies on the definition of the situation undertaken by symbolic interactionists) with the theoretical tradition of economics. By assumption, actors have limited cognitive capacity. Given that actors have goals and that action structures are complex, actors are forced to define a situation in real life decisions in terms of these goals. That is, in a given situation actors have dominant goals and restrict their attention to goal relevant aspects. The theory postulates that there is only one overarching goal that structures the situation and becomes the dominant goal, i.e., the frame, that defines the situation.30 For example, in a market situation it is typically assumed that gain-seeking or profit-maximizing is the dominant goal. This structuring is triggered automatically, i.e., the individual is unable to choose a frame. Contrary to the sociologist’s concept of the definition of the situation, the discrimination model assumes that goals potentially relevant in the situation but pushed into the background by the dominant goal still influence behavior, but only indirectly. They remain in the background of the decision maker's mind and affect the salience (strength) of the dominant goal. In a typical market situation, for example, the goal 'behave altruistically' will not be dominant but it might raise or lower the salience of 'gain'. This salience of the dominant goal determines the effect of the net utility of the alternatives (gi - g0) on the probability that the given alternative i will be chosen. Expressed algebraically: Pi = B (gi - g0) + 1/K, (4) where Pi is the choice probability of option i; gi is the expected utility of option i;31 g0 is the mean expected utility of all options. The parameter B denotes the salience of the respective frame, depending on other goals and K is the number of all options. Note, that formula (4) is about choice probabilities given frames. Hence, salience of frames in the discrimination model is exogenously given. As to frame selection, if no routine for handling certain decision situations has been developed, i.e., behavior is nonhabitual, the goal that discriminates best between the choice alternatives becomes dominant, structures the situation and pushes other situational goals in the background. If a routine for specific kinds of choice situations has already been developed, the dominant goal established previously prevails, until it fails to discriminate between the choice alternatives. Suppose, as an example, you bought an expensive ticket (say: f100) for a theatre show that received mixed reviews after you bought the ticket. While you are just leaving 30

'One overarching goal' is a somewhat misleading term, since in the theory 'one goal' may well comprise several attributes or goal variables. Suppose the goal is 'being a smart consumer'. Choosing between consumer goods being described by a variety of attributes the actor may well be able to compare goods by considering and weighing all the attributes. 31 As in the previous footnote, note that options may be described by several attributes. The expected utility of an option therefore weighs these attributes to determine the utility of an option. To arrive at an expected utility this weighted sum of the score of attributes is again weighed by the probability with which this options is to be expected.

Intertemporal choices

25

home to attend the show it begins to rain so heavily that you will get quite wet even when carrying an umbrella. The three most obvious goals in this situation are: a) to receive pleasure from attending the show, b) to avoid getting wet, and c) to avoid having thrown away f100 by not going to the show. Assume, you have no routine in handling such a decision situation. According to the model, alternative c should structure the situation. The enjoyment of the show is (due to the mixed reviews) unsure and thus not serving as a basis for a clear discrimination between going and not going. Because people in general are loss averse and because the subject already paid f 100, the utility difference between going and not going is likely to be larger in terms of losses than in terms of getting wet or not. However, the other goals do not vanish, but they remain in the back of the decision maker's mind and influence the salience of the structuring goal. In the example, the goal 'receiving pleasure from the show' raises the salience of the structuring goal 'loss-avoidance' and the goal 'avoid getting wet' lowers its salience. In this situation a change in the price for the theatre ticket (e.g., if your partner who bought the ticket tells you that the actual price was only f l0) has a much larger impact on the probability of 'going', than an increase in raining intensity. On the other hand, changes in raining intensity have a stronger impact if the dominant goal is ‘avoid getting wet’. Hence, changes in relative prices affect behavior quite differently depending on whether they occur on goods inside or outside the frame. To apply this theory to intertemporal choice, specifications are necessary as to what determines the salience of goals in an intertemporal context. Some goals promote behavior resulting in little attention to long-term consequences (say drug addicts search for means to overcome cravings) whereas other goals will make decision makers particularly sensitive to even small differences in long-term costs or benefits, e.g., when considering bank-offers to pay off your mortgage for the following 25 years. The remainder of this chapter lays out some mechanisms relevant in intertemporal choice. Note, that according to this model ‘rationality’ always is frame (or goal) dependent. Behavior is not necessarily rational across different frames. Behavior in one frame may appear ‘irrational’, if evaluated from another frame. 1.8. Application to intertemporal choice

To apply the theory laid out above, first the effect of goal salience on preference will be investigated, i.e., it will be tested for the effect of B on preference. Then, it will be investigated how changes in the choice set from which alternatives are chosen affect the discriminatory power of options and consequently preference. These tests are done in order to underline the general applicability of the discrimination model to intertemporal choice. Given that salience and discriminatory power are important in intertemporal choice, we will proceed by investigating other mechanisms for making goals salient, which lead to short-term or long-term orientation. For example, it will be argued that the salience of goals ignoring long-term consequences depends on characteristics of the good. More explicitly: hedonic goods, goods which provide direct utility, are more likely to lead to a frame in which ‘possession of the good’ is the dominant goal, than utilitarian goods are, goods serving as means to derive utility, even assuming that both goods have the same utility. Further, this

26

Chapter 1

likelihood depends on whether goods are presented vividly: staring at goods presented at a shop window affects preferences differently than choosing from a written list of items. It then will be argued that salience of frames depends on individual characteristics. The probability that in a specific situation a goal becomes dominant that ignores long-term benefits or costs, differs between individuals. Therefore, given identical resources and given an identical degree of deprivation, the propensity to frame this situation as ’end this miserable situation’ differs between people. Consequently, the probability of behaving impulsively differs between individuals, e.g., depending on whether how well they were socialized to delay gratification. Finally, the salience of rewards should decrease as a function of their distance from the decision maker. In intertemporal choice this is modeled by a discount function, may it be an exponential or a hyperbolic one. It will be argued that distance refers to several dimensions - outcomes might be risky (distant from certainty), temporally delayed (distant from ‘the present’), spatially distant (distant from ‘here’) or socially distant (distant from myself) – and that people discount outcomes according to their distance from a multi-dimensional reference point. While this is hardly a new statement – it is intuitively obvious that people prefer to receive benefits for themselves, ‘here’ and ‘now’ – it is also argued that these discounting processes follow similar discount functions and thus may yield almost identical preference shifts. Thus, preference reversals experimentally observed in intertemporal choice can be directly applied to interpersonal choice or risky choice due to a general influence of distance on salience. In the following sections, these considerations will be explained in somewhat more detail. 1.8.1 Framing theory, the discrimination model and salience in intertemporal choice

Lindenberg’s framing theory assumes that framing takes part habitually and automatically. If no decision routine for the respective decision situation has been established previously, the frame that discriminates best between choice options becomes dominant. The choice probability of the most favorable option within the dominant frame is enhanced or lowered by background goals depending on whether these goals are compatible with or run counter to the dominant frame. This model can incorporate hypotheses as to which people will be most impatient and when, as well as generate some new predictions. For example, the personal characteristics investigated in chapter 5 should influence the automatic triggering of the dominant frame. People with high selfcontrol thus more often ‘select’ frames which call for far-sighted behavior, i.e., have lower discount rates, than persons with low self-control. One could add an index capturing this to formula (4) above: Pij = Bj (gi - g0) + 1/K, j being an index for a specific personal characteristic, e.g., patience.

(5)

Because this dissertation is mainly concerned with intertemporal choices, only personal characteristics with respect to temporal discount rates are considered. However, personal characteristics should also influence the salience of frames in other domains. Lindenberg (1996) explicitly assumes framing in general to depend on institutions and norms. Adherence to norms as well as institutional effects differ between subjects - and if specifications of these influences analogous to those presented in Chapter 5 can be made

Intertemporal choices

27

- it may be reasonable to generally add such an individual index. For example, deviant behavior is more common and more pronounced for males than for females (Hirschi & Gottfredson 1994). This can be captured in formula 5 by assuming either that the expected utility of options available differs between sexes, that the salience of frames differs between sexes, or both. The hypotheses with respect to situation- and good-specific influences on discount rates likewise refer to the weight B in formula (4) and (5): for reasons elaborated below, hedonic goods and vivid presentations may lead to a larger salience of goals favoring short-term behavior, regardless of individual characteristics. However, it is possible to test the effect of discriminatory power on preference directly. For example, the attractiveness of the option ($100 in 4 weeks) according to the model should differ for the two choice sets C1: {($100 in 4 weeks), ($105 in 4 ½ weeks), ($120 in 6 weeks)} and C2 {($100 in 4 weeks), ($120 in 6 weeks)}. In the first choice set the option does not differ much from ($105 in 4 ½ weeks). Hence, a frame where ‘immediacy of receipt’ is a dominant goal does not discriminate between options and will therefore be replaced by a frame that discriminates better, while a frame where ‘amount of money’ is dominant does. The latter therefore is more likely to be dominant in this situation. 1.8.2. Good-specific determinants of patience: hedonic and utilitarian goods

It was already stated that the main shortcoming of the hyperbolic discounting perspectives is its failure to explain varying discount rates within an individual. Hyperbolic discounting heroically assumes that subjects reveal identical (hyperbolic) discount rates when buying designer clothes as in their gasoline purchases (see Chapman & Elstein 1995, and Dittmar, Beattie & Friese 1997 for counterevidence). Contrary to this position it will be argued that impatience is higher for hedonic than for utilitarian goods. Consumption of the former yields direct utility to a decision maker, while the latter serve as means to derive utility in future time periods.32 Hoch and Loewenstein (1991) argue that impulsivity is due to deprivation caused by the non-possession of certain goods. In this view, consumers adapt to the possession of a good and its actual non-possession then creates deprivation that is ended by buying the product. If hedonic goods provide direct utility, their non-possession elicits much higher levels of deprivation than the nonpossession of utilitarian goods. Recently, in fact Dhar and Wertenbroch (2000) demonstrated that loss aversion is more pronounced for hedonic than for utilitarian goods. Hedonic goods thus should create feelings of deprivation more easily and more strongly than utilitarian goods and subjects should therefore prefer to delay consumption of utilitarian rather than hedonic goods. In addition, impatience for hedonic over utilitarian goods should be stronger when subjects consider which good to receive sooner than when subjects have to determine adequate monetary compensation charges for delay. Asking subjects about their preference for immediate consumption raises the salience of hedonic frames, because immediate consumption is much more attractive when it provides direct utility. By contrast, asking for compensation charges raises the salience of utilitarian frames, because subjects have to consider alternative expenses, thereby distracting from the hedonic experience of immediate consumption.

32

Alinda van Bruggen (personal communication) suggested this definition.

28

Chapter 1

1.8.3. Situation-specific determinants of patience: vividness

In intertemporal choice the importance of vividness has been stressed frequently. A recurrent speculation is that vividness raises impatience. People walking along a shop front with nicely placed goods, maybe even smelling them, generally are much more eager to buy these goods than when sitting in their armchair spending time reading. It is thus often argued that forms of goods’ proximity raise discount rates (Loewenstein & Hoch 1991, Loewenstein 1996). This expected influence of vividness on discount rates is also supported by empirical evidence. Mischel and his co-workers conducted a series of experiments (summarized in Mischel et al. 1992) on children’s ability to delay gratification. They demonstrated that children are much less willing to wait for the reward when the reward was actually present. But this seems to be only part of the story. While standing in front of a bakery, viewing and smelling the products probably enhances one’s propensity to consume these goods. However, no such effects are reported, nor are they likely to be reported, for people at gas stations despite the fact that the smell of gasoline might provoke extremely vivid images. Intuition often quickly provides us with vivid examples of how and why vividness affects our judgments. Unfortunately, scientific scrutiny by and large has not corroborated this view. In a review on research about vividness Taylor and Fiske (1984) paint a rather dark picture and conclude that there is little evidence for vividness effects. It is argued here that vividness affects discount rates by affecting the deprivation felt by decision makers in two ways. First, vividness raises the perceived hedonic/utilitarian qualities of a good. That is, goods presented vividly are judged as more hedonic or utilitarian than non-vivid goods. Second, vividness raises adaptation to owning the good, thus creating stronger feelings of deprivation and more incentives to overcome this deprivation by buying the product immediately. Therefore, there is an interaction between type of good and vividness. Vividness is assumed to affect preferences for hedonic goods more strongly than for utilitarian goods, because the non-possession of hedonic goods inherently creates more deprivation. In addition, again there should be an effect of the elicitation method on impatience. As argued in the section above this effect should be dependent on the way impatience is measured. In choices which good to receive sooner, vividness should raise the attractiveness for hedonic goods more strongly than for utilitarian goods. When subjects determine financial compensation for the delay of these goods, vividness should affect the attractiveness of hedonic goods to a much lesser extent. 1.8.4. Individual determinants of patience

Economic theory usually regards preferences as tastes and thus as given. Due to the nonexistence of a reasonable theory of preference formation, tastes should be considered as 'similar among persons and stable over time' (Stigler & Becker 1977: 76). Only recently (e.g., Becker & Murphy 1988, Becker 1996) economists have begun to explore preference formation processes in more detail. As mentioned above preferences for the timing of rewards are an important part of a person's preference set. Fortunately, sociology has something to offer with respect to its formation: while tastes for chocolate ice cream instead of banana are probably distributed and acquired randomly among a population, delay of gratification is something that has to be taught. Consequently, children generally have more difficulties to realize that they cannot have all the pleasant

Intertemporal choices

29

things nicely displayed in supermarkets than do the standard (non-pathological) adult. To delay gratification is a major part of socialization, being taught mainly by parents but also by peers, institutions, such as schools or universities, and even professions. In this book the focus lies mainly on parental socialization for two reasons. First, with respect to later socialization factors it is not clear whether patience is a cause for further socialization or an effect of socialization. For example, a patient person, will be more likely to have higher educational achievements. He will be more likely to study instead of spending time on other things and more likely to value educational achievements, because they are seen as a valuable investment. This initial preference will again be reinforced by further attention of socializing institutions such as schools. Second, parental reinforcement and modeling influences children's preference formation more strongly than do other institutions (Bandura & Walters 1963). It is proposed here that parents differ in their ability to teach their children selfcontrol. In addition, incentives and restrictions for parents to teach their children selfcontrol differ. For example, parents with a high social status have more to lose when being blamed for misbehavior of their children than parents of low status (Coleman 1990, Cchapter 11). Hence, the former are expected to place greater value on a proper socialization. As to restrictions, parental time to teach children is reduced for single parents or when many children are present in the household. The latter reasoning is supported by results from research in areas where parental socialization is also relevant. For example, educational achievements generally are lower for children with many siblings due to limited parental resources, such as reading time or expenses (Downey 1995). 1.8.5. Outcome distance and discount rates

As stated above the relation between time delay and uncertainty has received considerable attention. It is argued here that empirical results obtained with respect to one preference dimension, e.g., risk, can in fact be applied to other preference dimensions as well, e.g., temporal delay. After a short comparison of the axioms underlying EU theory and DU theory, it will be demonstrated that the empirical regularities obtained by behavioral decision theorists apply to the outcome dimension (usually money) as well as to the distance dimension (time or risk). Then these regularities will be applied to other distance dimensions, viz. interpersonal and interspatial choice. For example, a person indifferent between $100 today and $150 in four weeks prefers the latter option when a time delay of four weeks is added to both options. Likewise, a person being indifferent between a sure win of $85 and a gamble offering $100 with a 90% chance prefers the later option, that is the higher amount, if a 30% chance of winning nothing is added to both options, i.e. the second choice is between ($85; .7) and ($100; .6). The addition of a four weeks delay is structurally equivalent to the addition of a 30% chance of winning nothing, because both alterations raise the distance of the considered options from the respective reference points (immediacy and certainty). Similar effects are expected to occur for other distance dimensions. Imagine the above-mentioned decision maker who is indifferent between $100 today and $150 in four weeks. Instead of choosing between temporally delayed outcomes adding an identical additional distance to both options should also change preference for interpersonal choices. Suppose, for example, the two amounts would be due to two persons being socially distant to varying degrees from the decision maker.

30

Chapter 1

1.9. Outline of the book

In the previous sections it has been stressed that intertemporal choices are relevant for real life decisions as well as rational choice theory. In addition, the DU-model, its accompanying anomalies, and competing models were explained in some detail. Further, a short overview of arguments used throughout this book was presented. Now, the rest of this book will be outlined briefly. Chapter 2 will test some of framing theory’s predictions concerning the effect of salience on preference and how salience depends on the choice set from which alternatives are chosen. In Chapter 3 hypotheses concerning good-specific influences on discount rates will be derived and tested experimentally. It is argued that hedonic goods are discounted more heavily than utilitarian or functional goods. In Chapter 4 it will then be investigated how a vivid presentation affects impatience for both types of goods. It will be argued that vividness raises impatience for hedonic goods more strongly when subjects have to state whether they would like to receive a hedonic or a utilitarian good sooner. For utilitarian goods vividness raises impatience more strongly when financial compensation charges have to be determined. Chapter 5 will specify how personal characteristics affect patience and how patience in turn has real consequences. It is suggested that parents differ in ability and incentives to teach their children to delay benefits. Experimental tests of these hypotheses are presented as well. Chapter 6 will compare similarities in the axiomatic systems underlying Expected Utility theory for choice under risk and Discounted Utility theory for intertemporal choices. Then the main empirical ‘anomalous’ regularities in intertemporal and risky choice will be compared and generalized to interpersonal and interspatial choice. It will be argued that decision makers compare outcomes to a multi-dimensional reference point and, due to diminishing sensitivity, discount deviations from this reference point in a similar fashion, viz. hyperbolically. Because anomalies in intertemporal choice and choice under risk are compared in this chapter there is some overlap between chapter 6 and section 1.4. in the current chapter. I am fully aware of the redundancies, but unfortunately presenting a full overview of anomalies in intertemporal choice together with a review of the main accompanying anomalies in choice under risk would mix up two different stories and therefore reduce accessibility of the text considerably. Chapter 7 will present empirical results of the generalizations of anomalies in intertemporal choice and choice under risk to interpersonal and interspatial choice. Chapter 8 will summarize and discuss the main ideas and results and present some suggestions for future research.

Chapter 2

Salience and the discrimination model in intertemporal choice

Summary

In this chapter it is first argued that the ‘anomalies’ in intertemporal choice described in chapter 1 can be interpreted as effects concerning the salience of decision attributes. Such changes in attribute salience and their effect on preference are compatible with the discrimination model underlying Lindenberg’s framing theory. The model postulates that preference and hence, choice probabilities depends on the salience of frames. This salience of frames in turn depends on their ability to discriminate between choice options. After demonstrating that attribute salience matters for determining preferences between intertemporal options, the postulates of the model are tested in several experiments specifying which factors influence discriminatory power. Because some of the tests require a manipulation of the choice context, i.e., the set of current and previous choice alternatives, previous research on context effects is briefly reviewed.

32

Chapter 2

2.1. Introduction

In Chapter 1 it was proposed that patience differs between individuals as well as for different types of goods and situations. Further, in Chapter 1 it was argued that such differences can be integrated in Lindenberg’s framing theory. According to this theory changes in the choice probability of an option depends on changes in the salience of frames. Hence, predictions concerning the salience of frames for different individuals and between different types of goods serve as bridge assumptions specifying which frame is likely to become dominant in a given situation. However, before choosing framing theory and the discrimination model as an overarching framework to answer the research questions raised in Chapter 1, their general applicability to intertemporal choice has to be demonstrated. Fortunately, it is possible to test some of the discrimination model’s postulates directly than only specifying appropriate bridge assumptions and this is the aim of this chapter. According to the discrimination model the probability of choosing an option depends on the salience of a frame and the utility of options within that frame. Frames structure the decision maker’s definition of the situation by determining which aspects are important in a given situation. For example, in choices between simple intertemporal options, offering different amounts of money at different points in time, the frame, will either focus on immediate returns (‘immediacy - get money now’) or on returns being as large as possible (‘outcome - get as much money as possible’). Salience of frames resembles an attribute's decision weight, a concept more common in decision theory. Because a frame in such simple choice situations consists of only one decision attribute,33 higher salience of a frame directly corresponds with a larger decision weight of this attribute. The following sections will briefly recapitulate some of the discrimination model’s postulates - beginning with frame selection and then turning to choice behavior given a frame - and then demonstrate how these postulates relate to intertemporal choice. The discrimination model postulates that in a routine decision the frame previously present in a similar situation prevails. For example, if in a choice situation between simple intertemporal options a decision maker is routinely concerned with maximizing profits, 'gain' will be the dominant goal independent of the offered interest rates. The decision maker therefore will always prefer the largest option, even if it is the most delayed. If no frame has been established previously, the frame that structures the situation best becomes dominant. That is, the frame will be dominant which allows the decision maker to discriminate most clearly between choice options. To illustrate the latter case, imagine you have to choose between two intertemporal options, offering different amounts of money at different points in time. It is a non-routine situation with two possible goals, ‘immediacy - get money now’ and ‘outcome - get as much money as possible’. Suppose the two options differ little with respect to ‘immediacy - get money now’. Then, it is likely that ‘outcome - get as much money as possible’ will become the dominant frame and the option preferable with respect to the outcome dimension will be chosen.34 Again, this can be interpreted as 33

This is not generally the case. Frames in this theory may well combine several attributes. Whether two options differ little or much also depends on the subject’s sensitivity to the differences, i.e., on the salience of the frame. A highly salient frame makes even small differences in utilities appear relatively large. This salience of the frame is also affected by background goals. These goals may or may not be compatible with the frame. If they are compatible the salience of the frame rises, otherwise it is

34

Salience and the discrimination model in intertemporal choice

33

changes in attributes decision weights due to the attributes' ability to discriminate between choice options. When a frame has been established, the choice option that maximizes utility within this frame is preferred and hence, chosen. However, this choice is stochastic depending on the strength of the frame and the utility of the other options under consideration. Suppose for example, a decision maker has to choose between two gambles and he is only concerned with maximizing the expected value of an option. That is, he is neither risk-averse nor risk-seeking. The probability of choosing the option with the highest expected value now depends also on the expected value of the other options. If expected values differ little between options, the probability of choosing the option with the highest expected value is much lower than when an option offers a clearly superior expected value. In this chapter, the influence of the salience of decision attributes on preference will be investigated since in the discrimination model the salience of frames is a crucial parameter to predict choices. To experimentally test for the effect of salience on preference predicted by the discrimination model three strategies are possible and will be carried out. First, one can directly ask for salience ratings of attributes, i.e., attributes' decision weights, and check whether preference corresponds to these ratings. Since in this approach ratings and preference are both elicited within persons, the problem arises whether subjects’ ratings reveal salience or the wish to be consistent (or a mixture of both). For example, they may argue ‘I preferred A over B, thus I have to rate the attribute on which A is better as more important’. If there is a homogeneous pool of subjects, one can also use a different strategy, viz. divide the pool and let one group of subjects rate the salience of attributes for a specific choice problem and another group state preference for this problem. As before salience ratings of attributes should predict preference. Finally, one can manipulate salience of attributes, e.g., by changing the discriminatory power of attributes, and check whether preference shifts accordingly. However, this final strategy does not elicit salience ratings explicitly. In the remainder of this chapter, first empirical results in the literature on intertemporal choice will be interpreted from a framing point of view and related to shifts in the salience of decision attributes. Then experimental evidence on the influence of attribute salience on preference will be presented. To do so, effects both within and between subjects will be investigated. Finally, attribute salience will be manipulated directly and we will test for accompanying changes in preference. These direct changes in salience require manipulations of the choice set from which alternatives are chosen. Hence, related research (mainly from consumer choice) where such manipulations of the choice context have received considerable attention will be reviewed briefly and these results will be linked to shifts in salience. Diagram 1 summarizes the line of reasoning put forth in this chapter:

lowered and may be more easily replaced by another frame. This interaction between background and foreground goals is not investigated more closely in this book.

34

Chapter 2

Diagram 1: Relation between choice set, discrimination, salience, and preference

Changes in Changes in the the choice set ⇒ discriminatory power of attributes (e.g., adding an option)

Changes in Changes in ⇒ salience ⇒ preference (choice probabilities)

2.2. A framing interpretation of ‘standard anomalies’ in intertemporal choice

Chapter 1 described several ‘anomalies’ to the DU-model. Some of them are dealt with in chapter 4, viz., differing discount rates for different objects, some of them have been satisfyingly modeled previously, viz. time-varying discount rates, sign effect, magnitude effect, and reference point effects.35 To see how these latter effects fit into framing theory, suppose that a decision maker is indifferent between two options yielding two outcomes x and y at two different future time points t and t': (x, t) ~ (y, t')

0 < x < y and 0 < t < t'

The major dimensions here are time and outcome and thus the choice could be framed in terms of immediacy or in terms of outcome. Time-varying discount rates, sign effect and magnitude effects now affect the salience of attributes and thus the choice probability36 (chapter 6 provides a detailed analysis of the effects of these manipulations on salience for intertemporal choice and choice under risk). For example, adding an additional temporal delay to both options due to diminishing sensitivity (Tversky & Kahneman 1992) lowers the salience of the temporal dimension and makes the outcome dimension more salient. Hence, the option that is more attractive with respect to this dimension, i.e., that offers a higher but later outcome, becomes preferred (see section 1.4.3 on time-varying discount rates for evidence). Adding such a common constant, makes both time delays more similar; the perceived temporal similarity between ‘364 days’ and ‘365 days’ is greater than between ‘now’ and ‘tomorrow’. Consequently, the temporal dimension fails to discriminate sufficiently between options, and thus the outcome dimension becomes decisive. This is in accordance with the discrimination model postulating that the salience of frames depends on their discriminatory power. Likewise, making both outcomes negative to the same amount raises the salience of the outcome dimension (Tversky & Kahneman 1991). The option that is more attractive with respect to this dimension, i.e., offers a lower but sooner loss, becomes preferred. Again, this is in accordance with the discrimination model postulating that people are loss averse and loss avoidance is likely to become the dominant frame in a loss-prone situation. Since such a greater sensitivity to losses has been demonstrated in a variety of settings including intertemporal choices, it will not be investigated in this book in more detail. The above interpretation is not unique to the discrimination model. Prelec and Loewenstein (1991) review some similarities between empirical regularities in risky and 35

Note again, that the framing effects reported in Chapter 1 are based on a different notion of framing, viz. changes in preferences due to reference point manipulations caused by rephrasing of the choice problem. Therefore, framing in this case refers to different matters. 36 Since the discrimination model is stochastic, changes in choice probabilities imply changes in preference.

Salience and the discrimination model in intertemporal choice

35

intertemporal choice and argue that exactly the mechanisms described above are responsible for the parallelism in ‘anomalies’ to the respective normative models.37 In a similar fashion Leland (1994, 1998) argued that similarity judgements cause ‘anomalies’ in risky and intertemporal choice. Such similarity judgements are invoked if, based on expected or present value, no clear preference between options (either risky or temporal) exists. In our interpretation, if choice is non-habitual and no frame initially is dominant, preference between choice options is established according to the discriminating power of potential frames (here: attributes present in the choice situation). It is not the aim of the present chapter or the book to separate these different existing theories from each other. Framing theory and the discrimination model are chosen as an overarching framework, because they appear to be flexible enough to incorporate the anomalies frequently observed in intertemporal choice as well as to provide heuristics to answer the general research questions posed in the previous chapter. It might be possible that other theories are also able to do so. 2.3. Effects of salience on preference

To recapitulate, the discrimination model postulates that choice probabilities depend on the salience of potential frames, which for simple intertemporal options are equivalent to weights of decision attributes. Suppose a frame favoring short-term aspects of a situation, e.g., hedonic consumption experience, is salient. Then the option offering immediate gratification is preferred. This may also be expressed by a larger decision weight of the temporal attribute compared to a decision situation where long-term aspects are more salient, e.g., functional considerations. In the latter case, the decision weight of the outcome attribute is larger, hence the option preferable with respect to this attribute becomes more favorable. The salience of decision attributes can easily be elicited by simply letting subjects rate the relative importance of either outcome or temporal attribute. Preference between simple intertemporal options then should correlate with these ratings: higher importance ratings for the temporal attribute should go along with a larger probability of preferring the more immediate option and vice versa. Hence: H1: Higher salience (or weight) of a decision attribute raises the choice probability of the option that is preferable with respect to this attribute.

This hypothesis is tested in the following experiment. Experiment 1 Subjects: Subjects were 89 students at the university of Groningen from various disciplines recruited via e-mail and by flyers. Mean age was 22 years, 44% were males and 56% females, average monthly income was ƒ1057 ($500), and the main fields of study were psychology (21%), management science (20%), natural science (11%), sociology (10%), educational science (11%), and economics/econometrics (6%). One subject had to be removed from all analyses, because he did not understand several of the decision tasks. 37

A more elaborate and detailed review and argumentation concerning similarities between and generalizability of empirical results in intertemporal and risky choice is provided in chapter 6 and 7.

36

Chapter 2

Design: Subjects were given a standard choice between two simple intertemporal options (x, t) and (y, t’). Subsequently, they had to state on a nine-point scale which aspect of the situation was most important for them. Ratings on this scale were used to predict their preference between simple intertemporal options. Higher salience ratings for 'immediacy' should raise the likelihood of preferring the sooner option while higher ratings for 'outcome' are expected to raise the likelihood of preferring the larger option. Subjects in group 1 were confronted with the following text (in Dutch, immediate condition):

'What do you prefer? (circle) A: ƒ100 now B: ƒ110 in 4 weeks Indicate on the following scale how important the two aspects of the situation (receiving the amount quickly and receiving a high amount) were to you. If you are only interested in receiving a high amount then choose the value 1. If you are only interested to receive the amount quickly, choose the value 9. Circle: Amount of money most important

Immediacy of receipt most important 1

2

3

4

5

6

7

8

9'

Group 2 was given a choice between ƒ100 in 10 weeks and ƒ110 in 14 weeks (remote condition). Subsequently subjects gave judgements on the same scale as group 1. We predict preference for the sooner option in both groups to increase with higher salience ratings for 'immediacy' and vice versa. Procedure: Subjects gathered in groups ranging from 2 to 23 persons to participate in the experiment. They were each provided with a booklet containing the experimental tasks and a questionnaire concerning the personal background of subjects. Experimental tasks comprised the choice problems described below, some other choice problems on intertemporal choices, presented partly in Chapters 3 and 4, as well as several unrelated decision scenarios. The questionnaire items are described in more detail in Chapter 5. A version of such a complete booklet is given in overall Appendix A at the end of this book. Since the decision scenarios were varied between subjects, only part of the sample received the choice problems presented below and n generally is smaller than 89. Subjects were told about the context in which the research took place and that there were no wrong or right answers. The experimenter was present during the whole experiment to answer questions concerning the questionnaire items or the questionnaire. Filling in the booklet took between 40 and 75 minutes, most people needed between 50 and 60 minutes. Subjects were paid ƒf10 (then approximately $4.00) and had an additional opportunity to win a gift certificate worth between ƒf15 and ƒ25 depending on their choices in another experiment (see Chapter 3).

Salience and the discrimination model in intertemporal choice

37

Experiment 1: results

In the immediate condition (n=50) preference was divided equally between options. Fifty percent preferred the sooner option, and 50% preferred the later but larger option. In the remote condition (n=39) only 42% preferred the sooner option, but 58% preferred the later and larger option. Adding a temporal delay of 10 weeks to both options, in accordance with hyperbolic discounting and framing theory and contrary to the DUmodel, reduces preferences for the sooner option by approximately 8% (from 50% to 42%). A logistic regression analysis was carried out with preference for the later and larger option as the dependent variable (coded as 0 = preference for the sooner option; 1 = preference for the later option) and salience of goals (importance ratings) and experimental condition (dummy variable indicating whether subjects faced the immediate or remote choice problem) as the independent variables. There is a clear effect of salience of the outcome dimension on the probability of choosing the later and larger reward (Wald: 18.62; p < .001). There is no separate effect of experimental condition on preference (Wald: 2.21; n.s.), indicating that salience is the main factor in explaining choices, irrespective of whether the choice is made for an immediate or a remote choice problem. Salience ratings themselves were not affected by experimental condition (t: .71; n.s.). That salience is the main determinant of preference is further underlined by the good model fit (χ2: 28.2; p < .0001).38 Hence, not too surprisingly, we find clear support for hypothesis 1. Because salience of frames or decision attributes is such an important theoretical ingredient of the discrimination model, we replicated experiment 1 with different options enlarging the choice set from two to three options. As has been demonstrated repeatedly (e.g., Herne 1996, see also below in this chapter) adding options and thus enlarging the choice set may change perceptions and evaluations of decision attributes and consequently preference between options. This may even hold in cases where the added options normatively should not influence preference. Because subjects and procedure of experiment 2 were identical to experiment 1, we turn directly to the experimental design. Experiment 2: design

Subjects were given a choice between three intertemporal options (x, t), (y, t’), and (z, t’’). Subsequently, they had to state on a nine-point scale which aspect of the situation was most important for them. Salience ratings on this scale were used to predict their preference between options. As in experiment 1, higher salience ratings for 'immediacy' should raise the likelihood of preferring the sooner option while higher ratings for 'outcome' are expected to raise the likelihood of preferring the larger option. The whole sample was divided into two groups to carry out another experiment (not reported here). However, both groups faced three simple intertemporal options. Subjects were confronted with the following text (in Dutch; preference proportions indicated):

38

A logistic regression with experimental condition as the only independent variable consequently yields a 2 much worse model (χ : .54, p = .46, pseudo R2 = .08). Adding experimental condition as an independent 2 variable to salience ratings in a logistic regression yields a residual χ of 2.13 (n.s.), i.e., model fit is not significantly improved.

38

Chapter 2

'What do you prefer? (circle) Options A: ƒ 100 in 2 days B: ƒ 105 in 3 days C: ƒ 120 in 14 days'

Group 1 (n = 50) 16% 24% 60%

Group 2 Options (n = 33) A’: ƒ 100 in 2 days 18% B’: ƒ 145 in 12 days 12% C’: ƒ 150 in 14 days 70%

Subsequently, subjects in both groups used the respective scale for salience ratings as in experiment 1 (see above). Experiment 2: results

Since in this experiment there are three options in the choice set, the dependent variable is no longer dichotomous and a linear regression is adequate. Preference for the option with the later and larger outcome is the dependent variable and salience of goals (importance ratings) and experimental condition (dummy variable indicating which choice set was presented to subjects) are the independent variables. As in experiment 1 with two options in the choice set, there is a clear effect of salience of the outcome dimension on the probability of choosing the later and larger reward (beta: .52; p < .001). Again, there is no separate effect of experimental condition on preference, indicating that salience is the main factor in explaining choices, irrespective of the specific choice condition. This is also underlined by the good model fit (Adjusted R2: .26, F: 15.35; p < .0001). Hence, again we find clear support for hypothesis 1. Table 2.1 summarizes results for experiment 1 and 2. Recall that in experiment 1 a logistic regression analysis was carried out, while in experiment 2 a linear regression analysis was carried out. Table 2.1 Regressions for preference between options depending on salience of frames, experiments 1 and 2

χ Model fit: Pseudo R2/Adjusted R2 Coefficients:

Constant Experimental condition Salience of ‘outcome’

2

Experiment 1 (n=88) B (S.E.) Wald Sig.

28.2 .37

Experiment 2 (n=83) F Beta T Sig.

.0000 15.35 .26 1.29 (87)

2.21

.1374

.77 (.53)

2.11

.1486

.68 (.16)

18.62

.0000

.0000

11.77

.000

-.042

-.44

.661

.52

-5.52

.000

Experiments 1 and 2 demonstrate that salience of decision attributes is an important predictor of preference in simple choice tasks. The next section investigates whether salience of frames depends on their ability to discriminate between choice options.

Salience and the discrimination model in intertemporal choice

39

2.4. Effects of discriminatory power on salience

The previous section demonstrated the relationship of attribute salience with individual preferences. Since these experiments elicited salience ratings and preference within subjects, results might be biased by subjects’ attempt to behave consistently. Subjects may argue that, if they preferred a smaller but sooner outcome to a larger and later outcome, they also should give more weight to ‘immediacy of receipt’. To further demonstrate the applicability of the discrimination model to intertemporal choices such a ‘consistency bias’ should be ruled out. In addition, it is also necessary to relate preferences to the discriminatory power of attributes more directly, since up to now only the influence of salience on preferences has been demonstrated. The better an attribute discriminates between choice options, the more likely it is that this attribute becomes decisive and that the option preferable with respect to this attribute is chosen. Note that discriminatory power of an attribute might differ from its salience. H2: the stronger an attribute discriminates, the higher its salience and the stronger the preference for the option preferable with respect to this attribute.

It was argued that standard manipulations in intertemporal choice affect the discriminatory power of attributes. For example, suppose a decision maker is indifferent between (x, t) and (y, t'): both attributes, temporal and outcome, discriminate equally between choice options. Adding a common temporal constant to both options usually leads to a preference for the later and larger reward reflecting hyperbolic discounting. In other words, this manipulation relatively lowers the discriminatory power of the temporal dimension and relatively raises the discriminatory power of the outcome dimension. Thus, if there are changes in preference between choice problems, such changes are expected to be caused by changes in the discriminatory power of attributes: H3: changes in the discriminatory power of attributes lead to changes in preference between choice problems.

Note, that hypothesis 2 refers to the effect of discriminatory power within choice problems, whereas hypothesis 3 refers to the effect of discriminatory power between choice problems. These are actually two different implications of discriminatory power. For instance, it might be possible that in a specific choice problem A one attribute discriminates clearly between options and that preference between options corresponds to the discriminatory power of attributes as specified in H2. Now suppose, this previously discriminating attribute discriminates less strongly between options in choice problem B than in A. Nevertheless, its discriminatory power may be stronger than the discriminatory power of the other attribute and be sufficiently strong for preference to be distributed correspondingly. In this case, H3 predicts preference between options to be distributed unevenly between options in both problems but more evenly in problem B. Both hypotheses are tested in the following experiment 3. Because subjects and procedure are identical to experiments 1 and 2, we turn directly to the design of experiment 3.

40

Chapter 2

Experiment 3: design

In experiment 1, subjects were divided into two groups and stated their preference for two intertemporal choice problems ('ƒ100 now vs. ƒ110 in 4 weeks' and 'ƒ100 in 10 weeks vs. ƒ110 in 14 weeks' - see above). Subsequently, subjects rated the salience of decision attributes. After completion of this experiment, subjects were asked to rate the difference in attractiveness of the temporal and the outcome attribute for the other choice problem (see the experimental booklet in Appendix A at the end of this book). That is, subjects in group 1 who stated preference and salience ratings for the problem 'ƒ100 now vs. ƒ110 in 4 weeks' subsequently rated the difference in attractiveness of attributes for the problem 'ƒ100 in 10 weeks vs. ƒ110 in 14 weeks'. Subjects in group 2 stated preference for the choice option for which group 1 rated differences in attractiveness for each attribute ('ƒ100 in 10 weeks vs. ƒ110 in 14 weeks') and then rated differences in attractiveness for the choice problem for which group 1 stated preference. Now these attractiveness ratings of one group are used to predict preference of the other group. It is argued that difference in attractiveness serves as a measure for the discriminatory power of attributes. The more strongly an attribute is perceived to differ in attractiveness between options, the better this attribute is able to discriminate between options. The option preferable with respect to this discriminating attribute thus should be more attractive (H2). Therefore, preference within choice problems for one group is expected to correlate with the other group's independent ratings of the discriminatory power of attributes. In addition, due to hyperbolic discounting, we expect preference for the later and larger reward to be stronger in the remote choice problem. This should be reflected in a relatively increased discriminatory power of the temporal attribute as opposed to the outcome attribute (H3). Note, that in experiment 3 the questions pertain to the discriminatory power of attributes whereas the scale in experiments 1 and 2 pertained to the salience of attributes. Note further, that in experiment 3 we have two groups, two choice problems and four attractiveness ratings (two attributes for two groups). With this design it is possible to use independent ratings (at a group level) of discriminatory power – and hence avoid possible consistency biases - as predictors for the preference of other subjects. This allows for a rather strong test of hypotheses 2 and 3. Subjects were divided into two groups and confronted with the following text: 'Imagine you have to choose between an amount of ƒ100 which you receive in 10 weeks and an amount of ƒ110 which you receive in 14 weeks. If you only look at the amount, how large do you estimate the difference in attractiveness between f100 and f110? To answer the question, use the following scale from 1 to 9 (1 means 'identical attractiveness, 9 means 'very large difference in attractiveness'). Circle: ‘f 100’ and ‘f 110’ equally attractive

1

‘f 110’ much more attractive than ‘f 100’ 2

3

4

5

6

7

8

9

Salience and the discrimination model in intertemporal choice

41

The same question was asked for the difference in attractiveness between 10 weeks and 14 weeks. Group 2 answered the same 2 questions but for a hypothetical choice between ƒ100 now and ƒ110 in 4 weeks. We expected the sooner option to be preferred when the difference in attractiveness for the temporal attribute by the other group was perceived to be larger than the difference in attractiveness for the outcome dimension and vice versa (H2). In addition, we expected changes in preference between the immediate and the remote choice problem to correspond to changes in attractiveness ratings (H3). Experiment 3: results

In both problems, the temporal dimension discriminates better than then the outcome dimension: For the ‘immediate’ problem with temporal delays ‘now’ and ‘4 weeks’ the mean value for the discriminatory power of the temporal dimension is 6.5, for the ‘remote’ problem with temporal delays ’10 weeks’ and ’14 weeks’ the mean value is 5.8. The outcome dimension discriminates less well: the mean value lies closer to the endpoint ‘equally attractive’ (4.5 for the ‘immediate’ and 5.0 for the ‘remote’ problem). Adding a 10 weeks delay to both options in the second choice problem does not lower the discriminatory power of the temporal attribute (6.5 vs. 5.8; t: 1.17; n.s.). Hypothesis 2 proposes stronger preference for an option for which an attribute discriminates more clearly. In the ‘immediate’ choice problem the temporal dimension discriminates much better than the outcome dimension (6.5 vs. 4.5 paired sample t-test: t: 3.84; p < .001) and therefore the sooner option should be preferred more often. Though actually 58% prefer the immediate option (preferable with respect to the temporal dimension), this proportion does not significantly differ from 50% (χ2: 1.26; n.s.).39 In the delayed choice neither the temporal nor the monetary dimension clearly discriminates between options (paired sample t-test: t: -1.24; n.s.) and hence, we expect preference to be distributed approximately evenly between options as is found in the data (50% preferring the larger option vs. 50% preferring the sooner option). Thus, hypothesis 2 receives only mixed support. With respect to hypothesis 3, stating that between choice problems changes in discriminatory power of attributes lead to changes in preference, the evidence is mixed as well. Since in the ‘immediate’ problem one attribute discriminates clearly and in the ‘remote’ problem both attributes discriminate approximately evenly, we expect that preference shifts towards an even distribution between options in the remote problem. Though there is a difference in preference between options (50/50 versus 58/42) this difference is not significant. Table 2.3 summarizes mean ratings for differences in discriminatory power (measured by differences in attractiveness of the respective attribute). Significance levels and t-values refer to differences between attributes. For example, in a choice problem where subjects choose between ƒ100 now vs. ƒ110 over 4 weeks, the mean ratings for difference in discriminatory power of the other group were 6.3 for the time dimension and 4.4 for the outcome dimension. The difference between these means is significant: the temporal dimension discriminates better.

39

Because in the immediate choice problem the temporal attribute discriminates more clearly than the outcome attribute, the sooner option should be preferred in this problem. Hence, preference between options should differ from a 50/50 random distribution which is tested by using a χ2-test.

42

Chapter 2

Table 2.3 Mean ratings of attributes’ discriminatory power in simple choice problems Means (Std. D. and t- values between brackets): Choice problem 100 now vs. 110 in 4 weeks 100 in 10 weeks vs. 110 in 14 weeks

Time (Std. D.)

Money (Std. D.)

Sig. (t-value)

Preference for the sooner reward

6.5 (2.4) 5.8 (2.6)

4.5 (2.5) 5.0 (2.7)

.000 (3.84) .225 (1.24)

58% 50%

Note that, changing the temporal context from 'now vs. 4 weeks' to '10 weeks vs. 14 weeks' slightly lowers the attractiveness ratings of the outcome dimension (4.5 in the immediate condition vs. 5.0 in the remote condition) despite the two amounts being actually identical in both choice problems. Subjects might evaluate the difference between outcomes against the background of the differences between time delays. Differences in temporal delays perceived as relatively large, as in the immediate problem, lead to an evaluation of the differences in outcomes as relatively small, an effect analogously observed in choices between standard consumer goods (Simonson & Tversky 1992). In terms of the discrimination model we argue that frames discriminate more strongly when favored by comparable other (past or previous) trade-offs between attributes. Thus, the attractiveness of options might change, when such comparisons take place. For example, consider a simple intertemporal choice problem where one frame 'immediacy' = the temporal attribute) discriminates strongly between options while 'gain' does not. Now, in another choice problem the salience of 'gain' should be raised, when two options are presented where 'gain' discriminates much better than in the previous problem while 'immediacy does not. The next experiments investigate more closely how subjects compare such tradeoffs between attributes when choosing between simple options. In experiment 3, comparisons took place within a choice problem. Experiment 4 below investigates tradeoffs between taking place between current and past choice options but with respect to identical attributes. Experiments 5 and 6 then manipulate the discriminatory power and salience of attributes by enlarging the choice set to three intertemporal options. Since similar manipulations have received considerable attention in consumer research, results from this area will be reviewed briefly in the following section. 2.5. Context effects in consumer choice

Standard economic theory, of which the DU model is part, assumes that preference between choice options is independent of irrelevant alternatives. If I prefer X over Y, this preference should be independent of other options in the choice set. At an aggregate level, this implies the so-called regularity principle. Regularity states that enlarging the choice set cannot increase the choice share of an option, i.e., the proportion of subjects preferring this option. If the choice set B is a subset of choice set A, the proportion P of subjects preferring alternative X the most cannot be larger for A than for B: P(X, A) ≤ P(X, B). This is a weaker statement than the one that choice should be independent of irrelevant alternatives. The latter refers to individual preferences while the former only concerns aggregates. Hence, regularity may hold, even if preference of all individual subjects does depend on irrelevant alternatives. Though regularity is a seemingly weak requirement, it has been violated repeatedly in studies on consumer choice and choices among gambles (see Herne 1996

Salience and the discrimination model in intertemporal choice

43

for a review). Already in the early eighties Huber, Payne and Puto (1982) demonstrated the asymmetric dominance effect. Here the market share of two products X and Y, each described by two attributes (a1, b1), (a2, b2) respectively, is altered by the introduction of a third product C1, the so-called decoy, in a specific way. This decoy C1 is dominated by X but not by Y. Imagine, for example, that the products, X, Y, and C1, are all described by two attributes: quality and price. Dominance of X over C1 then means that X either offers the same quality at a lower price, better quality at the same price, or is strictly superior on both dimensions. Adding C1, dominated by X on both dimensions, to the choice set X and Y, contrary to regularity, increases the popularity and market share of X relative to Y. Figure 2.1 shows possible placements of C1 as well as two other decoys whose impact on preference between X and Y is discussed below. Figure 2.1 Addition of decoys C1, C2, and C3 to the initial choice set of X and Y

C3 Attribute 1 (e.g., price: X offers a better price than Y)

X C1

C2 Y

Attribute 2 (e.g., quality) Later, this asymmetric dominance effect was extended by Huber and Puto (1983) by the addition of products that are relatively inferior to X. That is, the decoy C2 is much less attractive than X on dimension 1, e.g., much more expensive, but slightly superior on dimension 2, e.g., offering a slightly better quality. Though C2 is not dominated by X its addition again raises the attractiveness of X. Simonson (1989) further generalized this effect to cases where C is no longer inferior but, compared to X and Y, clearly superior to both on one dimension and clearly inferior on the other. For example, suppose that X is superior to Y in price but Y is superior to X in quality. C3 then offers a much better price than X but at even lower quality. Hence, the addition of C3 makes X a compromise in the continuum between Y and C3, again raising the attractiveness of X. Finally, Tversky and Simonson (1993) demonstrated the influence of the ‘background context’, i.e., previous choices, on preference. For example, they let subjects choose twice between brands of tires, differing in price and mileage. Initial choices between brands of tires differed between subjects. Group 1 chose between tires differing clearly in price but little in mileage, e.g., ($25, 30.000 miles) and ($49, 35.000 miles), group 2 chose between tires clearly differing in mileage but little in price, e.g., ($85, 55.000 miles) and ($91, 75.000 miles). Both groups then were given the same choice between tires differing considerably in price and mileage, e.g., ($60, 40.000 miles) and ($75, 50.000 miles). In this second choice, group 2 preferred the former option, superior in price, whereas group 1 preferred the second one, superior in mileage.

44

Chapter 2

Tversky and Simonson (1993) argued that between the two choices subjects compare tradeoffs between attributes (for an interpretation of these comparisons in terms of attribute salience see section 2.6 concerning context effects and similarity judgements). Tradeoffs between attributes specify 'exchange rates' between them. For example, in the problem concerning tires and mileage group 1 in the first choice faces a choice set where a large difference in price is accompanied by a small difference in mileage, thus mileage is relatively ‘expensive’ ($24 yield 5000 additional miles). The tradeoff between attributes in the second choice (smaller difference in price vs. a larger difference in mileage) then favors the option with the larger mileage ($15 yield 10.000 miles). Here, larger mileage is relatively ‘cheaper’ than in the initial choice. Conversely, for group 2 the tradeoff comparison of price and mileage favors the option with the lower price, because this group initially has made a choice where mileage was relatively ’cheaper’ (in the first choice $6 yield 20.000 miles, while in the second choice $15 yield 10.000 miles). Likewise, Tversky and Simonson (1993) attributed the asymmetric dominance effect to subject's comparison of tradeoffs with other options. A decoy raises the attractiveness of an option, if it offers a much better tradeoff between attributes than the decoy, while the other option cannot that easily be compared with the decoy. Hence the decoy raises the attractiveness of the option superior to it.40 Context effects have been found using a variety of consumer goods ranging from paper towels, over beer brands and micro-wave ovens to photocameras and car tires, but also on gambles (Wedell 1991). Recently some of these effects have been reinterpreted as rational inferences about product quality (Wernerfelt 1995; Prelec, Wernerfelt & Zettelmeyer 1998). Here, it is argued that consumers have rather good information about the relative location of their ideal products, e.g., whether they want a cheap product or one of superior quality, but rather bad information about the absolute location of their ideal product, e.g., what exact trade-off between price and quality is acceptable. Consumers therefore use context information to infer the absolute location of their ideal product, e.g., that X, since it dominates a decoy, offers reasonable quality at a good price. However, this account fails to explain why context effects also have been observed among gambles or gift coupons. Here, it makes no sense to assume that decision makers have no information about their absolute ideal option. The attractiveness of gift coupons or gambles normatively is clearly independent of the presence of irrelevant alternatives. People should know how much they are willing to pay for a gift and whether a gamble is attractive or not. For intertemporal options where observable interest rates exist which specify the appropriate trade-off between waiting time and increase of rewards the argument applies even more strongly: whether waiting for a larger amount is acceptable for subjects or not is not at all related to other options in the choice set, since subjects can compare offered trade-offs with interest rates. Other options thus provide no additional information. 40

Tversky and Simonson (1993) argue that next to trade-off contrasts another psychological mechanism is responsible for the occurrence of some context effects (e.g., the compromise effect): extremeness aversion. Extremeness aversion uses the general finding that people are loss averse (e.g., Kahneman & Tversky 1991) and extends it by arguing that people dislike disadvantages more than they like advantages. Compared to options with intermediate values options with extreme values such as the cheapest option or the option offering the highest quality offer advantages and disadvantages and are therefore less popular. Because, these effects will not be used for tests of the discrimination model in intertemporal choice, they have not been reviewed here in detail. However, extremeness aversion as an extension of loss aversion is compatible with framing theory as well, because loss aversion is a core assumption of framing theory.

Salience and the discrimination model in intertemporal choice

45

2.6. Linking context effects and similarity judgments

Tversky and Simonson (1993) attribute the context effects observed to two psychological features, viz. trade-off contrast and extremeness aversion. As argued above, we will concentrate on the relationship between the trade-off contrast and the discrimination model. The trade-off contrast can be accounted for by the argument that the similarity between attributes of different options is altered by the introduction of other options. Introducing an option which is dominated by one of the remaining options but not by the other alters the perceived similarity between the attribute positions of X and Y. For example, in Figure 2.2 below, option C1, e.g., ƒ90 in five weeks, is added to the original choice set consisting of X, ƒ100 in four weeks, and Y, ƒ20 in three days. This addition alters the perceived similarity between X and Y. C1 is relatively similar to X but not to Y. If the similarity of X compared to Y is judged against the similarity of X compared to C1, X and Y will be perceived as less similar in the triple than in the duo. More specifically, the perceived similarity between the two attributes of X and Y, ‘outcome’ and ‘temporal closeness’, will be affected differently by the introduction of C1. In the example below, with respect to outcome X is still the most favorable option, since it is better than C1 and Y. However, evaluating the distance on outcome between X and Y against the distance on outcome between X and C1 makes X and Y appear more different. The same holds for the distance between X and Y on temporal closeness. Preference now changes in favor of X, if the decrease in similarity for ‘outcome’ is stronger than the decrease in similarity for ‘temporal closeness’. In terms of the discrimination model the trade-off contrast and the accompanying changes in similarity affect the discriminatory power of frames. The frame (attribute) ‘outcome’, due to the changes in similarity explained above, after the addition of another option discriminates more strongly between options and hence, the option being preferable with respect to this frame becomes more attractive. Figure 2.2 presents several decoys introduced at different locations in the choice set. Consequences of these different locations are discussed below. Figure 2.2 Addition of decoys C1, C2, and C3 to the initial choice set of X and Y

Attribute 1 (e.g., outcome)

C2

X C1

C3

Y Attribute 2 (e.g., temporal closeness, Y is temporally closer than X) Exactly this relationship between perceived similarity of attribute positions and accompanying changes in preference after the introduction of decoys at different locations has been investigated by Dhar and Glazer (1996). They argued that in order for changes in preference between X and Y to occur, changes in the perceived similarity of X and Y are necessary. In a study on several products, e.g., automobiles, stereos, or

46

Chapter 2

apartments, each described by two attributes, Dhar and Glazer demonstrated that adding C1 raises the attractiveness of X only when the perceived similarity of the X and Y decreased. More specifically, the decoy had a stronger effect on the share of X, the stronger the negative effect of the decoy on the perceived similarity of the initial products X and Y. In a second study, Dhar and Glazer (1996) demonstrated that changes in perceived similarity also cause changes in preference when the decoy was not dominated by one of the initial options but also by none or both. Here, the decoy is located such that it alters the perceived distance on only one attribute in the core set. For example, adding C2 increases the range of the attribute ‘temporal closeness’ where option Y is superior. Hence, the relative distance between X and Y with respect to this attribute decreases and X’s disadvantage against Y with respect to this attribute decreases as well. Consequently, X becomes more preferable compared to the initial choice set. Likewise, adding C3, though not dominated, makes X and Y less similar on attribute 1 – outcome - and raises the share of X, as does C1 in the first example. Dhar and Glazer conclude that there might be decoys that do not alter perceived similarity of the options in the core set and hence do not alter the preference between them. In fact, in some previous studies (Huber, Payne & Puto 1982, Wedell 1991) shifting the location of the decoy had no effect on the occurrence of the asymmetric dominance effect. However, since none of these studies explicitly controlled for changes in perceived similarity of attributes, this non-occurrence might be due to the absence of changes in similarity of attributes. While the above reported effects of similarity judgements on preference are compatible with the discrimination model, the latter leads to an extra prediction concerning a limiting effect for choice sets where - after the introduction of a decoy - some options become so similar that it is difficult to differentiate between them. In such cases we expect a shift in attribute weights in favor of the attribute that after the introduction of a decoy discriminates better between options. Consequently, the option preferable with respect to this discriminating attribute becomes more attractive after the addition of options. This effect differs from the asymmetric dominance effect, since here none of the two similar options dominates the other. In the following, we will first look at the effect of comparisons between attribute trade-offs on attribute salience. Since subjects compare trade-offs between attributes with trade-offs offered by other attributes in previous choices, such comparisons affect the salience of goals. The previously salient goal offers a worse trade-off in the second choice. Due to loss aversion its salience is reduced and it is replaced by another frame. Options offering better trade-offs between attributes than offered in previous choices, become preferred more strongly, since the salience of the attribute or dimension where such favorable trade-offs are being made is raised. H4: Attribute salience - and preference for the option preferable with respect to this attribute - increases, if compared to worse previous trade-offs for this attribute.

To test this hypothesis the following experiment was carried out: Again, subjects and procedure were identical to experiments 1-3 and therefore we turn to the design immediately.

Salience and the discrimination model in intertemporal choice

47

Experiment 4: design

Subjects were divided into two groups and had to make two successive choices. The second choice problem was identical for both groups, while the initial choices differed. Group 1 initially chose between option A and B where the time dimension discriminated strongly between options while the outcome dimension did not. Group 2 initially chose between options A’ and B’ where the outcome dimension clearly differed but not the time dimension. Both groups then chose between options C and D where both dimensions differed. Since for group 1, in the second choice the temporal attribute offered a worse trade-off than in the first choice problem, the salience of the temporal attribute should decrease and the later and larger option should gain popularity. For group 2, the temporal attribute in the second choice offers a better trade-off, i.e., in the second choice time is 'cheaper', than in the first choice problem. Hence, its salience increases and the sooner option should be more popular in group 2 as compared to group 1. Subjects where confronted with the following choices (in Dutch; preference proportions indicated): ‘What do you prefer? (circle) Group 1 (n = 54) A: ƒ 100 in 7 days 67% B: ƒ 110 in 42 days 33% What do you prefer? (circle) Group 1 (n = 53)42 C: ƒ 100 in 10 days 17% D: ƒ 200 in 17 days 83%

(n = 34)41

A’: B’:

Group 2 ƒ100 in 7 days ƒ 200 in 10 days

C: D:

Group 2 ƒ 100 in 10 days ƒ 200 in 17 days’

(n = 34) 50% 50%

100%)

Preference for option D was expected to be stronger in group 1, since for this group in the initial choice the temporal attribute provided a better trade-off and time was 'cheaper'. Experiment 4: results

In group 1, 67% of the subjects preferred A in the first choice and 83% preferred D in the second choice. In group 2 none of the subjects preferred A’ in the first choice and preference for C and D in the second choice was divided equally. Hence, in the second group preference for the later option was significantly lower than in the first group (t: 3.26; p = .002). Apparently, for group 1 the dominant goal in the first choice, ‘immediacy’, provided such a worse trade-off in the second choice – 7 days sooner now cost ƒ100 – that, due to loss aversion, it had been replaced by ‘gain’ and the choice is made according to this frame. For group 2 again the initially dominant frame, ‘gain’, leads to worse trade-off in the second choice problem, becomes less salient and was replaced, resulting in a random choice. Thus, attribute salience can be raised by offering favorable trade-offs. 41

The two groups differ in size due to pragmatic reasons. Experiments reported in this chapter were carried out as a by-product of the experiments reported in chapters 3-5. Since it was difficult to collect subjects, data collection was stopped, despite of the difference in group size for some of the experiments presented here, when sufficient subjects participated to test the hypothesis presented in the following chapters. 42 One subject in group 1 did not answer the second question.

48

Chapter 2

If attribute salience depends on the attribute’s ability to discriminate between choice options, another context effects can be expected. Suppose, that there are three options in a choice set, e.g., (a consumer good, now), (ƒ135, in 20 days), and (ƒ140, in 22 days), each consisting of an outcome and a temporal attribute. The two attributes, outcome and temporal delay, now differ with respect to their ability to discriminate between the three options. While the temporal attribute discriminates and clearly favors one option, the outcome attribute does not. Not only are the two monetary amounts similar to each other, it is also difficult for the outcome attribute, if salient, to discriminate between the amounts and the consumer good. As a result, the temporal attribute should be more salient and the option preferable with respect to this attribute should be preferred more often. In the example above, the consumer good available immediately should be preferred more often than when compared to a choice set where one of the options with an amount is removed: H5: in choice set X, consisting of A, B, and C (all defined by 2 attributes), preference for A is stronger than in choice set Y including A and B, if the attribute where A is superior in X discriminates better between all options.

This hypothesis was tested in experiment 5. Again, subjects and procedure were identical to experiments 1-4 and hence, we immediately turn towards the design of experiment 5. Experiment 5: design

Subjects were randomly divided into two groups. Group 2 chose from a choice set including options A, B, and C. B and C were relatively similar monetary amounts available after time delays of one and two weeks respectively. The sooner available and distinct option A was a standard consumer good, viz. ‘an outstanding CD-player', of roughly equal value well known to the subjects. Group 1 chose from a set including only A and B, with the 'similar' option C being omitted. We expected option A, due to a larger salience of the temporal attribute in the triple, to be preferred more strongly in group 2. Subjects were confronted with the following text (in Dutch; preference proportions indicated):

‘What do you prefer? (circle) A:

An outstanding CD-player in two days

B:

ƒ 500 in one week

C:

ƒ 520 in two weeks

Group 1 (n = 34) 3% 97%

Group 2 (n = 21) 24% 62% 14%

Experiment 5: results

As predicted the distinct option A is preferred more often in the triple than in the duo (24% vs. 3%). However, this effect is only marginally significant (one-tailed test: t: 1.36,

Salience and the discrimination model in intertemporal choice

49

p < .1).43 This unexpectedly weak effect may be due to an overly strict testing of the hypothesis. Note, that to compare preference changes between duo and triple only preference for the distinct option was considered. Preference for the other two options in the triple was collapsed in order to analyze the data. Thus, it is assumed that all subjects (14%) who in the triple preferred C not present in the duo would have preferred B similar to it in the duo. Hence, in the analysis above, the effect of adding a similar option might have been underestimated. A final experiment was carried out to investigate the effect of adding options to the choice set that are ‘easily comparable’ with and inferior to one of the remaining options but not with the other. Adding such an option should make the trade-off between attributes in one option appear more favorable while leaving the trade-off between attributes in the remaining option unchanged. Consider the following example, taken from Simonson and Tversky (1992), concerning a choice between microwave ovens varying in price, quality and discount (note that these options are three-dimensional). Here, an initial choice set consisting of a cheap standard and an expensive high-quality microwave was enlarged by adding a(nother) high quality microwave. However, this additional expensive microwave offered a lower discount than its roughly equally expensive initial counterpart. Consequently, the attractiveness of the initial high-quality product was raised compared to a choice between two microwaves. H6: Adding an option which is easily comparable with and relatively inferior to one of the other options in a duo raises the attractiveness of the superior option.

This hypothesis is tested in experiment 6. Again subjects and procedure are identical to experiments 1-5, so we turn directly to the design of experiment 6. Experiment 6: design

Subjects were randomly divided in two groups. Group 1 chose from a choice set including two options (a consumer good and a monetary amount), both available after some time delay. Group 2 chose from a choice set including three options (two consumer goods among which one is clearly superior to the other and a monetary amount) all available at different points in time. In group 2, preference for the superior consumer good was expected to be stronger than in group 1. Subjects were confronted with the following text (in Dutch; preference proportions indicated): What do you prefer? (circle): A:

An outstanding CD-player in two days

B:

ƒ 500 in one week

C:

A standard cassette recorder tomorrow

43

Group 1 (n = 34) 3% 97%

Group 2 (n = 33) 15% 85% 0%

Throughout this book, significance levels are generally based on two-tailed test (as in the previous experiments). For some experiments, only one-tailed tests of significance have been carried out since the interest is on the extent to which predicted changes in preference are significant. However, the reader should bear in mind that such tests are less strict than the standardly used two-tailed tests.

50

Chapter 2

Note, that in the text above none of the options strictly dominates the other: while subjects should prefer an ‘outstanding CD-player’ to a ‘standard cassette recorder’ the latter is available immediately. Note further, that the CD-player and cassette recorder are distinct enough to ensure that, contrary to experiment 5, all options differ from each other sufficiently. Experiment 6: results

As predicted, in group 2 the CD-player is preferred more often (15.6% vs. 2.9%; t: 1.75; p < .05; one-tailed-test). The addition of a standard cassette recorder, while being not at all attractive in itself (zero preference), raised the attractiveness of the CD-player. 2.7. Overall conclusion and discussion

This chapter aimed at demonstrating the general applicability of the discrimination model in intertemporal choice. To do so we tried to relate preference to goal/attribute salience and to the discriminatory power of goals/attributes. Experiments 1 and 2 demonstrated the importance of goal/attribute salience in intertemporal choice. Salience ratings of attributes served as good predictors for preference in simple intertemporal choice problems. Further, it was argued that preference should depend on the discriminatory power of goals/attributes. In fact, in experiment 3 for a choice problem where both goals discriminated equally well between choice options preference was distributed evenly. However, for a choice problem where the temporal dimension discriminated better and one therefore would expect the sooner option to be preferred more often preference for the sooner option (58%) did not statistically differ from an even distribution. Thus, there is some evidence for discriminatory power of goals/attributes to affect preference. In addition, it was argued that preference changes between choice problems, such as e.g., the common difference effect, are due to changes in the discriminatory power of goals. That is, if in problem A the sooner option is preferred more often than problem B, then the temporal dimension must discriminate more strongly than the outcome dimension in problem A than in problem B. Therefore, to test this hypothesis it is necessary that changes in preference occur between choice problems. Yet, in experiment 3 the addition of a temporal delay of ten weeks to both choice options did not result in such changes. Because results from the other experiments were by and large consistent with the predictions, the absence of a relation between discriminatory power and preference changes between choice problems is probably due to a failure of the experimental manipulation instead of a wrong theory. Besides these more direct tests of the model, hypotheses concerning manipulations of the choice set from which options are chosen were tested. Such manipulations were chosen to indirectly change the discriminatory power of goals, and consequently, preference. Experiment 4 demonstrated that salience of goals/attributes is sensitive to subjects’ comparisons of current alternatives to previous choice options. The attractiveness of intertemporal options could be raised by previously letting subjects answer similar choice problems where the temporal or outcome attribute discriminated clearly between options. In the second choice problem subjects compared the attractiveness of the current options with the options for which they previously made a choice. In case an attribute, e.g., time delay, appeared to offer an improvement in the

Salience and the discrimination model in intertemporal choice

51

second choice, e.g., because subjects have to wait less long for an amount than in the first choice problem, its salience increases and preference changes accordingly. Experiment 5 introduced a limiting case of the asymmetric dominance effect. This effect states that preference for a target option increases after the addition of a similar but dominated decoy. However, in experiment 5 the addition of a similar but difficult to distinguish option marginally decreased the popularity of a target. The goal/attribute for which the two similar options are superior to the remaining option does not sufficiently discriminate between the two similar options. Therefore, the goal/attribute with respect to which the two similar options are inferior to the remaining option structures the choice situation and the remaining option becomes preferred marginally more often. Finally, experiment 6 demonstrated that the salience of a goal/attribute can be raised by the addition of an option where one attribute is similar to only one of the other alternatives. In addition, the added option has to be relatively inferior to this similar alternative. The attribute that appears in two of the three options in this experiment is able to discriminate between options and therefore this goal becomes salient. These results all run counter to standard economic theory, but can be predicted from the discrimination model. Hence, this model in fact appears to have predictive value for intertemporal choices and the question what determines salience in intertemporal choices appears to be useful. Given that some of the effects demonstrated in this chapter have analogies in choices between consumer goods and given that the discrimination model is a general model for frame selection, the question arises whether effects observed are restricted to intertemporal choices or are generalizable to other domains as well. Chapter 6 argues that 'anomalies' in intertemporal choice have analogies in choice under risk, not only with respect to effects, but also with respect to underlying cognitive mechanisms. Chapter 7 investigates similar cognitive processes in seemingly unrelated choice situations, viz. interpersonal and interspatial choices. In a similar fashion Medin, Goldstone and Markman (1995) review similarities in decision-making and similarity judgements. Likewise, they argue that behavioral regularities in these domains are highly similar and, even more important, due to similar underlying cognitive mechanisms. Further research seems warranted to investigate other parallels between seemingly unrelated areas of decision research.

Chapter 3 Intertemporal choices for hedonic and utilitarian goods

Summary*

Future outcomes are generally valued less than immediate ones. The normative DUmodel as well as the standard descriptive hyperbolic discounting model imply that this devaluation depends on the present utility of the reward, delay time and a delay gradient, a personal measure of impatience. In this chapter it is argued that only certain types of consumption, viz. hedonic consumption instead of utilitarian consumption, are associated with impulsive behavior. People, we assume, behave myopically when buying fancy clothes or alcohol, but only rarely when purchasing gasoline or washing machines. We present three experiments that show that people are more impatient to receive hedonic goods than utilitarian goods, even in cases where utilitarian goods are valued more highly. According to the framing theory introduced in Chapter 2 this effect is stronger in choices about which good to receive sooner than when adequate compensation charges have to be determined. In fact, we do not observe differences between hedonic and utilitarian goods for such compensation charges. *

This chapter benefited from discussions with and comments from Laurie Hendrickx, George Loewenstein, and Drazen Prelec. Marcel van Assen, Alinda van Bruggen and Renee van der Hulst helped to formulate the Dutch scenarios. None of them is responsible for any remaining errors. An earlier version of this chapter has been presented at the ICS Forum Day in 1998.

54

Chapter 3

3.1. Introduction

People generally have a preference to receive positive rewards as soon as possible. Delay of an outcome leads to a devaluation of the outcome's absolute value, i.e., it is discounted. As demonstrated by empirical research (e.g., Loewenstein & Elster 1992) this discounting of future outcomes is most appropriately modeled by a hyperbolic discount function: U (V 0) U (Vt ) = , (1) (1 + αt ) (V0) being the immediate value of an outcome, U(Vt) its value at time delay t, α being a constant determining the steepness of the discount function44, and t the number of time periods the reward is being delayed. This model implies that time delays close to immediacy lead to stronger discounting of an object's absolute value than identical time delays occurring later. People thus are relatively eager to receive rewards now instead of one day later but they care little whether they receive the same reward in 65 or in 66 days. The hyperbolic function describes preference for immediate rewards as well as impulsive or timeinconsistent behavior: people having the choice between a small reward available after some time and a later but larger reward initially may prefer the later reward. Their preference may, however, reverse, if both rewards become less delayed and the smaller reward is available immediately. In the hyperbolic discounting model discounting depends on the present utility of the reward, the delay time and discount factor α, which is usually interpreted as a personal measure of impatience. Hence, discount rates should be identical between objects of identical present utility. Hoch and Loewenstein (1991) and Loewenstein (1996) have pointed out two shortcomings of the hyperbolic discounting perspective. First, generally only certain types of consumption are associated with impulsivity. Common sense readily acknowledges impulsivity when drugs, alcohol or sweets are to be consumed, but only rarely are purchases of gasoline or kitchen facilities considered to be impulsive (for empirical evidence on good-specific discount rates see Chapman and Elstein 1995 and Dittmar, Beattie & Friese 1996; for evidence concerning impulse buying see Bellenger 1978). Casual inspection also suggests that people are quite sensitive to even small longterm consequences when considering which bank offer to accept in order to pay back their mortgage for the following twenty years. Second, hyperbolic discounting can also not explain why certain situational variables, e.g., direct contact with the object, sometimes raise impatience. If people discount the utility of delayed rewards hyperbolically, they would apply one and the same discount function regardless of situational or good-specific characteristics.45 If different goods would be discounted at different rates, this could result in preference reversals. That is, in a choice between two 44

In addition, α indicates whether a subject prefers immediate to future outcomes, if α > 0, future outcomes over immediate ones, α < 0, or is indifferent between receiving present or future outcomes, α = 0. See chapter 1 for an extensive discussion of hyperbolic discounting models. 45 Of course, in addition the utility the consumption of a good provides may be time-dependent. That is, my evaluation of a bottle of water may well depend on whether I am thirsty or not. But this is not a situationspecific influence of my discounting of the utility a bottle of water provides. Such a situation-specific influence would be, e.g., a vivid description of this water bottle (see chapter 4) leaving its utility from future consumption unaffected.

Intertemporal choices for hedonic and utilitarian goods

55

goods available immediately good A is preferred, while B might be preferred when both goods are being delayed the same amount of time. This chapter extends the standard hyperbolic discounting perspective by proposing that discount rates vary between ‘hedonic’ and ‘utilitarian’ goods. Hedonic goods are goods associated with fun and pleasure whose consumption provides direct utility. Utilitarian goods are instrumental goods that serve as means to reach higher utility levels. Because delaying goods means an immediate loss (one has to forego immediate use of the product) and due to the greater ease with which hedonic usage of goods can be imagined, we expect loss aversion and consequently impatience to be greater for hedonic goods. 3.2. Theory

Hoch and Loewenstein (1991) suggested that impulsivity is caused by shifts in the consumer’s reference point. When acting impulsively, consumers adapt to the reference state of owning a good. The actual non-possession of this good then creates deprivation, which is enhanced if the good is proximate, temporally or physically. Impulse buying consequently is more likely for proximate goods, because consumers want to end the experienced deprivation and buy the product. Likewise, Loewenstein (1988) and Shelley (1993) argue that delay charges, the standard measure for discount rates, reflect the decision maker’s adaptation to owning an object: telling a person that he will receive a certain object encourages him to adopt the reference state of owning this object. Asking to delay the receipt of a benefit to which he has partly or fully adjusted yields an immediate loss. This loss is compensated by the future discounted gain plus the delay charge.46 Hoch and Loewenstein model intertemporal choice as follows: D = a(P - r) + b(r-0),

(2)

a being the slope of the consumer's value function in the positive region, b its slope in the negative region,47 and r is the level of adaptation between purchase (P) and non-purchase (0). For example, consumers being able to imagine the possession of a good very well, adapt more strongly to its possession and are more likely to buy the product. Desire D represents the consumer’s motivation to purchase a product. Recently, Dhar and Wertenbroch (2000) investigated preferences for hedonic and utilitarian goods. They define hedonic goods as being characterized by an affective experience of sensual or aesthetic pleasure, fantasy, and fun. Utilitarian goods are those whose consumption is more cognitively driven, reasoned, and goal-oriented. They are thus more of an instrumental and functional nature. Although this distinction might seem somewhat vague, consumer goods have been successfully classified along these dimensions (Mano & Oliver 1993). In this book hedonic and utilitarian goods will be distinguished according to whether consumption of these goods provides direct utility as is the case for hedonic goods, or whether their consumption serves as means to derive

46

Note, that Hoch and Loewenstein’s argument refers to deprivation while the latter two deal with delay charges, i.e. monetary discount rates. 47 Hoch and Loewenstein assume linear utility functions for gains and losses. However, one could easily generalize the model to incorporate also S-shaped utility functions.

56

Chapter 3

utility from other goods as is the case for utilitarian goods.48 To give an example for the latter, for most people the use of a vacuum cleaner provides little utility in itself but instead is used to derive utility from a clean room. Dhar and Wertenbroch (2000) demonstrated that loss aversion is stronger for hedonic than for utilitarian goods. That is, people are less willing to forego hedonic goods instead of utilitarian goods of equal value. Stronger loss aversion for hedonic than for utilitarian goods implies that the slope of the value function in the negative region differs for hedonic and utilitarian goods: bh > bu. Hence, decision makers should feel more deprived when having to delay hedonic instead of utilitarian goods: Dh = a(P - r) + bh(r-0) > Du = a(P - r) + bu(r-0),

(2)

Because deprivation is stronger for hedonic than for utilitarian goods we hypothesize: H1: People prefer to receive hedonic goods sooner than utilitarian goods. 3.3. Method

Experiments used scenarios where subjects were told that they had won some reward in a lottery and had to delay receipt of this reward. Such lottery scenarios, introduced by Thaler (1981), were used in several previous experiments as well as in surveys and evolved into some standard paradigm for measuring discount rates. However, we did not investigate monetary discount rates as is usually done within this paradigm, but instead asked subjects which of two goods they would like to receive sooner when given the possibility. Subjects Subjects were 80 students at the university of Groningen from various disciplines recruited via e-mail and by flyers. They were confronted with decision scenarios presented on a computer screen and partly with decision scenarios and questions concerning their personal background presented on paper. Since the decision scenarios were varied between subjects, only part of the sample received the choice problems presented below and the number of subjects varies between problems. Mean age was 22 years, 42% were males and 58% females, average monthly income was ƒ1100 ($500), and the main fields of study were psychology (30%), management science (11%), languages (8.5%), sociology (8.5%), law (7%), economics/econometrics (8.5%) and natural sciences (7%). Method Subjects gathered in groups ranging from 6 to 15. They were assigned to a computer where the program concerning the experimental tasks described in chapter 7 had been installed and instructed about the context in which the research took place and that there were no wrong or right answers. In between these experimental tasks were regular breaks where subjects turned to the decision scenarios presented in this chapter as well as to the questionnaire. Decision scenarios varied between subjects to prevent order effects, hence not all subjects were confronted with all scenarios. The experimenter was present during the whole experiment in order to answer questions concerning the tasks.

48

This distinction was suggested to me by Alinda van Bruggen.

Intertemporal choices for hedonic and utilitarian goods

57

Completing the whole experiment took between 50 and 110 minutes, most people needed between 80 and 90 minutes. Subjects were paid ƒ25 for their participation. 3.4. Experiments and results Experiment 1: design

In the first experiment (n = 41) we tested whether there is indeed a stronger propensity to want hedonic goods sooner than utilitarian goods. We investigated preference for gift certificates for CD’s and computer diskettes. Both goods are comparable insofar as they both provide streams of utility and we expected subjects to be familiar with them. Gift certificates for CD's were considered hedonic, for computer diskettes they were considered to be utilitarian. After completion of the experiments part of the sample (n = 39) was asked to classify the consumer goods used in all three experiments as either type A: ‘hedonic’, type B: ‘utilitarian’, type C: ‘both’ or type D: ‘neither’.49 Subjects were told they had won two gift certificates in a lottery (each worth 20 guilders - approximately $9) and then asked which of the two certificates they would like to receive sooner. Forty-one subjects answered the following questions (in Dutch): Imagine: you participated in a lottery. You could not only win monetary prizes, but also consumer goods. You were so fortunate to win a gift certificate for CD’s and a gift certificate for computer diskettes. Both gift certificates are worth 20 guilders. What is the minimal amount for which you would sell the gift certificate for CD’s immediately? ƒ________ What is the minimal amount for which you would sell the gift certificate for computer diskettes immediately? ƒ________ Unfortunately there is a temporary shortage in prizes. Thus, not every winner can receive his or her prizes immediately. The organization of the lottery therefore asks you to wait for your prizes. You have two possibilities (circle what you prefer): A:

You get the CD certificate immediately and the certificate for computer disks in four weeks.

B:

You get the certificate for computer diskettes immediately and the certificate for CD’s in four weeks.

Order of presentation was varied. Half of the subjects stated selling prices for the computer diskettes certificate first and made their choice from an option set where Option A offered this certificate sooner. Eliciting selling prices is necessary to control for the trivial fact that subjects should wish to receive a higher valued object sooner. 49

The exact wording was: ‘Some people distinguish between goods. Some goods are pleasant and fun, give pleasure and appeal to your senses, e.g., perfume (type A). Some goods are useful, practical, functional, something that helps you to achieve a goal, e.g., a vacuum cleaner (type B). Could you indicate for each of the following goods whether they are type A, type B, both, or none. Circle.’

58

Chapter 3

Experiment 1: results

CD's were classified as primarily hedonic by 87%, as both hedonic and utilitarian by 7.7%, and as primarily utilitarian by 5.3% (χ2 = 35.89; p < .001). Computer diskettes were classified as primarily hedonic by 7.7%, as both hedonic and utilitarian by 7.7%, as primarily utilitarian by 82.6% and as none by 2% (χ2 = 31.54; p < .0001). Hence, the manipulation hedonic versus utilitarian appeared to have worked. Subjects on average valued a gift certificate for CD’s higher than certificates for computer diskettes (mean selling prices ƒ20 vs. ƒ18, t: 4.35; p < .001). Of the total sample 98% chose to receive the gift certificate for the CD sooner (χ2: 37.1; p < .0001). Note that for people valuing certificates for CD’s more than certificates for computer diskettes the choice between the two certificates resembles a choice whether to receive a smaller or a larger reward sooner. Since in such a choice situation the larger reward should be preferred sooner, it is more sensible to restrict the analysis to the part of the sample which values both goods equally. As it turned out for these subjects there is a clear preference to receive the certificate for CD’s sooner: twenty-four out of twenty-five subjects chose the CD certificate sooner (χ2: 21.6; p < .0001). Hence, we have a clear preference to receive hedonic goods sooner.50 Experiment 2: design

Since in experiment 1 the hedonic good on average was valued higher than the utilitarian good the result may partly be due to some unmeasured differences in value. For example, subjects may think that, despite greater preference for gift certificates for CD’s, they should be compensated with the nominal value of the certificate. Hence, we replicated experiment 1 with two other durable consumer goods (n = 66). This time we chose a sandwich toaster worth ƒ40 and a cassette with cutlery also worth ƒ 40. It was hypothesized that the sandwich toaster would be considered hedonic and the cassette with cutlery would be considered utilitarian. Subjects were given the same text as above but with other goods. Again, order of presentation was varied. Half of the sample received a question where the hedonic good was presented first, the other half answered a question where the utilitarian good was presented first. Experiment 2: results

The sandwich toaster was rated by 56% as ‘primarily hedonic’, by 23% as ‘both’ and by 23% as ‘primarily utilitarian’ (t: 2.49; p < .05). The cassette with cutlery was rated as ‘primarily utilitarian’ by 66.5%, by 20.5% as ‘primarily hedonic’, by 5% as ‘both’ and by 7.7% as ‘none’ (t: 4.24; p < .001). Consequently, in line with expectation, the sandwich toaster was rated as more hedonic than cassette met cutlery (t: 6.27; p < .001). 50

A pretest was done with 51 subjects who voluntarily answered an identical question. The answers were almost identical: gift certificates for CD’s were valued slightly more than for computer diskettes (mean selling prices ƒ18 vs. ƒ15; t: 3.18; p < .001). The number of subjects who preferred the certificate for computer diskettes sooner was slightly higher than in the experiment reported above (5 out of 51). Subjects valuing one certificate more than the other always preferred to receive this certificate sooner. Restricting the analysis to subjects who value both certificates equally again reveals strong preference (20 out of 21 subjects) to receive the hedonic good sooner (χ2: 17.19; p < .001).

Intertemporal choices for hedonic and utilitarian goods

59

Mean selling prices (sandwich toaster: ƒ35, cassette met cutlery: ƒ34) are equal (t: .61; n.s.). In a direct choice which of these two goods to receive immediately, 66% of the subjects preferred to receive the hedonic good sooner (χ2: 6.06; p < .05). Restricting the sample to the number of subjects who stated identical selling prices for both goods (n = 30) led to an identical distribution (2/3 chose to receive the sandwich toaster sooner) but reduced significance (one-tailed test, χ2: 3.33; p < .05). Experiment 3: design

Experiment 1 and 2 demonstrated that there is a greater tendency to prefer hedonic goods sooner than utilitarian goods. We conducted a final experiment (n = 76) with respect to whether there is a greater propensity to want hedonic goods sooner than utilitarian goods. This time we selected goods such that the utilitarian good was valued slightly higher. This was done in order to test whether the greater immediate attractiveness of hedonic goods could even outweigh differences in utility. In addition, we chose somewhat more expensive goods. We chose an ‘exclusive coffee machine with which you can also make espresso’ (ƒ300) and a ‘complete coffee and tea service’ (ƒ350). The coffee machine was considered to be hedonic and the service was considered to be utilitarian. Subjects were given the same text as above but with other goods. Experiment 3: results

The exclusive coffee machine was rated as ‘primarily hedonic’ by 59%, by 23% as ‘both’, by 15.4% as ‘primarily utilitarian’ and ‘none’ by 2.6%. Hence, for the coffee machine the manipulation worked (one-sample t-test: t: 2.49, p < .05). The coffee and tea service was rated as ‘primarily utilitarian’ by 38.5%, by 33.5% as ‘primarily hedonic’, by 18% as ‘both’ and by 10% as ‘none’. Therefore the coffee and tea service was not clearly rated as utilitarian (one-sample t-test: t: 1.53; p = .133). Because, as intended, the coffee machine was rated as more hedonic than the coffee and tea service (one sample t-test: t: 4.3; p < .001) the goods were used nevertheless. As expected, mean selling prices were higher for the tea service than for the coffee machine: ƒ264 vs. ƒ246 (t: 2.5; p < .05). In a direct choice which of these two goods to receive immediately nevertheless 71% preferred to receive the more hedonic good sooner (χ2: 13.47; p < .001). 3.4.1. Conclusion

There is clear evidence for a greater preference to receive hedonic goods sooner than utilitarian goods. In three comparisons between goods, subjects preferred the more hedonic one immediately. Given that in such pairwise comparisons generally there is a strong tendency to be more impatient for goods which are valued more, it is especially striking that in experiment 3 subjects actually preferred to receive a lower valued hedonic good sooner than a higher valued utilitarian good.

60

Chapter 3

3.5. The influence of measurement on framing

According to the Discounted-Utility model different measures of time preference should yield identical discount rates. However, already Loewenstein (1988) demonstrated that temporal discount rates differ between choices that are framed as speeding-up and choices framed as delaying the receipt of a good (see chapter 1 for an extensive discussion). Framing theory, as presented in chapter 1, in addition predicts greater impatience for hedonic over utilitarian goods to depend on the measurement of impatience. More specifically, a stronger preference is expected for hedonic goods under measures that stress their hedonic aspects and a stronger preference is expected for utilitarian goods under measures that underline the utilitarian features. Generally, measures to elicit preferences that are 'scale compatible' with one of the options' attributes raise the salience of this attribute (Tversky, Kahneman & Slovic 1988). Asking subjects which good they would like to receive sooner may make hedonic considerations more salient, because this elicitation method focuses on the usage of the good and subjects can more easily imagine using hedonic goods. Asking subjects to specify monetary amounts as compensation charges for delay may underline the utilitarian features of an option, especially when larger amounts are involved. People might consider alternative utilization of the money when thinking about the correct trade-off between waiting time and compensation. Further, asking for compensation charges may provoke considerations as 'appropriateness' of the charge or considerations about the monetary costs of delaying an object. When subjects are forced to make a choice, the attributes attached to the options will influence framing (see Chapter 2). This has consequences for the way impatience can be measured. When the measurement does not itself introduce a frame, it will reflect the effect of the given attributes (here: ‘hedonic’ and ‘utilitarian’). If, however, the measurement itself introduces prominent attributes or makes existing attributes more salient, it will interfere with the framing effect and thus not measure what it is supposed to measure. In order to demonstrate this effect, we added a replication of experiments 2 and 3 in which (im)patience is measured by the subjects’ willingness to accept financial compensation. We argue that being forced to think in terms of the monetary value of their feeling of impatience leads subjects to adopt an utilitarian frame on the choice task and which will eradicate the difference between hedonic and utilitarian goods. Experiment 4: design

In the previous experiments subjects always had to choose to receive hedonic or utilitarian goods sooner. We therefore replicated experiment 2 and 3 but told subjects that they had won one good in a lottery and the lottery organizer subsequently asks them to delay receipt of this good for four weeks. Subjects in turn are offered to receive some financial compensation and then state whether they accept the compensation and are willing to wait or whether they would like to receive the good immediately. Thus, subjects again make binary choices, but this time are forced to think in terms of appropriate monetary compensation. If subjects are more eager to receive hedonic goods sooner, they should be less willing to accept the financial compensation offered for weeks delay. Subjects in group 1 were confronted with the following text, subjects in group 2 answered the same question for a cassette with cutlery (both in Dutch):

Intertemporal choices for hedonic and utilitarian goods

61

'Imagine: you participated in a lottery. You could not only win monetary prizes, but also consumer goods. You were so fortunate to win a nice sandwich toaster (official price: ƒ 40). What is the minimal amount for which you would sell the sandwich toaster immediately? ƒ________ Unfortunately, there is a temporary shortage in prizes. Thus, not every winner can receive his or her prize immediately. The organization of the lottery therefore asks you to wait four weeks for your prize. You have two possibilities (circle what you prefer): A:

You get the sandwich toaster immediately.

B:

You get the sandwich toaster in four weeks plus ƒ15 in compensation for waiting.'

Subsequently subjects in group 1 answered the same question for 'a complete coffee and tea set (official price: ƒ350)' and a financial compensation of ƒ40 for a four weeks delay. Subjects in group 2 answered the same question for 'an luxurious coffee machine with which you can also make espresso (official price: ƒ300)' and also a compensation of ƒ40 for a four weeks delay. Therefore subjects in both groups made choice for hedonic and utilitarian goods. Again order of presentation was varied within groups. Experiment 4: results

Mean selling prices in this experiment did not differ from experiments 2 and 3 (sandwich toaster: ƒ32, cassette with cutlery: ƒ34, luxurious coffee machine: ƒ237, and complete tea set: ƒ21451). For all goods there is a strong tendency to wait for the receipt and receive the respective compensation instead of receiving the respective good immediately: 84% of the subjects preferred to wait for the cassette with cutlery (n = 26) as well as for the sandwich toaster (n = 19). For the coffee machine 90.6% preferred the later option (n = 53) and all subjects preferred to wait for the tea set and receive financial compensation (n = 29, t: 2.33; p < .05). As expected, combining answers for all four goods reveals no difference between hedonic and utilitarian goods (t: .75; n.s.) when offering financial compensation. Financial compensation seems to raise the salience of the utilitarian frame and can even, as in this case, lead to a frame switch. It is not an alternative way to measure impatience, even though it is often used in experiments to do just that. This point is so important that we conducted a second experiment where financial compensation is involved, but this time we decided to use an experiment with real payoffs.

The latter value appears to be much smaller than in experiment 3 (subjects then stated ƒ264), because of the reduced sample. Comparing the part of the sample who stated selling prices in both questions reveals no difference (t: .54; n.s). 51

62

Chapter 3

Experiment 5: design

Absence of greater impatience for hedonic goods in the previous experiment might be due to the relatively large compensation offered for a four weeks delay. The offered amounts may have been so high that the compensation outweighed any potential differences between hedonic and utilitarian goods. In fact, between 84% and 100% chose to wait for the good and receive financial compensation. To rule out this possibility we conducted a final experiment. Since up to now all experiments involved hypothetical questions, we this time also investigated intertemporal preferences for hedonic and utilitarian goods in a setting that potentially affects payoffs. Subjects Subjects were 89 students at the university of Groningen from various disciplines recruited via e-mail and by flyers. Since the decision scenarios were varied between subjects, only part of the sample participated in the experiment presented below. Mean age was 22 years, 44% were males and 56% females, average monthly income was ƒ1057 ($500), and the main fields of study were psychology (21%), management science (20%), natural science (11%), sociology (10%), educational science (11%), and economics/econometrics (6%). One subject had to be removed from all analyses, because he did not understand several of the decision tasks. Method Subjects gathered in groups ranging from 2 to 23 persons to participate in the experiment. They were each provided with a booklet containing the experimental tasks and a questionnaire concerning the personal background of subjects. Experimental tasks comprised the choice problems described below, some other choice problems on intertemporal choices, presented partly in chapters 3 and 4, as well as several unrelated decision scenarios. The questionnaire items are described in more detail in chapter 5. A version of such a complete booklet is given in overall Appendix A at the end of this book. Since the decision scenarios were varied between subjects, only part of the sample received the choice problems presented below and n generally is smaller than 89. Subjects were told about the context in which the research took place and that there were no wrong or right answers. The experimenter was present during the whole experiment to answer questions concerning the questionnaire items or the questionnaire. Filling in the booklet took between 40 and 75 minutes, most people needed between 50 and 60 minutes. Subjects were paid ƒf10 (then approximately $4.00) and had an additional opportunity to win a gift certificate worth between ƒf15 and ƒ25 depending on their choices in this experiment. Again subjects chose between gift certificates for Compact disks and computer diskettes, but this time they had to specify appropriate intertemporal trade-offs. Subjects were told that they had won a gift certificate in a lottery. Subsequently they made a series of choices where they could exchange this original gift certificate for a larger but later certificate. One group won a certificate for CD's worth ƒ20, the other group had a certificate for computer disks also worth ƒ20. When handing out the booklets subjects were told that they had a 5% chance that one of their choices in this experiment would actually be paid to them. In addition, it was also stated in the text that choices in this experiment could affect actual payoffs. Subjects were also asked to rate the two goods according to their hedonicness by the same question as in experiments 1-3. Since experiment 1 demonstrated a clear preference to receive gift certificates for CD’s sooner than gift certificates for computer diskettes, we expected that, if both measures for

Intertemporal choices for hedonic and utilitarian goods

63

impatience yield identical results, the initial certificate for CDs should be kept and the certificate for computer diskettes should be exchanged for a later but larger one. Subjects were confronted with the following text (in Dutch): 52 ‘Suppose you have won a gift certificate for CD’s (worth 20 guilders) in a lottery. What follows is a series of choices. Each choice offers you the possibility between keeping your initial gift certificate you will get in one week or to exchange it for a different certificate at a different point in time. For each choice, circle the option you prefer. That is, circle either A or B. For example, suppose on choice number 4 you chose A instead of B. Hence, you prefer your initial certificate of f20 in one week over a certificate of f25 in 19 days. When everyone has completed their form one person is picked randomly. This person will actually get paid his or her gift certificate. It is randomly determined who wins, the choices of the winner determine the prize. 1. A gift certificate for f 20 in one week (A) or a gift certificate of f 25 in 10 days (B) 2. A gift certificate for f 20 in one week (A) or a gift certificate of f 25 in 13 days (B) 3. A gift certificate for f 20 in one week (A) or a gift certificate of f 25 in 16 days (B) 4. A gift certificate for f 20 in one week (A) or a gift certificate of f 25 in 19 days(B) 5. A gift certificate for f 20 in one week (A) or a gift certificate of f 25 in 22 days(B) 6. A gift certificate for f 20 in one week (A) or a gift certificate of f 25 in 25 days(B) 7. A gift certificate for f 20 in one week (A) or a gift certificate of f 25 in 28 days(B) 8. A gift certificate for f 20 in one week (A) or a gift certificate of f 25 in 31 days (B) 9. A gift certificate for f 20 in one week (A) or a gift certificate of f 25 in 34 days (B) 10. A gift certificate for f 20 in one week (A) or a gift certificate of f 25 in 37 days (B)’ Experiment 5: results

Subjects on average were willing to change their gift certificate for the CD for a larger certificate after 28 days or less (mean choice: 7; n = 33), while they were willing to wait approximately 25 days or less (mean choice: 5.8; n = 25) for a larger certificate for computer diskettes. However, this difference is not significant (t: 1.22; n.s.). Hence, when subjects make trade-offs we did not find higher discount rates, i.e. shorter delays, for hedonic goods, even though choices potentially affect real payoffs. These results may also be due to value differences. Suppose a ƒ20 gift certificate for CD’s is more valuable than a ƒ20 gift certificate for computer diskettes as it turned out in experiment 1. Assuming decreasing marginal utility exchanging the initial certificate adds more utility in case of the certificate for computer diskettes. Thus, subjects are more willing to wait for a smaller increase in utility in case of the certificate for the CD. In this experimental session we also conducted two other experiments where subjects had to state appropriate compensation charges for delaying hedonic and utilitarian goods. In experiment A subjects again had to delay expensive television sets and washing machines (both worth ƒ2400). In experiment B subjects stated compensation charges for the delay of a television set and a vacuum cleaner (both worth ƒ600). In both experiments subjects had to delay goods either 5 weeks or 9 weeks and write down an 52

The experimental design was adapted from Loewenstein (1988).

64

Chapter 3

amount sufficient to compensate them. Again, in both experiments we found no differences in compensation charges between hedonic and utilitarian goods. Though washing machines were valued slightly higher than television sets (ƒ2129 vs. ƒ2026, t: 2.14, p < .05; n = 88), in experiment A delay charges were equal for the 5 weeks delay (t: .81; n.s.). In the 9 weeks delay compensation charges were slightly higher for washing machines (t: 1.75; p = .084). Comparing discount rates (compensation charges divided by selling price) for each good reveals no difference between goods for either delay (t: 1.32; n.s. for the 5 weeks delay and t: 1.33; n.s. for the 9 weeks delay). In experiment B likewise an ANOVA with selling price for the goods and time delay as covariates revealed no effect of type of good (hedonic or utilitarian) on absolute compensation charges (F: 2.37; n.s.) or discount rates (F: 2.3; n.s.). 3.6. Overall conclusion and discussion

In choices where subjects have to state whether they would like to receive rewards sooner or later we find clear preference to obtain hedonic goods sooner, even when the utilitarian good is valued higher. This runs counter to both the standard Discounted-Utility model and the hyperbolic discounting model. These results indicate that the distinction between hedonic and utilitarian goods is useful in intertemporal choice as far as deprivation and impulsive behavior is considered. We did not observe such differences in immediate preference for hedonic or utilitarian goods when subjects had to state monetary compensation charges independent of whether compensation charges were fixed or had to be stated by subjects. If at all subjects charge higher monetary compensation for the delay of utilitarian goods. These results suggest that the hedonic/utilitarian distinction is much less valuable when financial considerations play a role in decision making. These results are in accordance with framing theory, since they suggest a greater salience of hedonic attributes when hedonic usage of goods is more important, viz., in decisions which good to receive sooner, and a greater salience of utilitarian attributes when functional considerations are more important as when determining adequate financial compensation. As a consequence of these results intertemporal utility seems to be more complex than it has been treated up to now. First, there is the valuation of goods as measured by selling price in the experiments reported above. Second, there is an immediate preference for goods that might differ from the valuation of these goods and for which the distinction between hedonic and utilitarian goods is highly relevant. Finally, there is intertemporal utility as measured by monetary trade-offs that appears to be distinct from immediate preference. Here discounting is much less sensitive to the difference between hedonic and utilitarian goods. We suggest that this is due to a framing effect comparable to scale compatibility. Forcing subjects to think in terms of monetary compensation makes utilitarian considerations such as alternative usage of the money more important and lowers the prominence of hedonic characteristics. In a similar fashion Prelec and Loewenstein (1998) report effects of the payment structure on hedonic consumption experience. There, thinking about the payment reduced subjects' utility derived from the consumption of goods. However, more research with respect to these phenomena is clearly warranted. With respect to the fact that in choices which good to receive sooner, experienced deprivation is stronger for hedonic than for utilitarian goods the experiments reported here supplement previous research by Metcalfe and Mischel (1999) and Loewenstein (1996). Loewenstein (1996) argues that impulsive behavior often occurs when visceral

Intertemporal choices for hedonic and utilitarian goods

65

factors, e.g., cravings or hunger, are involved. Hedonic feelings involved in decisions whether to obtain goods now or later in some respects seem to play a similar role in intertemporal choice. Though hedonicness generally will have weaker effects than visceral factors, the former might become strong enough to produce similar results: one subject motivating a compensation charge of ƒ1500 for a five weeks delay of a television worth ƒ3000 to him argued: ‘ this is a very nice television and I want it immediately! It is a shame that the lottery promises such a nice thing and then does not deliver it immediately!’ Children being denied sweets at a supermarket also show reactions suggesting visceral factors instead of denial of hedonic consumption experience. Metcalfe and Mischel argued that in intertemporal choice the distinction between ‘hot’ and ‘cool’ aspects of a decision situation or presentation is relevant (Metcalfe & Mischel 1999). Hot decisions situations, e.g., presence of a reward, raise the need for immediate gratification, while cold aspects of a situation, e.g., abstract presentations of the reward, lower it. The hedonic/utilitarian distinction seems to be more precise than the hot/cool distinction, since goods can and have been rated consistently by a variety of subjects and a variety of goods, while Metcalfe and Mischel use their distinction to interpret three decades of Mischel's own research on children's willingness to delay gratification. However, the hot/cool distinction appears to be more generally applicable.

Chapter 4 Vividness in intertemporal choices between hedonic and utilitarian goods

Summary

In this chapter we investigate the influence of vividness on the desire to receive hedonic and utilitarian goods immediately. In analogy to chapter three also the influence of vividness on monetary compensation charges for delaying hedonic and utilitarian goods is investigated. Several researchers have suggested an influence of vividness on intertemporal preference, but there is disagreement about whether vividness raises or lowers impatience and how vividness affects the desired receipt of hedonic or utilitarian goods. In a series of experiments, we demonstrate no effect of vividness on monetary discount rates, but vividness strongly raises the attractiveness of hedonic goods when people choose whether to receive hedonic or utilitarian goods sooner. These results are in accordance with framing theory.

68

Chapter 4

4.1. Introduction

In the previous chapter we demonstrated greater impatience for hedonic goods than for utilitarian goods when subjects choose which of these goods to receive sooner. When subjects have to determine monetary compensation charges for delaying hedonic or utilitarian goods these differences in immediate preferences were much less pronounced. We now investigate intertemporal choices between hedonic and utilitarian goods in more detail, viz., how vividness affects impatience to receive either kind of good. An impact of vividness on intertemporal choices has been suggested by several authors (e.g., Mischel, Shoda & Rodriguez 1992; Becker & Mulligan 1997; Loewenstein 1996), but these suggestions are contradictory. Becker and Mulligan (1997) argue that vividness lowers impatience due to a perceived reduction in temporal distance, Loewenstein (1996) argues that vividness raises impatience when visceral factors, e.g., hunger, are involved, while Mischel et al. (1992) conclude that different kinds of vivid presentation, e.g., presence of the reward or photography of the reward, may either decrease or increase impatience, depending on whether the presentation of the reward focuses on its abstract or arousing qualities. We argue that vividness affects hedonic and utilitarian goods differently, since a vivid presentation raises the ease with which hedonic or utilitarian use of a good can be imagined. Because hedonic and utilitarian goods elicit different levels of deprivation, and consequently different levels of myopia, vividness is expected to raise impatience for hedonic goods more strongly than for utilitarian goods. As in the previous chapter, vividness should affect preference for hedonic over utilitarian goods stronger in choices about which good to receive sooner than when stating monetary compensation charges for delaying either good. In the following, we summarize the competing considerations about the influence of vividness on intertemporal preferences in more detail and then present some experiments testing their correctness. 4.2. Theory

In the psychological literature there has been extensive research on the effect of vivid information on behavior, yet the empirical evidence concerning this effect is, at best, mixed. Nisbett and Ross (1980: 45) define information as vivid ‘to the extent that it is a) emotionally interesting, b) concrete and image-provoking, and c) proximate in a temporal, spatial or sensory way’. Additionally, they state that emotional interest rises with social closeness and the hedonic relevance of events in a scenario to its participants. Nisbett and Ross (1980), mostly considering anecdotal evidence, take for granted that ‘people’s inferences on behavior are so much more influenced by vivid, concrete information than by pallid and abstract propositions of substantially greater probative and evidential value’ (1980: 44). Consequently, they immediately turn to the question what causes this effect, arguing that vivid information is more likely to be stored in memory and remembered than non-vivid information. Vivid information therefore is more accessible (Tversky & Kahneman 1973) than non-vivid information. In this view vividness potentially affects impatience for hedonic and utilitarian goods in several ways, making predictions concerning its overall effect difficult. First, this reasoning suggests a value effect. That is, vivid goods are valued higher, raising impatience for them, since people immediately prefer larger rewards to smaller rewards. Further, this value effect should be stronger for hedonic than for utilitarian goods, since hedonic goods are inherently more vivid and emotionally interesting than utilitarian

Vividness in intertemporal choices between hedonic and utilitarian goods

69

ones.53 Second, a vivid presentation may make delayed goods appear to be temporally closer than they actually are, i.e., it shrinks the perceived temporal distance. A vivid presentation thus, should lead to less discounting. Similarly, economists Becker (1996) and Becker and Mulligan (1997) argue that discount rates are influenced by the degree to which people can imagine the future vividly. People with a high stock of visualization capital can imagine the future better and therefore value it higher and discount less. Here again, vividness shrinks the perceived temporal distance of delayed rewards and therefore makes people more patient. Fiske and Taylor (1984) review the effect of vivid stimuli on behavior in general, but summarize: ‘there is little empirical evidence for vividness effects... in sum, vividness does not work well empirically, although intuitively it seems as if it should’ (1984: 391). This has been corroborated by recent research: Jenni and Loewenstein (1997) found no vividness effect when investigating why people are much more willing to save identifiable instead of statistical victims. It seems that vividness, if at all, only works under very special conditions (McGill & Anand 1989, Anand & Keller 1998), e.g., when subjects are explicitly told to imagine goods vividly and follow their emotions when making choices. From research on vividness in general, we thus would expect no effect of vividness for intertemporal choices. Previous experimental research on delay of gratification however suggests that vividness is at least sometimes important in intertemporal choice. Drawing on research on pre-schooler’s ability to delay gratification, Mischel and his colleagues distinguished arousing and abstract presentation of rewards. In an arousing presentation the motivating qualities of the stimulus were stressed and the completion of the action sequence was suggested, e.g., eating food. In an abstract presentation the focus was more on the informative, symbolic aspects of the stimulus. This distinction seems to match the distinction between hedonic and utilitarian goods. Hedonic goods provide direct utility derived from the completion of action sequences and therefore are potentially arousing, while the utilitarian goods serve as means to derive utility from consumption and are more abstract. Mischel et al. (1992) report several experiments about pre-schoolers’ ability to delay gratification. The children were presented with several rewards differing in value, e.g., toys or snacks. The experimenter then left the room and asked the children to wait an unspecified amount of time (typically about 15 minutes) for the experimenter to return. When the children were able to wait until the experimenter returned, they were promised to obtain the larger of the two rewards, e.g., five pretzels instead of two small cookies. Children who were unwilling to wait could call the experimenter at any time by ringing a bell to receive the smaller reward, the two cookies, immediately. Contrary to the investigators’ expectation, presence of either the immediately available smaller or the delayed larger reward decreased waiting time, i.e., the average time children waited before ringing the bell. Vividness, i.e., presence of either reward, thus increases impatience. Inclining subjects to think about the rewards yielded the same result, whereas providing distracting (’fun’) thoughts, unrelated to the objects, raised 53

'Raising impatience' here is a somewhat tricky term. One may argue that due to the magnitude effect presented in Chapter 1 higher valued goods are discounted less strongly, and hence people are more patient. Suppose a person has to choose between two goods presented non-vividly and is indifferent which good to receive sooner. A vivid presentation of one good now potentially raises its value and results in a preference for the good presented vividly to be received sooner. The person therefore is more eager to receive this good, though his discount rates may decline. In the experiments below both discount rates and preference for immediate receipt are investigated to separate these effects.

70

Chapter 4

waiting time. Distracting thoughts were also able to cancel out the effect of exposure to either reward: children waiting in the presence of the reward and being provided with distracting thoughts waited roughly as long as children waiting in the absence of rewards and distracting thoughts. Vividness therefore raises impatience. However, providing children with photos of the reward increased their waiting time, implying that vividness can also lower impatience. Mischel et al. (1992: 154) concluded: ‘Thus different modes of presenting rewards may either hinder or enhance self-control’. Unfortunately, this approach then shifts the problem to what makes a presentation arousing or abstract. However, generally we would expect less patience and more discounting for arousing hedonic goods and more patience and less discounting for abstract utilitarian goods when presented vividly. In addition, Loewenstein (1996), when discussing why myopic behavior occurs frequently when ‘visceral factors’ are involved, argues: ‘Immediate emotions arising from future events are inevitably linked to some mental image or representation of those events.... the more vivid such images are, and the greater detail with which they are recalled, the greater the emotional response’ (Loewenstein 1996: 280). This emotional response directly affects the decision maker’s level of deprivation due to non-possession of an object (see Chapter 3 for further discussion): higher emotional responses lead to stronger deprivation. Loewenstein (1996) interprets the Mischel et al. results such that presence of rewards creates deprivation, which children wanted to overcome by consuming the earlier but smaller reward. This view is consistent with the reasoning in chapter three as well as Hoch and Loewenstein (1991). However, since Loewenstein (1996) restricts his analysis to situations where ‘visceral factors’ are involved, his analysis does not easily transfer to hedonic and utilitarian goods. While hedonic goods appear to be clearly more visceral than utilitarian goods and we thus would expect vividness to raise impatience for them more strongly, it is doubtful whether hedonic goods are able to evoke the strong physical reactions, e.g., deprivation of drugs involves for addicts. Hence, the arguments concerning visceral factors cannot be applied to hedonic goods without additional assumptions and from a visceral factors perspective we therefore expect no effect of vividness on intertemporal preference between hedonic and utilitarian goods. The reasoning put forth in chapter three suggests an effect of vividness on intertemporal preferences. In chapter three we modeled deprivation for hedonic and utilitarian goods as follows: Dh = a(P - r) + bh(r-0) > Du = a(P - r) + bu(r-0),

(1)

a being the slope of the consumer's value function in the positive region, b its slope in the negative region, and r is the level of adaptation between purchase (P) and non-purchase (0). For example, consumers being able to imagine the possession of a good very well, adapt more strongly to its possession and are more likely to buy the product. Desire D represents the consumer’s motivation to purchase a product. This motivation depends on the consumer's deprivation due to current non-possession of a product. Consequently, impatience rises with deprivation and especially is stronger for hedonic than for utilitarian goods. A vivid presentation now affects impatience in two ways. First, making a good appear more hedonic or more utilitarian due to a vivid presentation affects loss aversion for this good as measured by b in formula (1) above. Since loss aversion

Vividness in intertemporal choices between hedonic and utilitarian goods

71

and deprivation due to non-possession are stronger for hedonic than utilitarian goods, deprivation and consequently impatience is raised more strongly when a vivid presentation makes a good appear more hedonic than when making a good appear more utilitarian. Second, because hedonic usage is easier to imagine, it is easier to adapt to the possession of hedonic instead of utilitarian goods when goods are presented vividly. This second effect of vividness thus pertains to the parameter a in formula 1. Extending equation 1 by assuming easier adaptation to the possession of hedonic goods and by assuming that vividness makes goods appear more hedonic or utilitarian respectively has two consequences. First, it raises deprivation for both goods as in equations 2 and 3: Dhv = av(P - r) + bhv(r-0) > Dh = a(P - r) + bh(r-0), for 1 ≥ av > a ≥ 0, and bhv > bh Duv = av(P - r) + buv(r-0) > Du = a(P - r) + bu(r-0), for 1 ≥ av > a ≥ 0, and buv > bu

(2) (3)

Second, it raises the inequality in equation 1 above. That is, a vivid presentation raises deprivation for hedonic goods more strongly than for utilitarian goods: Dhv = av(P - r) + bhv(r - 0) Duv = av(P - r) + buv(r - 0) > Dh = a(P - r) + bh(r - 0) Duv = a(P - r) + bu(r - 0) for 0 < au < ah ≤ 1, and bhv > buv.

(4)

Consequently, vividness raises impatience for both goods but more strongly for hedonic than for utilitarian goods (hypothesis 1). This reasoning is based on subject's experienced level of deprivation. In order to distinguish this ‘pure deprivation effect' from a framing effect (in the sense of framing theory), we will again use frame-sensitive measurement as in Chapter 3. Framing theory, as presented in Chapter 1, again predicts the effect of vividness to differ with the measurement of impatience. As in Chapter 3, we expect a stronger preference for hedonic goods to be received sooner under measures that do not stress utilitarian aspects and a stronger preference for utilitarian goods to be received sooner under measures that underline the utilitarian features. Again, monetary amounts should underline the utilitarian features of an option, because people might consider alternative utilization of the money when thinking about the correct trade-off between waiting time and compensation. Further, asking for compensation charges may provoke considerations such as 'appropriateness' of the charge or considerations about the monetary costs of delaying an object. Therefore, utilitarian goods should be more attractive when monetary compensation charges are involved, while hedonic goods are more popular when subjects simply state which of two goods they would like to receive sooner (hypothesis 2). Note, that this reasoning extends the 'pure deprivation effect', because the effects predicted in formulae 2-4 hold irrespective of the elicitation method. In the following paragraph we will have a closer look on the predictions of the other above-mentioned approaches on such a measurement effect. Applying Nisbett and Ross' considerations to intertemporal choices does not result in specific predictions how measurement affects impatience. Recall that their arguments concerning vividness pertained to a rise in value as well as to a shrinkage in perceived temporal delay. In may well be that different elicitation methods affect the valuation and perceived time delay for hedonic and utilitarian goods differently but to derive

72

Chapter 4

predictions, additional assumptions seem necessary. Fiske and Taylor expect no effect of vividness anyway and hence also no measurement effects, while for Becker and Mulligan, staying within the framework of neoclassical economics, different elicitation methods should yield identical results. Loewenstein's visceral factors perspective likewise suggests no effect of different elicitation methods. Since visceral factors exert physical influence on behavior, it is unclear why this effect should be dependent on the way impatience is measured. The only approach compatible with an effect of vividness depending on measurement is Mischel's arousing/abstract distinction. According to this view, impatience depends on whether abstract or arousing features of the reward are stressed. Asking subjects which of two goods they prefer to receive sooner then possibly focuses on the arousing features of the goods while asking for monetary compensation charges possibly focuses more on the abstract qualities of the good. Hence, Mischel's arousing/abstract distinction predicts similar measurement effects as framing theory. However, Mischel's position differs from the deprivation and framing arguments above, since it suggests greater impatience for hedonic goods when presented vividly and less impatience for utilitarian goods when presented vividly. This asymmetric effect of vividness on goods is due to the fact that for utilitarian goods when presented vividly the focus is likely to lie on the abstract qualities of the reward, thus making it even more abstract. Further, recall that presenting children with photos of the rewards lowered impatience. Since, the rewards were toys or snacks, they can safely considered to be hedonic. Likewise, photos can be considered more vivid than a pale description of the reward and should therefore, according to formulas 2 - 4 above, raise impatience. Table 4.1 summarizes the different approaches and their predictions with respect to the effects of vividness and measurement on discounting. Table 4.1 Different approaches and predictions on the effect of vividness on discounting Model Nisbett & Ross Fiske & Taylor Becker & Mulligan Mischel Loewenstein Deprivation/ Adaptation effect Framing theory

Effect of vividness on impatience Raises value and shrinks perceived time delay. Overall effect: unclear No effect of vividness Shrinks time delay, lowers discounting Depending on whether abstract or arousing properties are stressed: raises impatience for hedonic goods, lowers impatience for utilitarian goods. Effective when visceral factors are involved, no statement for hedonic or utilitarian goods Vividness raises impatience, but more strongly for hedonic goods Raises discounting for both goods, induced effects are confounded with measurement: greater impatience for hedonic goods in choices, greater impatience for utilitarian goods in financial compensation

Measurement effect No prediction No No Yes, depending on abstract or arousing presentation No No Yes

In the following, three experiments will be presented. Experiment 1, asking subjects whether they wold like to receive hedonic or utilitarian goods sooner, first investigates whether vividness affects impatience at all, second, it investigates whether vividness affects hedonic and utilitarian goods differently, and third, whether such a

Vividness in intertemporal choices between hedonic and utilitarian goods

73

potential effect of vividness is due to a rise in the valuation of goods. In experiments 2 and 3, it is additionally investigated whether vividness affects impatience identically irrespective of the way impatience is measured. The question explored is, whether the potential effect of vividness also holds when subjects have to determine adequate financial compensation for delaying hedonic and utilitarian goods. Experiment 2 offers a fixed financial compensation for delaying goods, whereas in experiment 3 subjects are asked to specify exact monetary compensation charges for the delay of hedonic and utilitarian goods. 4.3. Experiments and results Experiment 1 Subjects Subjects were 89 students at the university of Groningen from various disciplines recruited via e-mail and by flyers. Mean age was 22 years, 44% were males and 56% females, average monthly income was ƒ1057 ($500), and the main fields of study were psychology (21%), management science (20%), natural science (11%), sociology (10%), educational science (11%), and economics/econometrics (6%). One subject had to be removed from all analyses, because he did not understand several of the decision tasks. Method Subjects gathered in groups ranging from 2 to 23 persons to participate in the experiment. They were provided with a booklet containing the experimental tasks, i.e., choice problems, and a questionnaire concerning the personal background of subjects described in detail in Chapter 5. A version of the complete booklet is presented in Appendix A at the end of this book. The experimental tasks consisted of experiments 1 and 3 below, several other experiments on intertemporal choices partly described in Chapters 2 and 3, and several unrelated decision scenarios. Subjects were told about the context in which the research took place and that there were no wrong or right answers. The experimenter was present during the whole session to answer questions concerning the choice tasks, the questionnaire items, or the questionnaire. Filling in the booklet took between 40 and 75 minutes, most people needed between 50 and 60 minutes. Subjects were paid ƒ10 (approximately $4.50) and had an additional opportunity to win a gift certificate worth between ƒ15 and ƒ25 depending on their choices in another experiment (experiment 5 in Chapter 3). Experiment 1: design

The previous chapter demonstrated a greater preference to receive hedonic goods sooner than utilitarian goods when subjects have to choose which of the two goods to receive sooner. To investigate whether and how vividness influences intertemporal preference for hedonic and utilitarian goods subjects were told that they had won a hedonic and a utilitarian good in a lottery. Subsequently, they were asked to chose which good they prefer to receive sooner. The goods were described either both vividly or both nonvividly. A television set was considered a hedonic good, a washing machine was chosen

74

Chapter 4

as a utilitarian good.54 Washing machines and television sets were chosen because of subjects’ familiarity with them and because both goods provide constant streams of utility. In fact 80% of the subjects in a control question indicated that they owned a washing machine and 92% owned a television. Both goods were offered simultaneously. In the non-vivid condition subjects were only told that they had won ‘an expensive television (list price ƒ2400)’ and ‘an expensive washing machine (ƒ2400)’. In the vivid condition subjects were told that they had won 'a flat-screen television with Dolby stereo to create cinema atmosphere (official price ƒ2400)’ and 'a high quality long-durable Miele washing machine with low energy use and five-year guarantee (official price ƒ2400)’. Note, that for the hedonic good the hedonic attributes are made more vivid, while for the utilitarian good utilitarian features are made more vivid. Prices for both goods given in the text were 'real prices' of comparable goods available in stores. In addition, selling prices for both goods were elicited. Eliciting selling prices is necessary to control for possible differences in value between the vivid/non-vivid conditions as well as between the hedonic/utilitarian conditions. For example, subjects may value a television set presented vividly higher than a television set presented nonvividly as suggested by Nisbett and Ross (1981). This in turn may affect preferences for which good to receive sooner. To suppress such possible differences in valuation between the vivid and non-vivid condition, in the non-vivid condition the information that the good is expensive was given. An expensive good suggests superior quality and thus it has to have some special features. To control for the fact that waiting for a washing machine and waiting for a television set may create different costs - not being able to watch television is one thing, to dwell in a flat covered with smelly laundry another - current alternatives to the possession of these goods were specified (‘you used a laundry to wash your clothes’, and ‘you watched TV at your friends place’) associated with roughly equivalent costs. To further control for possibly different costs, at the end of the experimental part in the booklet subjects stated selling prices and monetary compensation charges for both goods and motivated their responses. In this control question goods were presented either vividly or non-vividly according to this experiment. That is, subjects who were confronted with a vivid scenario in experiment 1 also stated selling prices and compensation charges for vivid television sets and washing machines. If, due to laundry costs, subjects perceive the costs of delay to be higher for washing machines than for televisions, this should be reflected in their arguments. Note, that such potentially different costs for delay between washing machines and television sets do not affect the effect of vividness on preference for immediate receipt. If delaying washing machines creates higher costs for subjects than delaying television sets their preference for immediate receipt of the utilitarian good is raised in both the vivid and non-vivid condition. Time delay was varied in this experiment as well. In one condition ('immediate condition') both goods were available either ‘now' or 'after four weeks’. In the other condition ('remote condition') receipt of both goods was shifted in four weeks in the past. Hence, the respective time delays were ‘four weeks' and 'eight weeks’. Time delay was additionally varied, because the effect of vividness might be restricted to options that are available immediately. For example, Mischel’s arousing/abstract distinction and 54

The experiment was carried out in the absence of major sporting events that may cause temporary rises in the utility derived from immediately getting a television.

Vividness in intertemporal choices between hedonic and utilitarian goods

75

Loewenstein’s visceral factors perspective both seem to suggest that vividness raises impatience for the receipt of goods more strongly, if the goods are available immediately. After completion of the experiment the whole sample (n = 88) classified the goods used as either type A: ‘hedonic’, type B: ‘utilitarian’, type C: ‘both’ or type D: ‘neither’.55 Subjects were confronted with the following text (in Dutch, vivid condition): ‘Imagine: you have participated in a lottery. You could not only win monetary prizes but also consumer goods. You were so fortunate to win a television set and a washing machine. It is a flat-screen television set with Dolby stereo to create cinema atmosphere (official price of ƒ2400) and a high-quality long-durable Miele washing machine with low energy use and five-year guarantee (with an official price of ƒ2400, too). You did not have a television set or a washing machine at home. You watched television at friends or parents and did your washing at a laundry. What is the minimal amount for which you would like to sell this television set: _________? What is the minimal amount for which you would like to sell this washing machine: __________? Unfortunately, there is a temporary shortage in prizes, hence, not every winner can obtain his or her prize immediately. The organization of the lottery therefore asks you to wait for your prize. The organization is completely trustworthy. You have two possibilities (circle which you prefer): A:

You get the television set immediately and the washing machine in four weeks.

B:

You get the washing machine immediately and the television set in four weeks.’

Order of presentation for choices A and B was varied as well. That is, for part of the subjects getting the washing machine immediately was presented as option A. Note, that with this design it is only possible to distinguish whether vividness affects impatience for hedonic and utilitarian goods differently. When both goods are presented vividly either the hedonic good becomes more preferred, the utilitarian good becomes more preferred, or the relative desirability of both goods remains the same. It is not possible with this design to test whether impatience for both goods increases or decreases. This question will be addressed in experiments 2 and 3. Experiment 1: results Manipulation check The hedonic/utilitarian classification in fact worked as intended. For the whole subject pool washing machines were classified as 'primarily utilitarian' by 95.5% and as 55

The exact wording was: ‘Some people distinguish between goods. Some goods are pleasant and fun, give pleasure and appeal to your senses, e.g., perfume (type A). Some goods are useful, practical, functional, something that helps you to achieve a goal, e.g., a vacuum cleaner (type B). Could you indicate for each of the following goods whether they are type A, type B, both, or none. Circle.’

76

Chapter 4

'both hedonic and utilitarian' by 4.5% (χ2: 39.4; p < .001). Television sets were classified as 'primarily hedonic' by 75%, as 'both hedonic and utilitarian' by 20.5%, as 'primarily utilitarian' by 2.3% and as 'none' by 2.3% (χ2: 27.41; p < .001). There was no effect of vividness on classification. Thirty-nine subjects answered questions for the vivid goods, 49 for the non-vivid goods. As to different costs associated with the delay of the goods, there is some evidence that subjects in fact perceive the costs for delaying receipt of a washing machine to be higher than costs for delaying the receipt of a television set. In the control question at the end of the experimental booklet 42% of the subjects mentioned rental costs when motivating their financial compensation charge for delaying a washing machine, while only 1 subject mentioned rental costs in case of the television set. Selling prices Selling prices for the whole sample were approximately equal for hedonic and utilitarian goods (ƒ2010 vs. ƒ2032; t: 47; n.s.). Vividness did neither significantly affect selling prices for television sets (vivid condition: ƒ2072, non-vivid condition: ƒ1961; t: .88; n.s.) nor for washing machines (vivid condition: ƒ2072, non-vivid condition: ƒ2001; t: .55; n.s.). Thus, contrary to the suggestion of Nisbett and Ross, there is clearly no value effect present: if vividness affects intertemporal preferences, this is not due to differences in the valuation of goods. Comparing individual differences in the valuation of the two goods reveals a similar picture. In the non-vivid condition 26.5% of the subjects valued the television set more than the washing machine, 38.8% valued both goods equally, and 34.7% valued the washing machine more than the television. In the vivid condition the percentage of subjects who value the television set more than the washing machine remains stable (26.5%) and the percentage valuing both goods equally increases somewhat (44.1%). Consequently, slightly fewer subjects than in the non-vivid condition value the washing machine more than the television set (29.4%). None of these differences was significant. Impatience Preference to receiving hedonic or utilitarian goods sooner is affected strongly by vividness. In the non-vivid scenario where goods could be received either immediately or after a four weeks delay (non-vivid immediate condition) only 9% prefer to receive the television set sooner and 91% prefer to receive the washing machine sooner. In this immediate condition, preference for the hedonic good rises strongly when both goods are presented vividly: 56% now prefer to receive the television set immediately and to delay receipt of the washing machine. In the remote condition the picture is similar. In the nonvivid remote condition a majority of 60% prefers to receive the washing machine sooner and only 40% prefer to receive the television set sooner. Again, vividness strongly raises preference for the hedonic good. A vast majority (66%) prefers to receive the television set sooner when both goods are available after 4 or 8 weeks and only 34% prefer to receive the washing machine sooner. Table 4.2 summarizes differences in the valuation between goods (first 3 rows) and preference to receive the utilitarian good, i.e., the washing machine, sooner (rows 4 and 5). Values between brackets denote preference to receive the hedonic good, i.e., the television set, in the respective condition sooner.

Vividness in intertemporal choices between hedonic and utilitarian goods

77

Table 4.2 Selling prices for both goods and preference to receive the utilitarian good sooner (between brackets preference to receive the hedonic good sooner)

Vivid Non-vivid

TV > WM

Equal

TV < WM

26.5% 26.5%

44.1% 38.8%

29.4% 34.7%

Now vs. 4 weeks 44% (56%) 91% (9%)

4 weeks vs. 8 weeks 34% (66%) 60% (40%)

N 39 49

We carried out a logistic regression analysis with ’choice for getting the utilitarian good sooner’ as the dependent variable. Independent variables were 'differences in selling prices' (selling price for the TV set – selling price for the washing machine); 'time delay' (dummy: ‘now vs. 4 weeks’ and ‘4 weeks vs. 8 weeks’); and 'vividness' (dummy: yes or no). This model predicts preferences between hedonic and utilitarian goods very well (χ2: 13.47; p < .001, Nagelkerke Pseudo R2: .48). As predicted there are strong effects of 'differences in selling price' and 'vividness'. Valuing the television higher than the washing machine strongly lowers the likelihood of preferring the washing machine sooner (Wald: 11.95; p < .001). A vivid presentation of both goods likewise strongly lowers preference for getting the washing machine sooner (Wald: 10.77; p = .001).56 That is, in choices which good to receive sooner vividness relatively raises impatience for hedonic goods. Though longer time delay lowers the probability of preferring the washing machine sooner, this effect is not significant (Wald: .92; n.s.). Table 4.3 summarizes results. Table 4.3 Factors affecting preference for receiving the washing machine sooner B Model: Chi-square Pseudo R2 Coefficients: Difference in selling prices (TV-set - Washing machine) Time delay Vividness Constant

S.E.

Wald

38.73 .48

Sig.

.0000 -.0038

.0011

11.95

.0005

-.537 1.881 -1.7

.572 .573 1.28

.924 10.77 1.76

.3374 .0010 .1842

N = 88

Discussion: The results above clearly reveal a stronger effect of vividness for hedonic than for utilitarian goods. This effect is not due to changes in the valuation of both goods and is not dependent on the time delay after which goods can be obtained. This strong effect of vividness is especially striking, given that the vividness manipulation for both goods consisted of only one sentence and was not very strong. In addition, there is as strong tendency to prefer to receive the higher valued good sooner. This tendency is in

56

The coefficient for vividness is positive, because the coding was (1 = Yes; 2 = No). A higher value for vividness thus raises the likelihood of preferring the washing machine sooner.

78

Chapter 4

accordance with the results in Chapter 3 as well as the DU-model and is taken as a validation of the experimental design and the goods chosen. However, the large preference to receive the utilitarian good in the non-vivid condition sooner is surprising. While only 35% valued the washing machine higher than the television set, 91% preferred the washing machine sooner in the non-vivid immediate condition, i.e., in a setting that resembles the experiments in the previous chapter. It was already mentioned that delaying these two goods might involve different costs. Since a laundry involves monetary and temporal costs, subjects might be more aversive to use it for washing than to visit friends for watching television. In the non-vivid condition this effect may overweigh the stronger deprivation for hedonic over utilitarian goods, while vividness raised deprivation for the television set so strongly that the difference in costs is overweighed. There is indeed some evidence for this reasoning. First, in the control question at the end of the booklet 42% of the subjects mentioned rental costs when motivating their financial compensation charges for the washing machine, while only 1 subject mentioned rental costs in case of the television set. Hence, subjects do seem aware of different costs for delaying these two goods. Regressing selling prices for washing machines reveals that subjects who mention rental costs charge marginally higher selling prices (Beta: .21; t: 1.85; p = .07), indicating that the awareness of different costs affects variables relevant for delay.57 Further, subjects indicating that rental costs play a role on average value the washing machine higher than the television (difference in selling prices: ƒ203), while subjects who base their compensation charges on other grounds value the television higher (difference in selling prices: ƒ86). This difference is significant at p < .01 (t: 3.08), indicating that rental costs affect influential variables. Finally, it is possible that there is a bias towards equating selling prices for both goods. Recall, that both goods were presented simultaneously with identical official prices given. Subjects might argue that since goods are presented next to each other with equal official prices identical selling prices should be charged. This might compress existing differences in valuation, which are reflected in a greater preference to receive the higher valued good immediately. The following two experiments will investigate this final explanation more closely by using again washing machines as a utilitarian and television sets as a hedonic good. Because in these experiments the goods are presented to subjects simultaneously, we can check whether or not a bias towards equating the selling prices in experiment 1 is a reasonable interpretation of the results. Experiment 2

Experiment 1 demonstrates that vividness raises the attractiveness of hedonic goods more than of utilitarian goods. However, it did not allow us to judge whether vividness absolutely raises or lowers impatience. It may be possible that vividness lowers impatience for both goods, but more so for utilitarian goods and that vividness raises impatience for both goods but more so for hedonic goods. Further, from chapter three we 57

In a logistic regression analysis 'reason for delay charge' (dummy: mentioning rental costs yes or no) affects preference to receive the washing machine sooner, when vividness is included as the only other 2 covariate in the model (χ = 20.91; p < .001; both variables significant at p < .05). Additionally including difference in selling price leads to insignificance of mentioning rental costs (Wald: 1.72; p = .19) but 2 improves the model (χ = 31.27, p < .0001). Contrary to the model presented in table 4.3., vividness now has a stronger influence than differences in selling price (Wald: 10.77 vs. 6.79).

Vividness in intertemporal choices between hedonic and utilitarian goods

79

know that preferences in direct choices may not correspond to choice situations where subjects have to determine adequate financial compensation. To find out whether vividness raises or lowers impatience for both goods and whether vividness raises the immediate attractiveness of hedonic goods more strongly than for utilitarian goods also when subjects have to translate their impatience into adequate financial compensation charges another experiment was carried out. Subjects Subjects were 80 students at the university of Groningen from various disciplines recruited via e-mail and by flyers. Mean age was 22 years, 42% were males and 58% females, average monthly income was ƒ1100 ($500), and the main fields of study were psychology (30%), management science (11%), languages (8.5%), sociology (8.5%), law (7%), economics/econometrics (8.5%) and natural sciences (7%). Method Subjects gathered in groups ranging from 6 to 15. They were assigned to a computer where the program concerning the experiments described in chapter 7 had been installed and instructed about the context in which the research took place. They were told about the context in which this research took place and that there were no wrong or right answers. They were confronted with choice problems presented on a computer screen and with choice problems and questions concerning their personal background presented on paper.58 Choice problems on paper as well as the questions on paper were announced and presented to subjects as regular breaks in answering the choice problems at the computer screen. The experimenter was present during the whole experiment in order to answer questions concerning the tasks. Completing the whole experiment took between 50 and 110 minutes, most people needed between 80 and 90 minutes. Subjects were paid ƒ25 for their participation. Experiment 2: design

We again used the lottery scenario and offered subjects television sets (hedonic goods) and washing machines (utilitarian goods). Subjects then stated selling prices for the respective goods, thus again investigating whether potential effects of vividness on intertemporal preference are due to a change in the valuation of goods. Contrary to experiment 1, this time subjects were not offered both goods simultaneously, but subsequently. That is, subjects were told that they had won one good in a lottery that has to be delayed. Subjects then could either receive the good immediately or wait 4 weeks and receive a fixed financial compensation (ƒ100). Then the same text with the same choice tasks was presented to the subjects but with another good. Vividness was manipulated as in experiment 1. That is, in the non-vivid condition subjects were only told that they had won an expensive television (official price: ƒ2400) or an expensive washing machine (official price: ƒ2400). In the vivid condition subjects were told that they had won 'a flat-screen television with Dolby stereo to create cinema atmosphere 58

The choice problems on the computer screen are presented in chapter 7, the questions concerning the personal background of subjects are presented in chapter 5, and the other choice problems presented on paper are partly described in this chapter and partly consisted of other intertemporal problems not used for this dissertation.

80

Chapter 4

(official price: ƒ2400)’ and 'a high quality long-durable Miele washing machine with low energy use and five-year guarantee (official price: ƒ2400)’. Hence, this experiment tests not only whether there is a greater immediate attractiveness for hedonic than for utilitarian goods under a different measure for impatience, but also whether vividness absolutely raises the immediate attractiveness for both hedonic and utilitarian goods. Because in this experiment subjects made their choices for both goods subsequently, we additionally investigate whether the equation of selling prices observed in experiment 1 results from equal valuation or whether this result should be interpreted as a bias towards equal selling prices elicited by the equal monetary prices given in the experiment. Subjects were confronted with the following text (in Dutch, vivid condition, television set): ‘Imagine: you have participated in a lottery. You could not only win monetary prizes but also consumer goods. You were so fortunate to win a television set. It is a flat-screen television with Dolby stereo to create cinema atmosphere with an official price of f2400. You did not have a television at home. You watched television at friends. What is the minimal amount for which you would like to sell this television: _________? Unfortunately, there is a temporary shortage in prizes, hence, not every winner can obtain his or her prize immediately. The organization of the lottery therefore asks you to wait for your prize. You have two possibilities (circle which you prefer): A:

You get the television set immediately.

B:

You get the television set in four weeks plus ƒ100 as compensation for waiting.’

Subsequently subjects answered the same question for ‘a high quality Miele washing machine with low energy use and five-year guarantee (official price ƒ2400)’. Order of presentation was varied, part of the sample first answered questions for the hedonic good and part of the sample received questions for the utilitarian good first. In addition, also order of presentation for options A and B was varied. That is, part of the sample received a text where 'You get the television immediately' was presented as option B. Subjects either had to make choices for two vivid or two non-vivid goods. After completion of the experiments part of the sample (n = 39) was asked to classify the goods as either type A: ‘hedonic’, type B: ‘utilitarian’, type C: ‘both’ or type D: ‘neither’.59 Experiment 2: results Manipulation check The television set was rated as ‘being primarily hedonic’ by 59%, as ‘both’ by 38.5% and as ‘primarily utilitarian’ by 2.5% (t: 6.38; p < .001), Washing machines were rated as ‘primarily utilitarian’ by 97.5% and as ‘primarily hedonic’ by 2.5% (t: 18.5; p < 59

The exact wording was: ‘Some people distinguish between goods. Some goods are pleasant and fun, give pleasure and appeal to your senses, e.g., perfume (type A). Some goods are useful, practical, functional, something that helps you to achieve a goal, e.g., a vacuum cleaner (type B). Could you indicate for each of the following goods whether they are type A, type B, both, or none. Circle.’

Vividness in intertemporal choices between hedonic and utilitarian goods

81

.0001). Though the television is rated slightly less often as hedonic than in experiment 1, the manipulation hedonic versus utilitarian appears to have worked again. As in experiment 1, vividness does not affect ratings. Selling prices Mean selling prices for washing machines (ƒ2234) are higher than for televisions (ƒ1869; t: 5.45; p < .001), indicating higher valuation of the utilitarian good in this experiment. This does not affect the appropriateness of the experimental design, because subjects do not have to make direct comparisons between the two goods but compare them against financial compensation. However, this might provide some explanation why in experiment 1 in the non-vivid immediate condition 91% of the subjects preferred to receive the washing machine sooner. Subjects in this experiment in fact might have valued the washing machine higher than the television set, but felt they should be compensated for both goods equally, since official selling prices for both goods are equal as well. Again, in experiment 2 vividness has no effect on selling prices for television sets (t: .09; n.s.) or washing machines (t: .52; n.s.), thus again, ruling out a potential value effect of vividness on intertemporal preferences. Impatience There is a greater overall tendency to prefer the utilitarian good immediately – and to reject the offered financial compensation - than the hedonic good. For the television set only 14% chose to receive it immediately while 84% were willing to wait four weeks and receive additional financial compensation (n = 79). For the washing machine (n = 53) 40% wanted the good immediately and 60% were willing to wait (t: 3.87; p < .01). Since washing machines are on average valued higher than television sets, offered interest rates for washing machines are lower than for television sets leading to less patient behavior. Further, recall that in the control question for experiment 1 some subjects perceived the costs of delaying a washing machine to be higher than for the delay of a television set (see above). This also favors greater preference for immediate receipt of the former. More interesting is the effect of vividness on preference for immediate receipt. Logistic regression analyses were carried out for hedonic and utilitarian goods separately. Dependent variable in both analyses was 'preference for immediate receipt', independent variables were 'selling price' of the respective good and 'vividness' (dummy: yes or no). For the television set neither vividness (Wald: .62; n.s.) nor selling price (Wald: 2.3; n.s.) influences preference for immediate receipt (χ2 of the model: 1.3; n.s; Nagelkerke Pseudo R2: .067). Hence, contrary to experiment 1, vividness did not affect intertemporal preferences for hedonic goods when financial compensation charges are involved. For the washing machine, vividness (Wald: 5.3; p < .05) raised preference for immediate receipt, but selling prices did not (Wald: 2.27; n.s.; χ2 of the model: 8.45; p < .05; Nagelkerke Pseudo R2: .20). Thus, vividness raised the immediate attractiveness of utilitarian goods when financial compensation charges are involved. Consequently, using a financial measure for impatience reveals a stronger effect of vividness for utilitarian than for hedonic goods. In experiment 1 this effect was the reverse, as hypothesized (see Table 4.1 above). Table 4.4 summarizes preferences for experiment 2.

82

Chapter 4

Table 4.4 Intertemporal preferences for hedonic and utilitarian goods in experiment 2

Television set Washing machine

Preference for the immediate receipt Vivid Non-vivid Total 17% 11% 14% (n=79) 61% 27%* 40% (n=53)

* Differences in preferences for immediate receipt between vivid and non-vivid condition significant in a logistic regression analysis at p < .05. Discussion In experiment 2, vividness did not affect preference for immediately receiving the hedonic good. Subjects considered ƒ100 as enough compensation for a four weeks delay, irrespective of their valuation of the good and irrespective of whether the good was presented vividly or not. For a utilitarian good the appropriateness of this financial compensation depended on vividness but also not on the valuation of the good. For washing machines presented vividly ƒ100 were much less likely to be seen as an appropriate compensation than for washing machines presented non-vividly. This holds even though there are no effects of vividness on the valuation of the good, again ruling out the value effect suggested by Nisbett and Ross. Thus, vividness affects impatience for utilitarian goods more strongly than for hedonic goods. Apparently offering financial compensation lowered the deprivation effect for the hedonic good and increased the immediate attractiveness of the utilitarian good. Note, that in experiment 2 for the utilitarian good vividness had an even stronger influence on preference than valuation of the good. Experiment 3: design

We conducted another experiment investigating the effect on vividness on hedonic and utilitarian goods. The experiment was carried out in the setting of experiment 1, therefore subjects and method are identical and will not be reported here in detail again. This time subjects stated exact appropriate compensation charges for delaying hedonic and utilitarian goods, presented vividly and non-vividly. That is, as in experiment 2 they had to determine adequate financial compensation for delaying these goods, but they had to do so even more carefully, because contrary to experiment 2 the compensation was not fixed but had to be determined by the subjects. The experiment was carried out because fixing the offered compensation may suppress changes in preference, e.g., because the offered compensation is too high to overrule the rise in attractiveness due to a vivid presentation. Again, the goods were television sets and washing machines with the same prices as above. Both goods were presented to subjects in sequence both either vividly or non-vividly. Vividness was manipulated as in experiments 1 and 2 and again selling prices were elicited to control for possible effects of value differences. Subjects had to delay goods either 5 weeks or 9 weeks. That is, if they had to delay a television first for 5 weeks, in the second choice problem they had to delay a washing machine for 9 weeks. Order of presentation was varied. Part of the sample first received a question concerning the delay of washing machines and part of the sample first received a question concerning the delay of television sets. Subjects specified and wrote down an amount minimally sufficient to compensate them for the respective delay.

Vividness in intertemporal choices between hedonic and utilitarian goods

83

Experiment 3: results Manipulation check and selling prices Because experiment 3 was carried out in the same setting as experiment 1 the ratings of the two goods are identical to the ones reported above and the manipulation hedonic versus utilitarian appeared to have worked. As to selling prices, mean selling prices were higher for washing machines than for television sets (ƒ2129 vs. ƒ2026; t: 2.14; p < .05), thus replicating the results of experiments 2. This is striking, because the subject pool is identical to experiment 1 where selling prices were equal for both goods. It appears that in fact presenting goods simultaneously creates a biases towards equating selling prices, partly explaining why 91% preferred to receive the utilitarian good sooner in the non-vivid immediate condition in experiment 1. In experiment 3, vividness does not alter selling prices for either good (televisions: t: .59; n.s.; washing machines: t: .30; n.s.), replicating results of experiments 1 and 2. Impatience Compensation charges differ tremendously between subjects, making interpretation of the data difficult and suggesting that keeping the offered monetary compensation fixed is a more useful design. Several subjects charged unrealistically high amounts (more than ƒ1500 for a five or nine weeks delay of a good with a list price of ƒ2400) and were excluded from further analysis. For hedonic goods results are very noisy. Neither time delay nor vividness influences compensation charges. In the short delay condition mean charge was ƒ281 for a vivid television and ƒ295 for a non-vivid television, in the long delay condition mean charge for vivid television mean charge was ƒ273 and for a non-vivid television mean charge was ƒ224. T-values range from .01 to .5 and never approached significance. Hence, again vividness did not influence impatience for hedonic goods when measured by monetary compensation. For utilitarian goods likewise vividness did not affect compensation charges. Subjects mean charge for a five weeks delay was ƒ231 when the washing machine was presented vividly and ƒ194 when presented non-vividly. For the nine weeks delay the mean charge for a vivid washing machine was ƒ391 when presented vividly and ƒ527 when presented non-vividly. Though numerically these results appear to be impressive, due to the huge standard errors they are again not significant (short delay: t: .38; n.s.; long delay: t: .57; n.s.). These results do not change when the whole sample is used, i.e., when including the subjects charging unrealistically high amounts, or when discount rates are calculated instead of using absolute compensation charges. Table 4.5 summarizes compensation charges for experiment 3. Table 4.5 Compensation charges for hedonic and utilitarian goods (Standard error between brackets) TV vivid TV non-vivid WM vivid WM non-vivid

Mean compensation 5 weeks delay 281 (112) 295 (105) 231 (101) 194 (37)

Mean compensation 9 weeks delay 273 (25) 224 (39) 381 (101) 527 (161)

84

Chapter 4

Discussion It is obvious that vividness did not raise compensation charges for the hedonic good, thus replicating the result of experiment 2 where financial compensation was also involved. For utilitarian goods there appears to be no effect of vividness as well, however, the picture is very inconclusive due to the huge standard errors. The most interesting result is that in this experiment, as in experiment 2 selling prices were higher for utilitarian goods. This is striking because subjects and description of goods were identical to experiment 1 were subjects on average stated identical selling prices for both goods. This suggests that selling prices in experiment 1 were slightly biased downwards for washing machines. In experiment 1 average selling price for washing machines was ƒ2032, whereas in experiment 3 where goods were presented in sequence average selling price was ƒ2129. Selling prices for television sets on the contrary were approximately equal (ƒ2010 in experiment 1 vs. ƒ2026 in experiment 3). Presenting both goods simultaneously as in experiment 1 appears to lead to more equal selling prices of goods. This downward bias in selling prices for the utilitarian good might explain why in experiment 1, 91% in the non-vivid immediate condition preferred to receive the washing machine sooner. 4.4. Overall conclusion and discussion

An effect of vividness on intertemporal preferences has been suggested several times, but these suggestions are contradictory. Above, three experiments are presented where subjects have to express intertemporal preferences for hedonic and utilitarian goods presented both vividly and non-vividly. Intertemporal preferences could be expressed either in terms of monetary compensation or whether they prefer to receive hedonic or utilitarian goods sooner. In choices where subjects stated whether they would like to receive rewards sooner or later, vividness strongly raises preference to receive hedonic goods sooner. When monetary discount rates are involved, we did not observe effects of vividness for hedonic goods and higher impatience for utilitarian goods presented vividly. These results support the interpretation put forth in chapter 3 that in choices which good to receive sooner, the focus lies more on the hedonic features of goods whereas when financial compensation is involved, utilitarian aspects loom larger. Hence, vividness has a larger impact on hedonic goods when the former measure is used and a larger impact on utilitarian goods when the latter measure is used. These results further rule out several of the suggestions made in the literature about the effect of vividness. First, there is an effect of vividness on intertemporal preferences, running counter to Fiske and Taylor’s statement of a generally weak effect of vividness. Second, there is no effect of vividness on value, hence the effect of vividness is not due to a higher valuation of goods presented vividly, ruling out Nisbett and Ross' reasoning. Third, this effect of vividness is not due to a shrinkage in the perceived temporal distance as suggested by Becker and Mulligan, because vividness raises impatience for utilitarian goods when monetary compensation charges are involved. Since Becker and Mulligan predict less discounting for goods that can be visualized better, vividness according to them should lead to less impatience. In addition, their approach is unable to account for the different effect of vividness when using different elicitation measures for impatience. An effect of vividness on impatience in choices between consumer goods further extends Loewenstein's (1996) reasoning for the effect of vividness when visceral factors are involved to hedonic and utilitarian goods.

Vividness in intertemporal choices between hedonic and utilitarian goods

85

However, the arguments Loewenstein provides cannot account for an effect of the way impatience is measured. As to the compatibility of the results with the research of Mischel and his co-workers several statements can be made. First, vividness raising impatience for hedonic goods more strongly than for utilitarian goods in choices which good to receive sooner is in accordance with the arousing/abstract distinction. Since for hedonic goods in a vivid presentation the focus is on arousing hedonic features, impatience for them rises more strongly. Also a stronger effect of vividness on hedonic goods in choices than when determining monetary compensation is compatible with the arousing/abstract distinction. However, the increasing impatience for the utilitarian good when subjects have to determine monetary compensation in experiment 2 runs counter his theory. Since in the vivid presentation of the utilitarian good the focus lies on the utilitarian and abstract features of the good, according to Mischel one would expect less impatience for vivid goods in this experiment. The different effect of vividness for hedonic and utilitarian goods depending on measurement also extends the 'pure deprivation effect' of vividness caused by better adaptation to the possession of an object and by making goods appear to be more hedonic or utilitarian. The only account compatible with the data is framing theory. It correctly predicts a stronger influence of vividness on hedonic goods in choices which good to receive sooner and a stronger influence on impatience for utilitarian goods when impatience is measured by determining adequate financial compensation. It also predicts vividness to increase impatience when monetary compensation is involved as observed in experiment 2. Intertemporal utility, as a consequence of the results of Chapters 3 and 4, seems to be more complex than it has been treated up to now in the literature. First, there is the immediate valuation of goods as measured by selling price. People here determine the utility options provide to them. This appears not to be affected by vividness. Second, there is an immediate preference for goods, which might differ from the valuation of these goods. Here, the distinction between hedonic and utilitarian goods is highly relevant, since, as reported in Chapter 3, hedonic goods are sometimes preferred sooner despite the fact that the utilitarian good was valued higher. For this immediate preference vividness is important, because vividness raises the attractiveness of hedonic goods over utilitarian goods irrespective of differences in valuation. Finally, there is intertemporal utility as measured by monetary trade-offs that normatively should be equivalent to but empirically is distinct from immediate preference.60 Here, discounting is much less sensitive to the difference between hedonic and utilitarian goods and sensitive to vividness only for utilitarian goods. We suggest that this is due to a framing effect. Forcing subjects to think in terms of monetary compensation makes utilitarian considerations such as alternative usage of the money more important and lowers the prominence of hedonic characteristics. More research with respect to this phenomenon is clearly warranted. Besides considerations of different forms of utility, another interesting question is how the results above relate to the experiments of Mischel et al. (1992). Metcalfe and 60

In addition, determining financial compensation for waiting appears to be much less reliable. In another experiment, subjects stated selling prices for washing machines and television sets presented both vividly and non-vividly as well as compensation charges for delaying these goods. After being exposed to several other decision scenarios the questions were repeated. While correlations between selling prices for both goods were reasonably high (between .8 and .95), regardless of whether presented vividly or non-vividly, correlations were strikingly low for compensation charges and not even significant for non-vivid television sets.

86

Chapter 4

Mischel (1999) argued that in intertemporal choice the distinction between ‘hot’ and ‘cold’ aspects of a decision situation or presentation is relevant. Hot decisions situations, e.g., presence of a reward, raise the need for immediate gratification, while cold aspects of a situation, e.g., abstract presentations of the reward, lower it. The hedonic/utilitarian distinction seems to match but to be more precise than Metcalfe and Mischel’s hot/cold distinction, since goods can be and have been rated consistently by a variety of subjects and for a variety of goods. However, the latter distinction appears to be more generally applicable. It is striking, given the similarity of the hot/cold and hedonic/utilitarian distinctions, that contrary to Mischel et al. (1992) in experiment 2 vividness raised impatience for utilitarian goods. Since the focus of the presentation lies on the abstract/utilitarian features of the reward one would have expected the contrary. The question as to what makes a presentation vivid has been raised only briefly in this chapter. However, it is striking that in two experiments presented here vividness, i.e., adding a sentence that stresses some hedonic or utilitarian features of the good, raised the need for immediate gratification, while in Mischel’s original experiments providing children with photos of the reward, and thereby making the reward more vivid, lowered their impatience. Mischel et al. (1992) argue that photos actually focus on the abstract qualities of the reward, even though the rewards chosen, snacks or toys, would be classified as hedonic. Using an 'extremely realistic color photo' (Mischel, personal communication) appears to be more arousing than stressing some hedonic features of a good in a description. The different effect of vividness in the experiments reported here might be due to different perceptions of adults and children as to arousing or abstract presentations. While children might consider realistic photos as abstract, adults might consider more or less extensive descriptions of a good already as arousing. Future research might investigate whether the different impact of vividness is due to differences between photos and text or due to different perceptions between adults and children.

Chapter 5 Causes and consequences of patience

Summary∗

People differ with respect to patience. The present chapter seeks to explain such individual differences. Patience, as an ability to counterbalance the natural tendency to overweight short-term benefits, by assumption is taught during socialization mainly in the parental household. Consequently, patience will vary systematically depending on the parental incentives, restrictions and ability to teach it. For example, parents who have relatively more to lose from misbehavior of their children face higher incentives to socialize them properly. Parents with stronger time pressure, be it due to the number of children present in a household or to work requirements, face more restrictions for socializing their children. Strong evidence is advanced for higher parental status and larger parental wealth to result in more patient children, but there is little evidence for effects of family composition on patience, suggesting that parental restrictions are of minor importance in teaching children to delay gratification. Further, consequences of patience are demonstrated in a variety of settings: more patient respondents undertake more effort in studying, behave in a more healthy way, and are less likely to report deviant behavior. In addition, this chapter compares two different measures of patience, viz. a scale measuring preference to delay various amounts of money and a psychological scale measuring considerations of long-term consequences. The latter measure appears to be much better predictable and to be much more predictive. It is concluded that calculating discount rates from compensation charges for the delay of various hypothetical amounts of money measures other traits than patience. The consideration of future consequences scale appears to be a much more reliable, though possibly not exhaustive, measure for patience.

∗

This chapter benefited from discussions with and comments of Marcel van Assen, Peer Scheepers, Inge Sieben, Christian Steglich and especially Jeroen Weesie. None of them is responsible for any remaining errors. An earlier version of this chapter jointly with Jeroen Weesie has been presented at the ICS Forum Day in Fall 1997.

88

Chapter 5

5.1. Introduction

In Chapter 1 it was argued that there are profound individual differences with respect to patience, i.e., time discounting, and that such differences have considerable impact in a variety of settings. This chapter first briefly reviews treatment of patience in economics, sociology, and psychology. Second, it investigates the origins of individual differences in people's tolerance for delay of gratification. It is argued that since patience is of advantage in numerous situations, parents will try to socialize their children accordingly and attempt to ‘produce’ the necessary preferences within their children. Because incentives, restrictions, and ability to socialize children ‘properly’ differ between parents, children differ with respect to patience as well (see Lindenberg 1993). In the following some systematic differences with respect to socialization and their consequences for people's tolerance for delay of gratification are specified and tested. Third, it is investigated whether patient people differ from myopic subjects in a variety of settings, e.g., health behavior or engagement at school or university. Finally, two different measures of patience are discussed, a psychological scale measuring general concern for the future and a scale measuring monetary discount rates. By doing so, we continue the discussion of Chapters 3 and 4 of whether discount rates calculated from the specification of monetary compensation charges for delay do in fact measure myopia or are instead relatively unrelated to (im)patience. To investigate where individual differences in tolerance for delay of gratification stem from, data are used from experiments containing a sample of students at the University of Groningen from various disciplines. The data contain two alternative measures for the ability to delay gratification, i.e., a more general psychological measure and a measure based on the delay of monetary amounts, detailed information on the parental home of the respondent as well as on the techniques used for socialization. Thus, the data provide rather specific knowledge about individuals. Finally, the data contain information on subjects' behavior in several areas where differences in patience should lead to differences in behavior. The remainder of this chapter is structured as follows. In section 2, a short overview of the treatment of delay of gratification in various disciplines is given, theoretical considerations as to where differences in patience stem from are presented and research hypotheses are elaborated. Section 3 explains data and methods, section 4 presents results for causes of patience for the more general measure, section 5 presents results for consequences of different degrees of patience for the more general measure. Section 6 presents results for causes and consequences of patience as measured by delaying various hypothetical amount of money, and section 7 summarizes and draws conclusions. 5.2. Review and theory

Economics, sociology and psychology all assume that people prefer to receive benefits now and that waiting for them is aversive. In economics, differences in patience reflect themselves in different trade-offs between current and future consumption. These tradeoffs between current and future consumption are expressed by a person's time preference. A person with a high time preference, i.e., an impatient or even myopic person, will prefer a smaller but immediate reward to a larger but later reward. Thus, he strongly discounts future outcomes. The degree of time preference is measured by a discount rate specifying the devaluation of the delayed outcome per time period (see chapter 1 for an extensive discussion). In sociology and psychology, foregoing immediate benefits for later and

Causes and consequences of patience

89

larger ones is covered by the terms 'delay of gratification’ or ‘self-control’.61 Sociology and psychology explicitly regard the delay of gratification as problematic (see Ainslie 1992 for extensive support of this view). Individuals may fall prey to short-term temptations, i.e., they choose immediate small benefits, and later regret their choices. In classical economics, the assumption of rationality excludes such regretting one’s informed choices, but recently, behavioral economists focussing on behavior qualified by standard economic theory as ‘anomalous’ have also begun to investigate sub-optimal choices due to myopia (Loewenstein & Elster 1992).62 Although patience is potentially important for sociology and economics, both disciplines largely neglected the question where individual differences come from. Economists have mostly avoided to study individual determinants of time preference by making time preference part of individual tastes to which economists generally remain agnostic.63 If more precise statements are necessary for modeling purposes, the common assumption is made that all commodities are freely exchangeable, and hence trade-offs between current and future consumption should be made at the market interest rate, no matter what commodities are involved, (e.g., Loewenstein & Thaler 1989). Sociology treats delay of gratification as part of the general acquisition of norms and role expectations learned via socialization (Parsons 1951). The socialization process creates an internal sanctioning system via the internalization of norms and is mainly mediated by parents in early childhood. Differences in socialization then may lead to differences in the internalization of norms and role expectations and, in case of undersocialization, to deviant behavior. It is therefore not surprising that research on deviant behavior has taken the question which parents are successful in establishing 'appropriate' behavior, e.g., selfcontrol, most seriously (e.g., Gottfredson & Hirschi 1990). However, few attempts have been made to generalize these findings to other areas of research or to preference formation in general. Furthermore, research on deviant behavior provides little or no theory with respect to how different socialization techniques affect internalization of norms. In the following it is assumed that people are naturally shortsighted and have to be socialized in order to take long-term consequences of their actions into account.64 In line with classical sociology it is argued that children's learning is mainly mediated by parents. Such learning within childhood mainly takes place via reinforcement and modeling (see e.g., Bandura & Walters 1963). Reinforcement includes the reaction of parents towards behavior of their children either by punishment or by rewards. Modeling includes instruction of children by the parents and imitation of parental behavior by the children. 61

Self-control is a broader concept than time preference or delay of gratification. Self-control includes the delay of gratification (patience or time preference), restriction of aggressive behavior, tolerance of frustration, and the orientation towards risk. 62 In this stream of research also self-command is discussed (Schelling 1984, Lindenberg 1993). Selfcommand investigates the use of efficient means, e.g., commitments, in order not to fall prey to temptations. For example, a person with a low degree of self-control may know this and, because of high self-command, successfully try to prevent himself from actions he would regret later. He may develop rules such as ‘never shop when hungry’, ‘don’t shop with a credit card’, or ’always wait one day to make expensive purchases’. 63 Recently Gary Becker has argued in a series of articles (summarized in Becker 1996 that parents being altruistic towards their children try to influence their children's preference sets ('tastes') such that these tastes in turn benefit their children. However, no link is made to what makes parents successful in such preference manipulations. 64 Though most people learn to some extent to take long-term consequences into account, situations where hedonic consumption experience is important are especially prone to impulsive behavior (see chapters 3 and 4).

90

Chapter 5

Hence, for modeling parental ability with respect to the trait is also important, while reinforcement should be unaffected by different degrees of parental ability. Though there are also cultural effects with respect to reinforcement and modeling, e.g., by schools and religious communities, there is general agreement that parental behavior exerts the strongest influence on young children’s behavior (Parsons 1951, Bandura & Walters 1963) and therefore here the focus is on parental characteristics. How do parents differ with respect to modeling and reinforcements? In this chapter it is argued that, first, parents differ with respect to patience themselves. Parents who are better able to delay gratification serve as a better model than parents who frequently behave in a myopic fashion. Second, incentives to reinforce and instruct children differ between parents, depending, e.g., on the degree to which they are held responsible for their children’s behavior (Lindenberg 1986, Coleman 1990, Chapter 11). Parents, for example, may gain or lose status depending on whether their children behave appropriately or not. Parents having more to lose from children's inappropriate behavior thus have higher incentives to reinforce them. Finally, parents face different restrictions due to family composition. The amount of parental time available per child, for example, varies with the number of children in a household and with professional obligations of the parents, thus affecting opportunities to teach children complex abilities such as delay of gratification. In the following sections, first parental characteristics and then effects of family composition on patience will be investigated. Since it is argued that parental characteristics and family composition affect patience via socialization techniques, in another subsection these techniques will be investigated more directly by looking at the family environment in which the respondents grew up and it will be demonstrated how socialization techniques affect patience. 5.2.1. Effects of parental characteristics on patience

Following higher education is one of the most prominent examples of investment behavior where immediate benefits are forgone for the sake of larger future returns. Parents with high educational achievements should be more able to delay gratification, because of self-selection and because of the training higher education provides. Both effects make these parents provide a better model for the value of delayed outcomes and make it more likely that they value patience and place more value on its being learned by their children. Hence, higher parental education raises patience of the respondents. Consistent with this reasoning, higher parental education affects several potential consequences of (im)patience. For example, it raises children’s education (Downey 1995) and lowers their risk to become delinquent (Gottfredson & Hirschi 1994). Thus: H1: Patience increases with the level of parental education.

The parental profession likewise affects parental socialization efforts in several ways. Parents who regularly have to delay gratification at work, are trained to value this ability higher, and consequently teach it to their children more strongly. In addition, these parents also serve as better models. We argue that especially autonomy and responsibility at work serve as such training factors and consequently influence patience of children. Increasing autonomy at work creates more opportunities to delay unpleasant tasks and makes it more necessary to delay gratification. Increasing responsibility increases potential costs of myopic behavior. Hence, patience is expected to increase with increasing autonomy and responsibility. These effects will be enhanced by self-selection.

Causes and consequences of patience

91

Initially more patient people reach higher educational levels and will be more autonomous and responsible at work: H2: Patience increases with increasing parental autonomy and responsibility at work.

A special case where the above reasoning applies is self-employment. To ensure the long-run survival of the business, self-employed will have to pay greater attention to the future consequences of their immediate actions than employed people. Most often, for example, it is not possible for the owner of a business to use the firms assets to peruse immediate and personal goals, since this might endanger the long-run survival of the business. Such a restriction in the pursuit of immediate and personal goals provides a continuous training to delay gratification. This effect again will be strengthened by selfselection: patience may lead to self-employment, since the long-term advantages of selfemployment (more autonomy, higher income in the long run) then are more likely to outweigh the short-term costs (more work, possibly less initial income). Both, the training and the self-selection argument suggest that self-employed parents value patience higher, serve as a better model, and will take more efforts to teach their children to delay gratification. In addition, since more employees create greater responsibility, these effects should be stronger, the more employees there are in a parental business: H3: People whose parents are self-employed are more patient. H4: Patience increases with the number of employees in the parental business.

Most religions promote behavior that is directed towards reaching long-term goals, e.g., eternal life, and stimulate and support behavior to achieve these goals, for example, by establishing religious laws. Religious context therefore serves a prominent socializing factor, at least with respect to self-control. Thus, people socialized by religious parents are expected to be more patient. In line with Weber’s classical argument on the effect of religious beliefs on economic organization (Weber 1905), we expect a better ability to delay gratification for people socialized by more strict religions, e.g., protestant versus catholic. In the catholic doctrine (minor) deviations are not allowed but to some degree absolved in case of repentance, while the protestant doctrine takes a more restrictive position on them. This suggests that Protestants are more restrictive against their own impulsivity. These considerations result in the following two hypotheses: H5: Children socialized by religious parents are more patient. H6: Patience of children rises with stricter parental denomination.

With respect to the incentives to teach delay of gratification, Coleman (1990, Chapter 11) reasoned that socialization efforts are higher for families of a high status, as these families have relatively more to lose from inappropriate behavior of their children. Such higher socialization efforts should result in higher internalization of norms and better ability to delay gratification:

92

Chapter 5

H7: Patience increases with increasing parental status and wealth.65

A related argument pertains to social cohesion in a neighborhood. Higher cohesion raises negative consequences for parents from inappropriate behavior of their children, because they will be held more responsible. Hence, stronger cohesion raises incentives to teach children to delay gratification: H8: Increasing cohesion in the neighborhood raises patience.

As to restrictions, mothers working during early childhood of the respondents are expected to have less time to supervise and socialize the children. Junger (1994) actually reports that having a working mother in early childhood raises the probability of deviant behavior. Since in research on deviant behavior this effect in general disappears after controlling for parental supervision, this is mainly due to time restrictions. H9: Patience decreases, if the mother was working during the youth of the child.66 5.2.2. Effects of family composition on patience

Differences in socialization and consequently in patience can be expected between singleparent and two-parent families (see already Mischel 1961). Being raised in a single-parent household potentially affects patience for several reasons. First, again there is selfselection. Higher parental impatience raises the risk of separation, because it is more attractive to jeopardize the relation for short-term advantages (see Snijders & Weesie 1998 for empirical evidence). In this case, parents serve as a bad model with respect to patience. Second, after a separation, less time and effort can be put into socialization of the children by the remaining parent, if single parents face the same professional demands as twoparent families. Time and effort put into socialization of children should thus be lower the sooner the parents divorce. Actually, Hirschi (1994) reports that children in single-parent households report lower monitoring levels than children from two-parent families. Research on school achievement (Astone & McLanahan (1991) likewise reports lower levels of parental supervision and monitoring in single-parent households.67 According to Junger (1994) children of divorced parents run a higher risk of having an accident, which is taken to be an indicator for a neglect to take long-term risks of immediate pleasures into account. This effect disappears after controlling for parental supervision. Third, parents absent from their family have fewer incentives to teach and supervise their children, because they will receive only a small fraction of the benefits of children's proper behavior. Again, this effect should be stronger the sooner the parents divorced. Finally, not only absolute levels of supervision are lower for single-parent households but also the 65

Note, that the argument presented here pertains to parental incentives while the argument concerning the effect of parental education, which is obviously correlated with status and wealth, referred to parental modelling. It is therefore useful to test both hypotheses separately. 66 This effect is counterbalanced by the greater likelihood for higher educated mothers to re-enter the labor market after child birth. As argued above, these mothers have a higher likelihood of teaching their children patience. In the analysis below, we therefore add mother's education as another control variable when investigating this hypothesis. 67 The influence of divorced parents on educational achievements while pronounced in the short run, is not well established with respect to long-run consequences, especially after controlling for the economic consequences following separation (see, e.g., Ryan & Adams 1995).

Causes and consequences of patience

93

relationship with the remaining parent might be closer in these families. This reduces his or her ability to impose sanctions in case of children’s improper behavior. In fact, delinquents are afraid of the parental sanctions more often in two-parent families than in single-parent families (Hirschi 1994). To conclude, there are several reasons to expect stronger impatience for children with divorced parents and among these stronger impatience the sooner the parents divorced.

H10: Children with divorced parents are less patient.

H11: Patience decreases the younger the child's age at which parents divorced.

The number of siblings should also affect parental socialization. Two competing arguments can be developed. On the one hand, the more siblings are present, the more they have to take care of each other and they more they are restricted by the needs of the other siblings. This raises incentives to control themselves and increases patience. Coleman (1990, 1993) argued further that the more children in the household, the higher are the incentives for the parents to socialize their children with respect to self-control in general. If children are not properly socialized, parents will experience the costs of this behavior more directly. For example, having contact with the children more often, e.g., when sharing a common bedroom, creates severe annoyance in case of undersocialization (see Lindenberg1986, 1993 for further elaboration of this argument). However, this latter argument concerning externalities pertains to space restrictions and not to the number of siblings. Therefore:

H12: Patience increases with decreasing space for children at home.

On the other hand, a larger number of children reduces parental supervision and control per child, at least if parents do not supervise all children at once all the time. Hence, children will be surrounded by other children more often. Despite the strong incentives to monitor each other, it is unlikely that children teach themselves self-control, since complex and motivational skills are learned better from adults (c.f. Downey 1995). Consequently, Downey (1995) argues that the negative influence of the number of siblings on educational achievements is due to limited parental resources such as time, effort or expenses. He explicitly states that resource dilution outweighs the eventual effects of economics of scale. Research on deviant behavior supports this view. Here the number of siblings persistently has a positive influence on deviance (e.g., Hirschi 1994). These arguments suggest that patience decreases with the number of siblings. Therefore:

H13: Patience decreases with increasing number of siblings.

94

Chapter 5

5.2.3. Effects of family environment on patience

The previous sections focussed on relatively ‘hard’ but only indirect measures for the underlying socialization processes. In this section hypotheses on the effect of parental socialization techniques on patience will be formulated. To test these hypotheses, retrospective data will be elicited. Though these measures are by nature ‘softer’ than say, the number of siblings, the more direct testing of processes that is possible with this approach should outweigh the disadvantages of the noise inherent to these variables. Above it was argued that parents differ with respect to patience themselves. Consequently, parents who are perceived to be more patient and who place more value on delay of gratification, serve as better models and can teach their children more efficiently. This reasoning leads to two hypotheses. It might be possible that parents are patient themselves but nevertheless regard this ability as less important and hence, place little value on teaching it to their children. It is also possible that parents are impatient themselves but regret this and thus value this ability especially strong: H14: Patience increases with parental patience. H15: Patience is higher for respondents whose parents value patience more strongly.

It has been argued above that people's tolerance for delay of gratification to a large extent depends on parental effort and ability to teach it to their children. Parental supervision of children's behavior then is a necessary means to teach this ability. Parents who supervise their children more intensely should be better able to direct behavior of their children as intended. Such supervision can occur in two forms. First, there is strictness, i.e., number and enforcement of rules for the children. Second, parental engagement in leisure or school activities of the children also serves as a means to supervise and teach children. In addition, the stimulation of general intellectual abilities should also increase patience. Already Mischel et al. (1992) speculated that an early family environment that stimulates cognitive abilities in general also leads to a higher ability to delay gratification, because the latter is an important part of general cognitive abilities. All three possibilities will be explored in this chapter: H16a: Patience increases with increasing strictness of rules at home. H16b: Patience increases with increasing parental engagement in leisure or school activities. H16c: Patience increases with increasing intellectual stimulation of children. 5.2.4. Control variables

Because socialization continues during the life course, increasing age should improve delay of gratification and we thus should control for age when investigating these hypotheses. Further, we should control for sex, since socialization patterns for boys and girls differ in many respects. It is unclear, however, how these differences affect patience. Intuitively one would associate impatience with males, because males are much more likely to exhibit different forms of deviant behavior such as criminal behavior, excessive alcohol use or disregarding speed limits (Hirschi & Gottfredson 1994). On the other hand,

Causes and consequences of patience

95

in some previous studies gender worked in the other direction: women exhibited higher discount rates than men (Donkers & van Soest 1997; Snijders & Weesie 1998; Cairns & van der Pol 2000). Because of a lack of theoretical arguments pointing in either direction, sex is included only as a control variable. 5.3. Tests of the hypotheses: subjects, procedure, and measurement 5.3.1. Subjects

One-hundred and seventy subjects participated in the study.68 Subjects were students at the university of Groningen from various disciplines recruited via e-mail and by flyers. Mean age was 22 years, 44% were males and 56% were females, average monthly income was ƒ1087 ($500), and the main fields of study were psychology (25%), management science (15%), natural science (9%), sociology (9%), educational science (6.5%), and economics/econometrics (6%). One subject had to be removed from all analyses, because he did not understand several of the decision tasks. 5.3.2. Procedure

Data gathering took place in two experimental series. This was done to link the experiments used for Chapters 2-4 and 7 with the research described in this chapter. Differences with respect to subjects between the two series were small. In the first series 90 subjects had to answer questions in a questionnaire containing several choice problems, a measure of patience (see below), a measure concerning monetary discount rates (see below) and questions concerning the personal background of subjects. In the second series, 80 subjects partly were confronted with choice problems presented on a computer screen described in Chapter 7 and partly with choice problems on paper, a measure of patience, a measure for monetary discount rates, and questions concerning their personal background presented on paper. In the first series, subjects gathered in groups ranging from 2 to 23 persons to participate in the experiment. They were provided with a booklet containing the choice problems presented in Chapters 2-4 and the questionnaire used for this research. Subjects then were told about the context in which the research took place and that there were no wrong or right answers. They were also asked not to look back to check previous answers. A version of such a complete questionnaire is included in Appendix A at the end of this book. The experimenter was present during the whole experiment to answer questions concerning the tasks. Filling in the booklet took between 40 and 75 minutes, most people needed between 50 and 60 minutes. Subjects were paid ƒ10 (approximately $4.50) and had an additional opportunity to win a gift certificate worth between ƒ15 and ƒ25 depending on their choices. In the second series, subjects gathered in groups ranging from 6 to 15. They were assigned to a computer where the program concerning the experiments described in Chapter 7 had been installed. Then they were instructed about the context in which the research took place and that there were no wrong or right answers. In between the choice tasks on the computer were regular breaks where subjects turned to the other decision scenarios and to the questionnaire, both presented on paper. The experimenter was present during the whole experiment in order to answer questions concerning the 68

Analyses presented below did not use all 170 subjects. Five subjects participated in both experimental series (see below), reducing the number of usable measures for patience to 165.

96

Chapter 5

tasks. Completing the whole experiment (on the computer and on paper) took between 50 and 110 minutes, most people needed between 80 and 90 minutes. Subjects were paid ƒ25 for their participation. 5.3.3 Measurement

This section explains how the hypotheses presented above have been operationalized and how they were measured. We begin with an overview of the independent variables used in this study following the order of hypotheses and then turn to the measurement of patience. We are well aware of the fact that eliciting retrospective data concerning, e.g., the family climate during the respondents' youth, is less reliable than eliciting 'hard' parental characteristics such as education. However, since we argued that tolerance for delay of gratification is mediated by the family climate these variables are necessary to corroborate the reasoning put forth in this chapter. a) Independent variables

Measurement of the independent variables is presented in tables referring to 'parental characteristics', 'family composition', and 'family environment'. Some of the independent variables introduced above were measured by several items in order to construct scales for these variables. For them, their respective scale values, measured by Cronbach’s alpha (henceforth α), are reported as well. Generally, items were considered to form a scale when α for the initial scale, i.e., all items, exceeded .5 and could not be raised by more than .05 after the elimination of one other item. In this case the scale was used as an independent variable. If α for the initial scale could be raised by more than .05 after the deletion of a variable, this deletion was carried out and this procedure was reiterated until α increased less than or exactly .05. The remaining scale was used for analysis, if α exceeded .5, otherwise, no scale was used. If no scale could be used, the single variables intended to form a scale were introduced simultaneously into a linear regression analysis. Whether the complete or the reduced scale was used is indicated in the table by an asterisk (*) in the respective column. Tables below specify the variable name, their response categories, and, if relevant, whether the items formed a scale. Some questions were asked for parents separately. This is indicated by 'sex'. A detailed overview of measures used in the questionnaire is presented in Appendix A at the end of this book. In the text below, variables are indexed by a subscript referring to their item number in the questionnaire.

Causes and consequences of patience

97

Table 5.1. Variables, response categories, and scale values for parental characteristics and family composition Variable

Response categories

Scale values (α) Complete Reduced

Parental characteristics Parental education13 (sex) Parental autonomy19 (sex) Parental responsibility20 (sex) Parental profession at respondent's age 1417 (sex) Number of employees18 (sex) Religious affiliation15 (sex) Frequency of church visit16 (sex) Scales: a) Parental status23 (three items)69 b) Cohesion of the neighborhood23 (seven items)70 Mother working during respondents' youth22 (elicited for four ages)

Scale from 1 to 9 (lower education to Ph.D.) Four: 'very little', 'little', 'somewhat', 'much', 'very much' Four: 'self-employed', 'employed by the government', 'employed elsewhere', 'does not apply' Five: from 'none' to 'more than 24 persons' Five: 'no religion', 'catholic', 'Dutch reformed', 're-reformed', 'else' Five: from 'never' to 'once a week or more often' (including 'missing') .19 Five: 'disagree completely', 'disagree', 'do not agree or disagree', 'agree', and 'agree completely'

.47

.45 No scale used .51 No scale used

Three for each age: 'not working', 'parttime', 'full-time'

Family composition Parents divorced11 Dummy: 'yes' or 'no' Age of divorce12 Respondent's age at parents' divorce Space at home (four measures, all elicited for ages 5 and 14): 'Number of people at home'6 'Number of rooms at home'7 'Rooms minus people' 'Type of home'5 Number of siblings3 Addition of brothers and sisters including the ones who died

69

The respective items were: ‘My parents had a good reputation in the neighborhood’, ‘It was a wealthy neighborhood’, and ‘We were wealthier than the average neighborhood’. 70 Variables: ‘Other adults than the parents told children what they were allowed to do’, ‘I felt at home in the neighborhood’, ‘People knew each other’, ‘I visited the neighbors at home’, ‘I felt insecure when I was alone at home’, ‘Many of my friends came from the same neighborhood’, ‘It was a neighborhood with much fluctuation’.

98

Chapter 5

Table 5.2. Variables, response categories, and scale values for family environment Variable Agreement to statements a) Patience of parents23 (sex)71 b) Importance of patience for parents23 72 c) Strictness of child raising: general2 (nine statements)73 Strictness of rules (in comparison with friends during primary school. Five items each) a) 'playing'3274 b) 'playing'32 standardized values c) 'eating'3375 'Attending kindergarten'26 Parental engagement at school a) 'Engagement parents at parental school activities'28 b) 'Visits of parent's evening'29 c) 'Frequency of homework control by parents'31 (elicited twice: for primary and secondary school) How strongly stimulated parents a) children's 'values'38 (seven topics)76 b) children's 'knowledge'37 (five topics)77

71

Response categories

α Scale Complete Reduced

Five: 'disagree completely', 'disagree', 'do not agree or disagree', 'agree', and 'agree completely' .19 Five: ‘almost no’, ‘somewhat less’, ‘just as much as at friends’, ‘somewhat more’, or ‘much more’ rules

.34 No scale used

.44 .67 .62*

.56* .74* .67

.53*

.56

.51* .

.49

Dummy: 'yes' or 'no' Four: 'never', 'regularly', 'almost always', and 'always' Four: 'never', 'regularly', 'almost always', and 'always' Five: 'daily', 'once a week or several times a week', 'once a month or several times a month', 'exceptionally', and 'never' Five: 'very little', 'little', 'somewhat', 'strongly', and 'very strongly'

Statements: 'My father has much self-control' and 'My mother has much self-control'. Statement: ‘My parents find it important that I learn to look ahead’. 73 Statements: 'at home problems were hidden instead of solved', 'I was allowed more than my friends', 'I had more tasks at home than my friends', 'my parents asked what was happening at school', 'I do not want to disappoint my parents', 'my parents knew with whom I hang around', 'my parents knew where I hang around', 'I often talked to my parents about things I found important', and 'I had a strict raising'. 74 Items: ‘making noise’, ‘what you were allowed to play’, ‘where you were allowed to play’, ‘with whom you were allowed to play’, and ‘when you had to be home'. 75 Items: ‘washing hands before eating’, ‘eating with the whole family’, ‘table manners’, ‘eating sufficient fruit and vegetables’, ‘finishing your plate'. 76 Topics: ‘relate to other children’, ‘relate to your family’, ‘relate to strangers’, ‘good manners’, ‘selfdevelopment’, ‘body health care’, and ‘general moral standards’. 77 Topics: general development’, ‘practical abilities’, ‘social abilities’, ‘cultural abilities’, ‘work ethics’. 72

Causes and consequences of patience

99

Table 5.2. Variables, response categories, and scale values for family environment (continued) Variable

Response categories

Stimulation of children (four different leisure activities)78 a) at primary school35

Five: never', 'once a year', 'once a month', and 'once a week'

b) at secondary school36 c) at both ages combined Stimulation of children: general30 Number of children's book present at home

α Scale Complete Reduced .24

.24 No scale used .25 .38 No scale used .56 .71*

Four: ‘less than 10’, ‘about 10 –20’, ‘about 20-50’, ‘more than 50’

b) Control variables

Age2 was measured by years. To additionally test for the reliability of the CFC-scale, we constructed a scale measuring the respondents perception of their self-control in general. The reasoning behind this question is that if our measure for patience (see below) is reliable, then it should also be related to self-control in general, since patience is part of a person's (perceived) self-control. Response categories and scale values for this variable are presented in table 5.3 below. Table 5.3 Response categories and scale values for respondents' self-control Variable Respondents’ self-control25 (nine items)79

Response categories Five: completely disagree, disagree, do not agree or disagree, agree, and agree completely

α Scale Complete Reduced .49* .49

c) Measurement of patience

To measure patience as a general personal disposition, the ‘Consideration of future consequences’ scale (CFC-scale) was used. This scale was developed and validated by psychologists Strathman, Gleicher, Boninger, and Edwards (1994). The scale consists of 12 statements pertaining to the subject's general concern for positive and negative future outcomes and is presented in full detail in Appendix D of this chapter. Subjects express agreement or disagreement to each of the 12 statements on a five-point scale.80 Strathman 78

Leisure activities: ‘going the zoo, sea or forest’, ‘going to the theater or cinema’, ‘how often played your parents games with you’, ‘how often did you have serious talks with your parents’. 79 Items: ‘I still have problems to do unpleasant tasks’, ‘I am somebody who wants to try everything, even if it might be dangerous’, ‘ I have difficulties to get up in the morning’, ‘it is difficult for me to stay away from unhealthy things’, ‘I am very disciplined ‘, ‘I do not worry much about tomorrow’, ‘I can deal well with money’, ‘I should control myself more when I have fun’, and ‘I like to do things other people find strange or dangerous’. 80 Based on four subsequent studies with a total n of 1100, Strathman et al. (1994) provide evidence for consistency of the scale (α ranging between .8 and .86) and temporal stability (two studies were replicated after a two-weeks and five weeks-delay with correlations of .76 and .72 respectively). They also report high

100

Chapter 5

et al. (1994) report predictive value of the scale for subjects' general concern for health as well as for subjects' actual behavior such as number of cigarettes smoked, amount of alcohol consumed, and their likelihood to engage in environmental behavior. This scale potentially serves as a general measure for patience, because it includes positive and negative outcomes and in previous studies had predictive value for a variety of settings where weighing of future outcomes is important. In the present research, α for the ‘CFCscale' was .8, thus resulting in a neat scale and replicating a result of Strathman et al. (1994). This underlines the reliability of the CFC-scale (see also Chapman 1998 for a similar result). Later, patience will also be measured by what economists call 'time preference' linked to discount rates for money. There are theoretical reasons to doubt the validity of such a measure, presented in Chapters 3 and 4 as well as below, but since it is used in most research done by economists on the subject, it will be included. 5.4. Causes of patience: results for the CFC-measure

Hypotheses were mainly tested by using linear regression analysis. In some cases ANOVA was used, this is indicated in Appendix A at the end of this chapter. In all cases independent variables were entered separately resulting in separate regressions for each independent variable. That is, the independent variable under study, irrespective of whether it was a single variable or a scale, and the relevant (see next section) control variables were entered as independent variables to predict scores on the CFC-scale and then this procedure was repeated with another independent variable. On the whole, results for the CFC-scale are consistent with the reasoning put forth in section 5.2, many variables are significant at conventional levels and have the correct sign. A complete overview of variables used for analysis and corresponding t-values is given in Appendix B of this chapter. T-values and significance levels are given in the text when p < .1, otherwise the reader is referred to the Appendix. We now turn to the hypotheses in more detail. Note, that in the results presented below positive t-values indicate higher patience, unless indicated otherwise. Control variables

Sex did not influence patience and was therefore not included as a control variable. Age marginally lowered patience running counter to the argument that socialization continues during the life course (t: -1.86; p < .1), but was included in all analyses presented below. The scale measuring respondents' self-control predicted values on CFC-scale in a linear regression analysis very well (t: 5.7; p < .001). This is especially due to the items ‘I do not worry much about tomorrow’ (t: -4.58; p < .001),81, ‘I am very disciplined’ (t: 4.48; p < .001), and ‘I still have problems to do unpleasant tasks’ (t: -1.89; p < .1). This result again underlines the reliability of the CFC-scale, because, as assumed, it demonstrates that patience as measured by this scale is in fact related to self-control in general.

scores on the CFC-scale to lead to less favorable attitudes towards offshore oil drilling when negative consequences were presented as being likely to occur in the long run. 81 This is not too surprising given this item is similar to some of the items used for the CFC-scale, for instance, items 1 and 3 see Appendix C of this chapter.

Causes and consequences of patience

101

Parental characteristics

Hypothesis 1 stated that higher parental education raises patience of children. There is no effect of father’s highest education on patience, but higher education of the mother raises patience (p < .05). Since mothers generally do most of the work related to child raising and socialization, a stronger effect of mother's education on patience is in line with hypothesis 1. Autonomy of either parent at work had no effect on patience, but there is strong evidence for parental responsibility at work to raise patience (father: p < .01; mother: n.s.; parents combined: p < .01), thus supporting the second part of hypothesis 2. Self-employment of either parent contrary to hypothesis 3 did not affect patience, but in line with hypothesis 4 there is weak evidence for higher number of father’s employees to raise patience (p = .1). Religion did affect patience less strongly than expected. For mothers, results concerning frequency of church visit and strictness of the religion are not significant. Separating fathers into frequent and non-frequent church visitors revealed that frequent church visits of fathers resulted in more patient children (p < .05), but no effect of strictness of religion is found. Hypothesis 6 thus is partly supported while hypothesis 5 is rejected. For hypothesis 7 there is strong evidence. As predicted higher parental status (p < .05) and greater parental wealth (p < .05) lead to more patient children. As to cohesion of the neighborhood, of the 7 items only the number of friends from the neighborhood affected patience. In line with the prediction, having more friends from the neighborhood, indicating stronger cohesion, leads to more patient respondents (p < .05). Given that the other 6 variables measuring cohesion had no effect, there appears to be little evidence for stronger cohesion to affect patience. Hypothesis 8 is therefore rejected. As to the working history of the mother, larger work activity of the mother before birth of the children (p < .1)82 and at age 14 (p < .1) raise patience providing weak evidence against the hypothesis. Because, in addition, for all other analyses mother’s work activity had no effect on patience, hypothesis 9 has to be rejected.83 To sum up the results on parental characteristics, there are clear effects of parental responsibility, status and wealth on patience. Further, patience increases with increasing education of the mother and increasing number of employees for self-employed fathers. As to religion, only frequency of fathers church visit affected patience, the other variables concerning the effect of religion on patience were not significant. No effects were found for cohesion of the neighborhood and self-employment of parents. Finally, there is some evidence for increasing labor market participation of the mother to raise patience, but results are not very strong. Family composition

With respect to family composition, whether parents divorced or not did not affect patience, rejecting hypothesis 10. Contrary to hypothesis 11, patience is higher the sooner the parents divorced (p < .05), however, since only 20 of the 165 respondents had divorced parents, this results may be an artifact. There is little evidence that space available at home affects patience as expected in hypothesis 12: neither number of people present in the household at age 5 and 14 nor the number of rooms present at both ages 82

Higher scores on this variable indicate higher working activity, thus the hypothesis predicted negative tvalues. 83 In these analyses, in addition to respondents' age, also mother’s education was included as a control variable. Since mother's education affected respondents' patience and is likely to affect mother's labor market position and activity, it was included as an additional control variable.

102

Chapter 5

affect patience. Subtracting the number of people from the number of rooms, i.e., testing for the influence of rooms per person, also had no effect. Likewise having an own room either at age 5 or age 14 did not affect patience. The only effect weakly in line with hypothesis 12 was that larger homes raised patience (age 5: p < .1; age 14: p < .1). Hypothesis 12 therefore is rejected. As to hypothesis 13, having more siblings did not affect patience. To sum up, effects of family composition on patience are only weak. Family environment

Recall that the influence of family environment on patience was measured by scales when possible and by single items otherwise. In addition, single items used for the scales were entered simultaneously into a linear regression analysis, irrespective of whether they did form a scale or not. First, results for the scales are reported and then we turn to individual variables in the following subsection. Self-control of either parent did not affect patience, leading to rejection of hypothesis 14. Parents placing greater importance on farsightedness raise patience (p < .05), thus supporting hypothesis 15. With respect to hypothesis 16a, stricter rules with respect to ‘eating’ raise patience (t: p < .05). However, stricter rules with respect to ‘playing’ did not. With respect to parental engagement at school and children's leisure activities, contrary to hypothesis 16b, no effect on patience of visiting a kindergarten, parental control of home work at lower or high school, or stronger parental activities at school were found. However, more frequent parental visiting of 'parents evening' at school (p < .1) and the number of children’s books in the household (p < .1) both raised patience, but these effects are not very strong. As to other parental stimulation, having serious discussions with children at age 5 more often raised their patience (p < .05). Hence, there is some evidence for higher parental engagement to raise patience, but results are mixed. For hypothesis 16c the picture is mixed as well, stimulating ‘values’ raised patience weakly (p < .1), but stimulating ‘knowledge’ of children did not affect patience nor did the scale combining ‘leisure activities at age 5 and 14’. To sum up, the results for the effect of family environment on patience, there is in fact evidence for importance of parents modeling and valuation of self-control on patience of their children as suggested in hypotheses 14 and 15. Further, there is evidence of stricter raising and increasing stimulation to raise patience, both with respect to scales (for ‘eating’ and ‘values’) and single items such as number of children’s books present in the household and having serious discussions with the children more often. Evidence with respect to parental engagement was mixed: more engagement as measured by visits of "parents' evening" at school raised patience but no effects were found for visiting kindergarten, parental control of homework, and parental activities at school. Table 5.4 summarizes results significant at p < .1 for effects of parental characteristics, family composition, and family environment on patience.

Causes and consequences of patience

103

Table 5.4 Significant effects of parental characteristics on subject's patience Effects Parental characteristics Mother's education Parental responsibility at work Number of father's employees when self-employed Father frequent church visitor (dummy) Larger parental wealth Number of friends from the neighborhood Larger labor market participation of the mother at respondents age 14 Family composition Larger homes at age 5 Larger homes at age 14 Family environment Parents placing greater value on farsightedness Stricter rules concerning 'eating' Stimulation of 'values' Higher frequency of visits at parents evening Number of children's books in the household Frequency of serious discussions with the parents

t-values 2.13* 3.06** 1.66! 2.26* 2.5* 2.4* 1.86! 1.76! 1.77! 2.1* 2.39* 1.81! 1.8! 1.92! 2.36*

! p < .1; * p < .05; ** p < .01 Effects of single items

Because scaling properties for most scales were only modest and because we were interested how strongly the single variables contributed to the success or failure of scales, we also used these variables as predictors for the values of the CFC-scale in a linear regression analysis. Hence, the following section reports results for these single variables, irrespectively of whether they formed a scale or not. All variables initially considered to form a scale were entered simultaneously as independent variables. Though the items on ‘child raising’ did not form a scale, several items strongly affected patience. Stricter agreement to ‘I had more tasks at home than friends (p < .01) and ‘I had a strict raising’ (p < .01) raised patience, again supporting the positive influence of strictness of rules on patience. In addition, there is also weak evidence for ‘I do not want to disappoint my parents’ (p = .1) to raise patience. The positive effect of stricter rules concerning the scale for ‘eating’ on patience is mainly due to the items ‘eating with the whole family‘ (p < .05) and ‘finishing your plate’ (p < .05). Surprisingly, there is also weak evidence for stricter rules with respect to ‘washing hands before eating’ (p < .1) and, for the items concerning the scale for ‘playing’, ‘making noise when playing’ (p < .1) to lower patience. As to single items for the scales concerning stimulation of ‘knowledge’ and ‘values’, stimulating work ethics (p < .05), general moral standards (p < .05), general development (p < .01), and social competencies (p < .05) raised patience. Surprisingly, stimulating practical competencies lowered patience (p < .05). Table 5.5 summarizes results significant at p < .1 for single items on patience.

104

Chapter 5

Table 5.5 Significant effects of single items on subject's patience Effects (scales to which the items belong are given between brackets) Having more tasks at home than friends (child raising) 'I don't want to disappoint my parents' (child raising) 'I had a strict raising' (child raising) Stricter rules with respect to 'eating with the whole family' (eating) Stricter rules with respect to 'finishing your plate' (eating) Stricter rules with respect to 'washing hands before eating' (eating) Stricter rules with respect 'making noise when playing' (playing) Stimulating work ethics (values) Stimulating general moral development (values) Stimulating general development (knowledge) Stimulating social competencies (knowledge) Stimulating practical competencies (knowledge)

t-values 3.0** 1.8! 3.1** 2.1* 1.98* -1.77! -1.73! 2.57* 2.9* 2.6* 2.57** -2.2*

! p < .1; * p < .05; ** p < .01 5.5. Consequences of patience

To investigate how socialization affects patience of course is useful only, if the chosen measure, in our case the CFC-scale, is in turn able to predict behavior in some relevant settings. To test whether this is indeed the case, is the aim of the second study. In addition to potential causes of patience, in the questionnaire several potential consequences of myopia, e.g., health behavior, investments in education, propensity to engage in deviant behavior, and life styles, were included as well. In case the CFC-scale had predictive value in some of these settings, this would also support the notion of patience being a general personal trait and being measured, at least to some extent, by the CFC-scale. In this second study, the scores on the CFC-scale were used as independent variables to predict outcomes in the respective domains in a linear regression analysis. Because subjects and procedure were identical to the study investigating causes of patience and were already described in section 5.3, the following section directly turns to the measurement of the dependent variables. In addition, the rationale for choosing the respective domains is briefly laid out in the beginning of each section. Again, variables are indexed referring to their item number in the questionnaire presented in Appendix A at the end of this book. 5.5.1 Measurement of the dependent variables

As in study 1, several variables were considered to form a scale. To decide whether these variables formed a scale or not the same criteria as in study 1 was used. For other questions, e.g., on health behavior (see below), the items were intended to be used as separate dependent variables. Nevertheless, these latter items were checked with respect to their scaling properties and, in case they formed a scale, this scale was used as an additional dependent variable. This happened, for example, for the four items on alcohol consumption (see below).

Causes and consequences of patience

105

a) Health behavior The link between health behavior and patience has been made several times and seems obvious. A healthy life-style often means refraining from immediate pleasures, such as smoking or excessive consumption of alcohol, for the sake of future benefits. More patient and self-controlled people thus should be less likely to engage in such behavior. Empirically this relationship is less well established than the theoretical considerations suggest. Strathman et al. (1994) as well as Chapman (1998) provided modest empirical evidence for an effect of patience, measured by the CFC-scale, on health outcomes, while Fuchs (1982) and Barsky et al. (1997), using different measures, could not demonstrate such a relationship at all. In study 2, health behavior is investigated with respect to health generally, sporting activities, nutrition, smoking, alcohol and drug consumption. These different aspects of health behavior were measured by the following items presented in Table 5.6.

106

Chapter 5

Table 5.6 Measurement and response categories for the variables concerning health behavior

Variable

Response categories

Self-perception of general health Five: 'much worse', 'somewhat worse', 'about the compared to other people of the same same', somewhat better', 'much better' age41 Contact with a doctor during the last Four: 'never', '1-2 times', '3-4 times', '5 times or more' year42 Nutrition (five items): α five items: .18; α four items: .34; no scale used How often breakfast per week43 How often per week junk food as a warm meal44, How often per day do you eat snacks45, such as chips or nuts How often per day sweets46, How often do you eat fruit47 Five: 'never', 'less than once per week', 'once a week', 'every two or three days', 'at least once a day' Alcohol (standardized values): α four items: .7; α three items: .79*84 Frequency: How often do you drink alcohol48 Six: : 'no, never', 'yes, less than once a month', 'yes, a) one to three times a month', 'yes, once a week', 'yes, b) How often are you drunk51 several times per week', 'yes, daily' 'Four: never', 'one or several times per year', 'one or several times per month', 'one or several times per week' Quantity: a) How much alcohol do you consume Four: 'nothing', '1-2 glasses per time', '3-5 glasses per during the work week49 b) How much alcohol do you consume time', '6-10 glasses per time' during the weekend including Friday evening50 Smoking and drug use How much do you smoke52 Four: 'not smoking', '1-10 cigarettes/cigars per day', '11-20 cigarettes/cigars per day', 'more than 20 cigarettes/cigars per day' Age first smoking cigarette/cigar53. How often do you use drugs54 Four: 'never', 'once or several times per week', 'once or several times per month', 'once or several times per year').85 Age first drug use55

Sporting activities (standardized values): α three items: .5* How often do you engage in sporting Five for each age: 'never', 'one or several time per activities (measured at primary school, year', 'one or several times per month', 'one or several secondary school, and now)56 times per week', 'daily'

84

After elimination of 'how often are you drunk'. In the data file these answers were recoded ('never' getting the highest number) to ensure that higher numbers reflect a decreasing drug use.

85

Causes and consequences of patience

107

b) Investments in education

Again the link between investments into educational achievements and patience is obvious. Spending time and effort to achieve educational outcomes usually implies foregoing immediate benefits for later ones. More patient people should be more willing to undertake such investments. It is therefore not surprising that Mischel et al. (1992) report that children who were better able to delay gratification at age four achieved better educational results than their comrades more than a decade later. In this study, investments in education are measured with respect to school and university for the items presented in Table 5.7 below. Table 5.7 Measurement and response categories for the variables concerning education Variable Effort at school How carefully did you do homework at secondary school71 Could final results at high school have been better72 Did you have to repeat classes a) at primary school78 b) at secondary school78 Effort at the university How often do you read literature not required for their study67 How carefully do you prepare courses and exams73 How often do you party before exams74 How often do you follow extra courses75

Response categories Five: 'bad', 'casually', 'average', 'careful', 'very careful'. Four: 'much better', 'somewhat better', 'about what could have been expected', 'I had to work very hard to achieve results'. Dummy: yes or no Dummy: yes or no Five: 'never', 'several times per year', 'several times per month', 'several times per week', 'daily'. Five: 'bad', 'casually', 'average', 'careful', 'very careful'. Four: 'never', 'sometimes', 'often', 'almost always'.

Have you ever applied for a job or Dummy: Yes or no traineeship76 How well did you prepare this application77 'bad', 'casually', 'average', 'careful', 'very careful'

c) Deviant behavior

The link between deviant behavior and patience is much less obvious than the links for health behavior and education with patience. However, the sociologists Gottfredson and Hirschi (1990) explicitly argue that a lack of self-control is responsible for the high correlations of deviant behavior in several otherwise unrelated settings. Self-control in their view includes (among other aspects such as moderate risk taking and refraining from aggressive behavior) tolerance for delay of gratification. Refraining from deviant behavior means foregoing immediate benefits for the sake of long-term benefits. Thus, lower levels of patience in this line of reasoning result in a higher likelihood to engage in deviant behavior. Deviant behavior in the present study was investigated by using a scale combining several forms of deviant behavior and by investigating subjects proneness to become involved in accidents both measured by the variables presented in Table 5.8 below.

108

Chapter 5

Table 5.8 Measurement and response categories for the variables concerning deviant behavior Variable

Response categories

Deviant behavior: α nine items: .51; α 8 items: .59* Five: 'never', 'once', '2-3 times', '3-4 times'. '6 Engagement in nine forms of deviant times or more' behavior6286 Involvement in accidents: α six items: .38 (no scale used) a) Traffic accidents6587 Five: 'never', 'once', '2-3 times', '3-4 times'. '6 b) Other accidents66 times or more'

c) Life styles

Patience potentially affects a variety of life styles. In the present study the focus lies on relationships (partner relationships), several leisure activities and the likelihood to exceed one's bank account. As to relationships, impatient people are expected to have more partners and less stable relationships than patient people (see Snijders & Weesie 1998 for evidence). In a relationship situations repeatedly occur where partners have to rely on each other. If in such situations one partner first trusts the other and the other successively has to decide whether to honor or abuse trust, in game-theoretic terms this situation resembles an iterated trust game. For cooperative behavior in such iterated games the weight actors place on future outcomes is crucial (Snijders 1996). Higher degrees of patience thus make cooperative behavior more likely and therefore raise the stability and duration of relationships. As to leisure activities, some leisure activities, e.g., enjoying art displayed at museums or reading literature, require previous investments to enjoy them (see Becker 1996 for an economic analysis of such 'beneficial addictions'). More patient people therefore should be more likely to engage in these activities. Finally, impatient people should be more prone to impulse buying, especially when hedonic consumption is considered as demonstrated in Chapters 3 and 4, and therefore are more likely to exceed their bank account. Table 5.9 summarizes variables and measurement concerning the respondents' life styles.

86

The respective topics were: 'stayed away from school or courses without valid reasons', 'going by bus without paying', 'driving a car or motorcycle without license', 'driving under influence', 'deliberately crossed a red light or a stop sign', 'left a restaurant without paying', 'went to a concert or cinema without paying', and 'stole something in a shop'. 'Deliberately crossed a red light or stop sign' was excluded in the final scale. 87 The respective accidents were: 'accidents as a pedestrian', 'accidents as a cyclist', and 'accidents with a car or motorcycle'. We controlled for traffic participation, when investigating this topic.

Causes and consequences of patience

109

Table 5.9 Measurement and response categories for the variables concerning life styles Variable Response categories Relationships Age at which subjects first had a relation which lasted at least three months58, Number of such relations59 Do you currently cohabitate57 Dummy: Yes or no Leisure activities: α four items: .18 (no scale used) How often do you67: a) visit museums or theatres Five: 'never', 'several times per year', 'several b) read literary books times per month', 'several times per week', c) read newspapers 'daily'. Impulse buying How often do you exceed their bank account Four: 'never', 'sometimes', 'regularly', 'almost by more than ƒ 500 always'.

5.5.2. Consequences of patience: results a) Health behavior

Higher scores on the CFC-scale neither affected subjects' subjective self-perceived health state nor their frequency of contact with a doctor. The scale concerning nutrition was marginally related to patience (p < .1.), with more patient subjects consuming healthier food. They were especially less likely to have junk food as a warm meal (p < .05). Also, more patient subjects reported less alcohol consumption with respect to quantity of alcohol. This effect holds both for alcohol consumption during the week (p < .05) and during the weekend (p < .05). Consequently, subjects reporting higher values on the CFCscale scored less on the scale for alcohol consumption (p < .05). In addition, more patient subjects report lower frequencies of smoking (p < .01) and drug use (p < .01). Finally, more patient subjects scored much higher on the scale measuring sporting activity (p = .001). To conclude, there is strong evidence for patient subjects to lead a healthier life style in all of the investigated areas. However, their self-reported health was not related to patience. This is somewhat surprising given that such health self-reports are usually taken to be reliable (e.g. Ross & Wu 1995) b) Investments in education

As to investments in school, subjects reporting higher values on the CFC-scale did their homework during high school much better than subjects classified by this scale as impatient (p < .001). In addition, they were more likely to work hard in order to achieve their final results at school (p < .05). However, using scores of the scale in a logistic regression analysis to predict the likelihood of repeating classes at high school revealed no effect. As to engagement at the university, the picture is similarly clear. Subjects classified as patient report to prepare exams better (p < .001), do follow topics that are not required more often (p < .001), and read literature that is not required more often (p < .05). In addition, when applying for a job or traineeship, patient subjects again prepare this application much more careful (p < .05). The only effect not in accordance with the prediction was propensity to party before exams. To sum up, there is strong evidence for more patient subjects to invest more strongly in education.

110

Chapter 5

c) Deviant behavior

Subjects classified as patient, scored much lower on the scale measuring deviant behavior (p < .05). That is, these subjects report to engage in deviant behavior much less often. This effect is most pronounced for 'stayed away from school or courses without valid reasons' (p < .05). There was no evidence for patient subjects to become involved in accidents more often. To sum up, there is evidence for more patient subjects to engage in deviant behavior but not for a lower propensity to become involved in accidents. d) Life styles

As to life styles, the results with respect to relations did not confirm the arguments put forth above. Subjects did not differ with respect to the age at which they had a first relation. Moreover, subjects classified as patient were less likely to cohabitate (p < .05) and had more relations than impatient subjects. The latter difference was marginally significant (p < .1). Patient subjects were, however, more likely to visit museums or theaters (p < .05) and less likely to exceed their bank account (p < .05).88 To sum up, though there is no evidence for patient subjects to have fewer relations, these subjects differ with respect to leisure activities and financial management. Table 5.10 summarizes the significant results with respect to consequences of patience for all four domains. Table 5.10 Significant effects for consequences of patience Effects Scale 'nutrition How often do you eat junk food as a warm meal? Scale 'alcohol consumption How much alcohol do you consume during the week? How much alcohol do you consume during the weekend? How often do you smoke? How often do you use drugs? Scale 'sporting activity' How carefully did you do your homework during secondary school? How hard did you work to achieve final results at school? How carefully do you prepare exams? How often do you follow topics that are not required? How often do you read literature that is not required? How careful did you prepare application for a job or traineeship? Scale 'deviant behavior' How often did you stay away from school without valid reasons? Do you cohabitate? (Logistic regression analysis: Wald) 89 How many relations did you have? How often do you visit museums or theaters?

t-values -1.79! -2.4* -2.54* -2.57* -2.47* -2.64** -3.05** 3.27** 3.75*** 2.07* 4.42*** 4.0*** 2.34* 2.14* -2.35* -2.5* 5.1* 1.66! 2.06*

! p < .1; * p < .05; ** p < .01; *** p ≤ .001

88 89

We controlled for subjects income when doing this analysis. The Wald coefficient is positive because the dummy variable was coded as '1= yes' and '2 = no'.

Causes and consequences of patience

111

5.5.3. Discussion: patience as a general trait?

The results presented above indicate that patience, at least to some extent, can be measured, predicted and used for prediction in several otherwise unrelated areas. Therefore, it appears that patience as measured by the CFC-scale is a general personality trait, a view shared by researchers from several disciplines. However, since empirical results with respect to the stability of different measures for patience are weak, a discussion concerning the stability of patience, and if so, how to measure it best, has evolved in the literature. This discussion will be summarized briefly below. The sociologists Gottfredson and Hirschi (1990) argue that lack of patience and self-control in general is responsible for deviant behavior in a variety of otherwise unrelated settings. They also provided evidence for strong correlations between criminal behavior, accident risk, drug abuse, driving under influence, and aggression (Hirschi & Gottfredson 1994). Laub and Sampson (1994) find that individual differences in deviant behavior, may this be crime, pushing comrades on a playground, using drugs or alcohol etc., remain stable over the life course, suggesting that these differences reflect a general trait.90 The economist Victor Fuchs, when summarizing three decades of research on health economics, attributes the high correlations in schooling and health to patience (Fuchs 1996), again implying that patience is a general trait. The psychologist Walter Mischel who did pioneering work on tolerance for delay of gratification, argued that there are stable personality dispositions as well as situational influences at work (Mischel 1974). Further, Mischel et al. (1992) successfully used waiting time in an experiment conducted at age four as a predictor of school success ten years later, implying that patience is both a general and stable trait. However, if patience was a stable and general trait, then empirical studies comparing subjects' patience for various monetary amounts (discount rates) with their tolerance to delay other outcomes, e.g., health outcomes, should yield similar results. However, such a similarity in results was found only at a very general level. For example, the discount function for temporal delay is similar for several goods. People are more sensitive to time delays occurring in the close future than to time delays occurring in the remote future, resulting in a hyperbolic discount function, no matter whether money, health outcomes, vacation trips, or life-saving programs are involved (Cropper, Aydede & Portney 1994; Chapman & Elstein 1995). Unfortunately, the evidence with respect to whether the same people always act myopically irrespective of commodities is mixed. On the one hand, intrapersonal correlations for discounting only financial gains or only losses are usually high (Chapman & Elstein 1995). In one of the earliest studies investigating the discounting of receipts or payments Benzion, Rapoport and Yagil (1989) reported remarkable individual degrees of consistency within and between scenarios, i.e., for discounting gains and losses (Benzion et al. 1989: 278). Likewise, in a study investigating discount rates for money and lifesaving programs, Cropper, Aydede and Portney (1994) compared individual consistency in discounting monetary outcomes and life-saving programs. Respondents who were most eager to receive money today also preferred life-saving programs that saved few people now instead of more later. While this evidence supports the stability patience, on the other hand, correlations between discount rates for monetary rewards sometimes drop to zero when gains and 90

In addition, Laub and Sampson (1994) explicitly attribute such differences to differences in socialization, a view consistent with the arguments and evidence presented above.

112

Chapter 5

losses, speed-up premiums or delay charges are investigated together. Also, in the literature low correlations in discount rates for different commodities, e.g., money, vacation trips and health outcomes, are reported (Chapman & Elstein 1995, Chapman 1998, Loewenstein personal communication). Moreover, though money is the most general commodity and in the literature monetary discount rates are used almost exclusively, the predictive effect of monetary discount rates for actual health behavior is weak (Fuchs 1982, Barsky et al. 1997, Chapman 1998). Chapman (1998), for example, reports no evidence of discount rates for health on actual health behavior such as appointment keeping or regular visits at a hospital. Further, Chapman (1998) reports no correlations between monetary discount rates and either discount rates for health outcomes or the CFC-scale. However, in her research the CFC-scale moderately correlates with discount rates for health outcomes. Given these mixed results, we replicated studies 1 and 2 with a measure mainly used by economists, viz. preference to delay monetary outcomes.91 5.6. Patience for monetary outcomes (discount rates)

Up to now, in this chapter patience was measured by using a scale investigating the respondents' occupation with the future. To the reader familiar with research on intertemporal choice this approach might appear strange. The overwhelming majority of empirical studies in this area uses monetary outcomes to measure patience or, since this measure is mainly used by economists, time preference. However, there are good reasons to doubt the reliability of this measure. People generally have little experience in stating adequate compensation charges for delaying amounts ranging from $5 to $50000 over periods varying from days to years. Moreover, the results in Chapters 3 and 4 suggest that determining monetary compensation charges does not measure impatience or myopia but instead leads people to think about alternative usage of the foregone amount. This at best results in an underestimation of subjects' impatience when monetary discount rates are used. At worst, it results in the estimation of discount rates that have nothing to do with impatience at all. Calculating discount rates from monetary compensation charges would then not serve a proxy for impatience but for knowledge about current market interest rates and calculating abilities. This might explain why Barsky et al (1997) in an otherwise careful study were unable to relate preferences for sequences of outcomes to consequences of myopia. This suggests that questions about general concern for the future might serve as a more general measure for patience and a much better predictor for behavioral consequences than discount rates calculated from delayed monetary outcomes. To test whether this is indeed the case, studies 1 and 2 were replicated with this common measure for impatience, using a scale constructed from the delay of monetary outcomes. In addition, by using monetary discount rates as well we hope to contribute to the discussion reviewed above. If patience measured by monetary amounts is much less predictable and predictive, this would suggest that the inconclusive results are at least to some extent due to the inadequacy of this measure. 91

Other measures of time preferences are even more problematic. Economists frequently calculate discount rates from purchase decisions of consumers (Hausman 1979; Gately 1980), e.g., whether a consumer buys a cheap air conditioner with higher energy costs or a more expensive one with lower energy costs. The finding that subjects with a low income buy cheaper durable goods can, besides time preference, be due to liquidity constraints: subjects with a low income simply cannot afford to purchase expensive goods. Related concepts in social psychology usually comprise too many items for questionnaires, e.g., Block’s (1961) Q-sort test measuring ego-control and ego-resiliency consists of more than 100 items.

Causes and consequences of patience

113

5.6.1. Measurement of patience by monetary discount rates

In an early experiment on human temporal discounting, Thaler (1981) told subjects to imagine they had won some money in a completely trustworthy state lottery, so that they could be certain to receive it. Subjects then were asked how much they would charge to delay receiving this amount of money x some time t. Time delay and prize money were varied. This scenario has evolved into some sort of a standard measure of discount rates often used in experimental (e.g., Benzion, Rapoport & Yagil 1989) and survey research (e.g., Cropper, Aydede & Portney 1994; Wahlund & Gunnarsson 1996)92. In this study, a measure introduced in the HiN95Exp data set (Bruins & Weesie 1996) was used. HiN95Exp is a set of experiments, e.g., on bargaining and trust games, presented to a nationwide sample of Dutch households. This measure yielded six combinations of amount x and delay t presented in Table 5.11. The full text presented to the subjects is given in Appendix D at the end of this chapter. Table 5.11. Combinations of amounts and time delay Lottery win, now Delayed choice

ƒ1000 Y1, 3 months

ƒ500 Y2, 1 year

ƒ25 Y3, 1 month

ƒ100 Y4, 3 months

ƒ10 Y5, 6 months

ƒ10000 Y6, 2 years

Delay time and amount were varied because previous research has constantly demonstrated that discount rates depend on both variables (Loewenstein & Thaler 1989; Loewenstein & Prelec 1992). Subjects were asked to specify the amount y that would make them indifferent between the options. In the investigation of predictors for this measure, we additionally controlled for the respondents’ income measured as monthly net income. Income and current wealth affect the utility derived from money and therefore may influence the amount of money charged in the experimental setting. The DU-model underlying standard economic analysis makes predictions only for utility levels. Suppose, for example, that for one person the utility function for money is linear, while for the other it is logarithmic, i.e., decreasing at the margin. When asked to determine adequate financial compensation for delaying monetary outcomes, the latter person will charge higher compensation to accept waiting, even if both subjects are identical with respect to patience. It is therefore unclear whether different money charges in our experiments are due to differences in the utility function or differences in patience.

92

Barsky et al. (1997) let respondents choose between sequences of life-time income and calculated discount rates from preference between these sequences. However, they were unable to relate discount rates obtained in this way to either potential consequences of discounting (e.g., health behavior) or potential causes of discounting (e.g., age or income).

114

Chapter 5

5.6.2. Corrections on the data

Some corrections on the data had to be made. Subjects were asked to specify an amount y that would make them indifferent between the immediate specified amount x and the delayed amount y. Because neither from a psychological nor from an economic point of view it is very reasonable to wait for nothing, subjects should require a higher amount to wait. Suppose now a person stated that ƒ50 (y) after three months would make him indifferent to ƒ1000 now (x). This answer can be interpreted in two ways. First, he values future outcomes much more strongly than immediate ones and second, he stated the additional amount of money required to wait. It was assumed that if y was smaller than x, then this was due to misunderstanding and y was added to x. This happened in 15 cases in total. 5.6.3. Causes of discount rates: results93 a) Results for the scaling of the dependent variable

Cronbach’s alpha was .87 for the ‘money-scale’ after standardization of variables and could not be raised by deletion of one item. Thus, the six items resulted in a good scale. b) Comparisons between measures

Correlations between the CFC-scale and the money-scale are low and almost non-existent (.096; n.s.). This replicates a result by Chapman (1998) and indicates that these scales in fact do measure different aspects of patience. Further, in no case do the independent variables, i.e., potential causes of patience, affect both scales in the predicted direction. In three cases there are significant effects on a conventional level (p < .05) for both scales, but with an unpredicted sign for the money-scale (see below). This result strongly supports the notion that the two scales measure different aspects of patience. In addition, contrary to the CFC-scale, the money-scale was not related to the scale measuring subjects' perceived self-control (t: .25; n.s.). In combination with the results in Chapters 3 and 4 these results cast doubt on the usefulness of using hypothetical questions concerning delaying monetary outcomes. We will now look at the effects of socialization on the willingness to delay monetary outcomes in detail. To distinguish results for the CFC-scale and the money-scale, in the following the latter will be headed under the term 'discount rates'. c) Causes of patience for discount rates

On the whole results concerning causes of discount rates are weak: few effects are significant at p < .05, and 50% of them have a sign that runs counter the predicted direction. This result again casts doubt on the usefulness of using hypothetical questions for delaying monetary outcomes as a measure for impatience. As in the section presenting the results for the CFC-scale, t-values are given only when they are significant at p < .1 and the reader is again referred to Appendix A of this chapter for a detailed description of 93

Analyses presented below did not use all 170 subjects. Faulty preparation of the questionnaire for the first experimental series led to the omission of the money scale in some questionnaires, thus reducing the number of money-scales to 135.

Causes and consequences of patience

115

the results. Note, that, contrary to the analyses for the CFC-scale, in this analysis negative t-values indicate higher patience (lower discount rates), unless indicated otherwise. Again, we start with the effect of the control variables on discount rates and again variables are indexed to refer to their number in the questionnaire. Control variables

Sex did not influence discount rates94 and was therefore not included as a control variable. Age did not affect discount rates either, but higher income lowered discount rates (t: -2.3; p < .05). This result is consistent with lower utility derived from money with increasing wealth. Parental characteristics

Neither parental education, status, or wealth affected discount rates. Autonomy, responsibility and number of employees of either parent at work had no effect on discount rates as well. Parental religion and their frequency of church visit did also not affect discount rates. With respect to self-employment, an ANOVA revealed less discounting for fathers employed by the state but not for self-employed (F: 2.43; p < .05). As to cohesion of the neighborhood, having more friends from the neighborhood, being one indicator of cohesion, contrary to expectation and contrary to the CFC-scale, weakly raised discount rates (t: 1.8; p < .1). The other variables measuring cohesion had no effect and hence, similar as to the CFC-scale, there is little evidence for stronger cohesion to affect discount rates. With respect to the working history of the mother, having a mother working at the age of 5 strongly lowered discounting (t: -2.67; p < .01).95 Similar to the CFC-scale results, other aspects of mother’s work activity had no effect discount rates. To sum up, contrary to the results found with the CFC-scale, there are almost no effects of socialization on discount rates. If variables are significant, the effects run counter the hypothesis. Family composition

With respect to family composition a divorce of the parents weakly lowered discount rates (t: -1.88; p < .1), running counter to the hypothesis. The number of siblings of the respondent did not affect discount rates. There is some evidence that space available at home affects discount rates: though number of people present in the household at age 5 and 14 does not affect patience, more rooms present at home lower discount rates (age 5: t: -1.92; p < .1). More space at home, i.e., subtracting the number of people from the number of rooms, also lowers discounting (age 5: t: -1.76; p < .1; age 14: t: -2.07; p < .05). Finally, having an own room either at age 5 or age 14 had no effect. Therefore, effects of family composition on discount rates again are only modest. The strongest evidence being on more space available for children lowering discount rates.

94

In some previous studies males were found to have lower discount rates than females. If this hypothesis is tested, there is weak evidence for higher discount rates for women (t: 1.55; p < .1). Since the text did not relate time preference to sex the null hypothesis (no difference between sexes) is taken not to be rejected. 95 Higher scores on the variable indicate higher working activity.

116

Chapter 5

Family environment

Recall that the influence of family environment on discount rates was measured by single variables, single variables being transformed into scales when possible, and, additionally, by individual variables for the scales. As in the report for the results for the CFC-scale, first results for single variables and the scales are reported before we turn to the individual variables intended to form a scale. Self-control of either parent did not affect discount rates nor did higher valuation of patience by the parents. As to stricter parental supervision, stricter rules with respect to ‘playing’ as hypothesized marginally lowered discounting (t: -1.93; p < .1), while stricter rules with respect to 'eating' did not. The same holds for parental engagement at school. As to children's stimulation, stimulating ‘knowledge’ and ‘values’ of children had no effect on discounting as did the scale on ‘leisure activities at age 5 and 14’. To sum up, contrary to the CFC-scale, there is little evidence for family environment to affect discount rates, the only exception being for stricter rules with respect to playing which, in line with the hypothesis, marginally lowered discounting. We now turn to effects of single items. Effects of single items

None of the items on ‘child raising’ approached significance. This is clearly at variance with the CFC-scale where several items strongly influenced discount rates in line with the hypotheses’. As to ‘parental stimulation’, results for discount rates on these items are inconclusive. Though having serious discussion with the children at age 14 more often (t: -1.7; p < .1) and visiting cinema or theatre at age 5 more often (t: -1.8; p < .1) marginally lowered discount rates, accompanying children at their leisure activities more often at age 14, contrary to the hypothesis, led to more discounting (t: 2.5; p < .05). Parental stimulation on the variables measuring ‘knowledge’ and ‘values’ also had effects not in line with the predictions. Stimulating social competencies (t: 2.3; p < .05) and good manners marginally (t: 1.75; p < .1) raised discount rates, thus running counter the hypotheses. To sum up the results, there are no systematic effects of family environment on discount rates, and results, if any, are inconclusive. Table 5.12 summarizes results for both scales for all hypotheses.

Causes and consequences of patience

117

Table 5.12 Summary of results for causes of patience and discount rates Hypothesis CFC-scale Money-scale H1: Patience increases with the level of parental Father: no effect No effect education. Mother: yes H2: Patience increases with increasing parental Autonomy: no effect No effect autonomy and responsibility at work. Responsibility: yes H3: People whose parents are self-employed are No effect No effect more patient. H4: Patience increases with the number of Yes No effect employees in the parental business. H5: Religious parents have more patient children. Father: Yes No effect H6: Stricter parental denomination raises patience. No effect No effect H7: Patience increases with increasing parental Yes No effect status and wealth. H8: Stronger cohesion in the neighborhood raises Yes (marginally) Marginally lower patience. H9: Patience decreases, if the mother was working Marginally increasing Increasing during the youth of the child. H10:Parental divorce decreases patience. Marginally increasing Yes (marginally) H11: The younger the child's age at which parents Marginally decreasing To few cases divorced the lower patience. (few cases) H12: Patience increases with decreasing space for Yes (marginally) Yes children at home. H13: More siblings lower patience. Yes (marginally) No effect H14: Respondents with patient parents are more No effect No effect patient. H15: Respondents whose parents value patience Yes No effect more strongly are more patient H16: Increasing parental supervision raises Yes Inconsistent patience.

5.6.4. Consequences of discount rates: results

Though the scale values constructed from compensation charges for delaying monetary outcomes could not be predicted by socialization, they might serve as a good predictor for real life behavior. For example, Cairns and Van der Pol (2000) as well as Dronkers and Van Soest (1997) successfully used a similar measure to predict real life behavior. On the contrary, Barsky et al. (1997) as well as Chapman (1998) failed to empirically establish such a relationship. In this section, the same consequences of discount rates as in section 5.7 measured by the same variables were investigated. That is, study 2 was replicated with scores for the money scale as the independent variable. Since the discount rates turned out not only to be difficult to predict but also to be a bad predictor, results are reported here only briefly. The reader is referred to Appendix B to this chapter for a detailed account. As to health behavior, neither general health of the respondent nor frequency of contact with the doctor were predicted by discount rates. The same holds for the variables concerning alcohol consumption, smoking, and sporting activities. As to the items concerning 'nutrition', subjects revealing low monetary discount rates, i.e., the more patient subjects, were marginally more likely to have breakfast regularly (t: -1.83; p < .1) and marginally less likely to eat junk food as a warm meal (t: 1.8; p < .1). In addition,

118

Chapter 5

these subjects marginally report a lower age of first drug use (t: 2.0; p < .1). The latter results runs counter the expectation. As to investments in education, subjects discounting the future less, party before exams marginally more often (t: -1.89; p < .1) and prepare their application for a job or traineeship less careful (t: 2.05; p < .05). Both results run counter the expectation. As to deviant behavior, subjects discounting less report marginally more bicycle accidents after controlling for traffic participation (t: -1.73; p < .1) and marginally less sport accidents (t: 1.72; p < .1). As to life styles, subjects discounting less report having had marginally less partners in the past (t: 1.66; p < .1). To sum up the predictive value of discount rates, effects generally are weak and often in direction not expected. 5.7. Summary and discussion

It was argued that patience is of importance in many situations. Parents altruistic towards their children realize that and thus attempt to produce adequate preference sets within their children and socialize them such that they are able to delay gratification. However, some parents are better suited to teach this ability, have more to lose from improper behavior of their children, or face fewer restrictions to teach their children. Indeed, evidence was found for parental characteristics to affect scores on the 'consideration of future consequences scale' (Strathman et al. 1994), which was taken as a measure for patience. Consistent with the argument that higher status and wealth raise the costs of children’s improper behavior, these variables also raise patience. Consistent with the argument that some parents provide a better model to delay gratification, higher education of mothers and higher parental responsibility at work raised patience as well. Little evidence was found for family composition to affect patience, suggesting that restrictions such as parental time are of lesser importance for teaching delay of gratification. If socialization is the carrier of patience, then the family environment should be of prime importance for the acquisition of this ability. More specifically, stricter supervision of children, stronger stimulation of children and stronger valuation of patience by the parents then should affect patience of children. In fact, there is evidence for stricter raising and stronger stimulation of children to raise patience. In addition, parents who value patience higher raise more patient children. To investigate causes of patience is of interest only when differences in patience in fact lead to differences in behavior. A second study was carried out to find out whether this is indeed the case. In this study the effect of patience on several potential consequences of myopia, e.g., health behavior, investments in education, propensity to get involved in various kinds of accidents etc., was investigated. For health behavior, scores on the CFC-scale correctly predicted alcohol consumption, smoking and drug use, i.e., more patience positively correlated with more healthy behavior. Higher patience increased the likelihood to engage in sporting activities as well as the intensity of these activities. Further, the CFC-scale correctly predicted deviant behavior. More patient subjects were less likely to engage in deviant behavior. Finally, subjects expressing greater concern for the future on the CFC-scale were much more engaged in school and at the university, e.g., following topics that are not required, preparing exams more intensely etc. Another question raised (again) in this chapter is how to measure patience and whether patience is a stable general trait. Besides the CFC-scale also a measure often used by economists, viz. discount rates calculated from compensation charges for the delay of money, was investigated. More specifically, causes and consequences of patience were investigated again but with this other measure. The results for the money-scale suggest

Causes and consequences of patience

119

that is does not serve as a reliable measure for patience. As to causes of patience, few effects are significant and even fewer are in line with the prediction. Also with respect to consequences of patience, few effects are significant at conventional levels and of these many are in a direction other than predicted. This result is in accordance with the results presented in Chapters 3 and 4. Already in these chapters discount rates calculated from monetary compensation charges did not serve as convincing measures for impatience. Patience measured by the psychological scale was much better predictable and served as a much better predictor. An interesting problem following from this research is how the variables investigated here do affect other features of preference formation, e.g., risk aversion or social orientation. Since the arguments on socialization referred to very general processes, similar effects for other important features of preference sets, e.g., risk-aversion, social orientation, can be expected. This hypothesis can be tested with the data gathered for this dissertation by combining the data on socialization effects in this chapter with the data on choices under risk presented in Chapter 7.

120

Chapter 5

Appendix A Overview of causes of patience The table below gives the results for the respective linear regression analyses (unless indicated otherwise) with scores for each scale as the dependent variable. Independent variables were introduced separately plus the control variables relevant for the respective scale. Generally, for the CFC-scale this was age and for the money scale it was income. For some analyses there were other control variables added, this is indicated in the text. As to significance levels, the table gives much more details than is usually done in the social sciences. Because very little research has been done with respect to this topic, the details might provide useful hints for future research. Results of t-values (two-tailed test) for the regression analysis are presented (significance levels are given if p < .2, boldface if p < .1). Table A1: Causes of patience Control variables Sex Age Income 97 Parental characteristics Highest education father Highest education mother Religion mother Religion father Religious parents (dummy) Frequency of church visit father Father frequent church visitor (dummy) Frequency of church visit mother Mother frequent church visitor (dummy) Type of work mother (ANOVA: F-values) Type of work father (ANOVA: F-values) Autonomy at work father Autonomy at work mother Responsibility at work father Responsibility at work mother Responsibility parents combined Self-employment father Self-employment mother

96

CFC-scale t-value Sig.

Money scale t-value Sig.

-.35 -1.86 .23

n.s. .06496 n.s

1.55 -.37 -2.3

.124 n.s. .024

.43 2.13 .4 .1 1.2 1.39 2.26 .59 1.29 .25 .7 -.12 .79 2.76 1.53 3.04 .66 .8

n.s. .035 n.s. n.s. .25 .165 .025 n.s. .20 n.s. n.s. n.s. n.s. .006 .127 .003 n.s. n.s.

-.58 .2 .8 -.67 -1.12 -.7 -.32 .29 -.85 -.23 2.43 1.19 .20 1.32 1.29 1.81 .37 -1.27

n.s. n.s. .n.s n.s. n.s. n.s. n.s. n.s. n.s. n.s. .025 .237 n.s. .188 .2 .072 n.s. .207

Not corrected for different sessions: session 2 was conducted almost one year later Several people gave nonsensical answers such as 'month income is f300'. These are included in the analysis.

97

Causes and consequences of patience

121

Table A1: Causes of patience (continued) CFC-scale t-value Sig. Parental characteristics (continued) Employees father98 1.66 .1 Employees mother .04 n.s. Mother working before birth 1.54 .126 Mother working in the year of birth .3 n.s.99 Mother working age 5 .71 n.s. Mother working age 14 1.86 .064 Wealthy neighborhood 1.2 .231 At home in the neighborhood -1.6 .114 Friends from the neighborhood 2.4 .017 Status parents in the neighborhood 2.6 .011 Visiting neighbors -.4 n.s. Wealthier than the neighborhood 2.5 .013 Family composition Number of siblings -1.51 .133 Parents alive .45 n.s. Year of death father (5 cases) 1.5 .222 Year of death mother (9/6 cases) 1.56 .114 Age of divorce (20/16 cases) -2.77 .013 Parents divorced -1.39 .167 People at home (5) -.48 n.s. People at home (14) -.51 n.s. Number of rooms at home (5) .64 n.s. Number of rooms at home (14) .51 n.s. Rooms minus people (5) .94 n.s. Rooms minus people (14) .92 n.s. Type of home (14) 1.76 .081 Type of home (5) 1.77 .08 Own room age 14 .6 n.s. Own room age 5 -1.34 .182 Family environment Scale ‘ statements over parents’ .87 n.s. Self-control father .7 n.s. Self-control mother Parents find farsightedness important 2.1 .036 Raising More tasks at home than friends 3.0 .003 I do not want to disappoint my parents 1,8 .1 I had a strict raising 3.1 .002 Other variables concerning 'raising' are not significant 98

Money scale t-value Sig. .6 -1.0 -.58 -37. -2.67 -1.5 -1,3 .53 1.8 -.94 -1.6 .99

n.s. n.s. n.s. n.s. .006 .137 .11 n.s. .07 n.s. .14 n.s.

-.84 n.s. .027 n.s. Not computed 2.5 .063 .74 n.s. -1.88 .074 -.061 n.s. .44 n.s. -1.92 .057 -1.55 .124 -1.76 .081 -2.07 .04 -.002 n.s. -1.39 .167 -.21 n.s -.37 n.s. 1,0 -1.3

n.s. .19

1.58

.12100 n.s. n.s. n.s.

Some people stated ‚none‘ despite having a father/mother who was not self-employed. Analysis does not correct for this. 99 All four variables concerning working history of the mother were put into one regression. Values given in the text refer to this regression. Including only one variable in the regression does not significantly alter values. 100 For the ‚money-scale‘ values differ strongly depending on the combination of variables, but no significance levels.

122

Chapter 5

Table A1: Causes of patience (continued) CFC-scale Money scale t-value Sig. t-value Sig. Eating Scale ‘eating’ 2.39 018 -.56 101 Washing hands before eating -1.77 071 1.49 Finishing your plate 1.98 .049 -1.46 Table manners 1.5 .13 .75 Eating sufficient fruit and vegetables 1.52 .131 -1.0 Eating with the whole family 2.1 .041 .22 Playing Scale ‘playing’ -1.1 n.s. -1.93 Standardized values -.8 n.s. -2.0 Making noise -1.73 .086 -.7 Other variables concerning 'playing' are not significant Engagement parents Kindergarten .78 n.s. -1.35 Type of school (ANOVA) .24 n.s. 1.84 Parental activities at school .94 n.s. .07 Visits parents evening 1.8 .073 -.58 Number of children's books 1.92 .057 -.7 Parental control of homework at ages 5 and 14 had no effect on either variable Who enforced rules at home (ANOVA) .56 n.s. .68 Scale 'leisure at primary school' .4 n.s. -.12 Scale 'leisure at high school' 1.14 n.s. .04 Scale 'leisure activities combined' .6 n.s. .24 How often accompanied parents leisure activities at age 5 -.35 n.s. 1.5 How often had parents serious discussions with you (age 5) 2.36 .02 -.9 How often visited parents cinema or theatre with you (age 5) 1.3 .197 -1.8 How often accompanied parents leisure activities (14) .6 n.s. 2.5 How often had parents serious discussions with you (age 14 ) -.2 n.s. -1.7 How often visited parents cinema or theatre with you (age 14) 1.27 .204 -1,1 Stimulation of knowledge and values Scale 'knowledge' 1.05 n.s. .45 Scale 'values' 1.81 .073 1.06 ‘Work ethics’ 2.57 .011 1.25 ‘Good manners’ .4 n.s. 1.75 General moral development 2,9 .003 .63 ‘General development’ 2,6 .01 -.67 ‘Social abilities’ 2.57 .011 2.3 ‘Practical abilities’ -2.2 .028 -.66 ‘Cultural abilities’ -.48 n.s. -1.58

101

n.s. .139 .146 n.s. n.s. n.s. .056 .052 n.s. .179 .124 n.s. n.s. n.s. n.s n.s. n.s. n.s. .137 n.s. .07 .017 .09 .25 n.s n.s. .212 .083 n.s. n.s. .022 n.s. .116

T-values for single items on the scale are given only, if on at least one scale they exceed 1 or –1.

Causes and consequences of patience

123

Appendix B Consequences of patience102 In the analyses presented here the CFC-scale and the money-scale were entered separately as independent variables into a linear regression analysis. Again, the table below gives much more details than it is usually done in the social sciences, because these details might provide useful hints for future research. Results of t-values (two-tailed test) for the regression analysis are given, all variables entered separately (significance levels are given if p < .2, boldface if p < .1) Table B1: Consequences of patience CFC-scale t-value Sig. Health Index How is your health generally Frequency of contact with doctor Nutrition Scale ‘junk food’ How often do you eat junk food as a warm meal How often do you have breakfast per week How often do you eat fruit per week Hoe often do you eat snacks How often do you eat sweets per week Alcohol consumption How often do you drink alcohol Average alcohol consumption during the week Average alcohol consumption during weekend How often are you drunk Scale ‘alcohol use’ Smoking and drug use How much smoking Age first smoking How often drug use Age first drug use Sporting activities Scale ‘sport activity’

102

Money scale t-value Sig.

.46 -.54

n.s. n.s.

-.22 .53

n.s. n.s.

-1.2 -2.21 1.06 -.55 .3 -.77

.23 .045 n.s. n.s. n.s. n.s.

.21 1.8 -1.83 -1.0 1. .55

n.s .074 .07 n.s. n.s. n.s.

-1.65 -2.57 -2.47 -.04 -2.54

.101 .011 .012 n.s. .012

-1. -.12 .03 .05 -.55

n.s. n.s. n.s. n.s. n.s.

-2.64 .72 -3.05 .95

.009 n.s. .003 n.s.

.18 -.76 .73 2.0

n.s. n.s n.s. .052

3.27

.001

.51

n.s.

Both scales are entered as independent variables simultaneously.

124

Chapter 5

Table B1: Consequences of patience (continued) CFC-scale t-value Sig. Engagement in school and university How well did you your homework at high school Final results at school could have been better How well do you prepare exams How often do you party before exams How often did you follow topics not required103 Ever applied (Logistic regression: Wald) Preparation application Deviant behavior Scale ‘crime’ Staying away from school without valid reasons104 Bus, tram or Metro use without paying Driving under influence Deliberately crossed a stop sign or red light Stole something in a shop Life styles Age first relation How many relations Living together (Logistic regression: Wald) Frequency of using contraceptives when having sex How often do you exceed your bank account Scale ‘leisure activities’ How often do you visit museum or theater Reading not required literature for study Deliberately crossing red-light

103

Money scale t-value Sig.

2.05

.175 n.s. n.s. .061 n.s. n.s. .043

.02 .014 .178 n.s. .283 .149

-1.3 -.5 -1.47 -1.4

.2 n.s. .145 .164

n.s. .098 .029 n.s. .017 n.s. .041 .021 n.s.

.24 1.66 -.12 -.8 -.34

n.s. .099 n.s. n.s. n.s.

-1.2 -.12 -1.4

.223 n.s. .16

3.75 2.07 4.42 -1.49 4.0 .004 2.14

.000 .04 .000 .138 .000 n.s. .034

-2.35 -2.5 -1.4 -.34 -1.08 -1.45 -.17 1.66 5.1 .47 -2.41 .77 2.06 .23 -.84

1.36 .24 .44 -1.89 -.52

The four items ‘following topics that were not required for studying‘, ‘doing homework during high school', ‘final results at school could have been better‘, and ‘how well do you prepare exams' yield a scale with an alpha of .6. Eliminating ‘following topics....‘ raises alpha to .72. I have not yet included such a scale. 104 Other variables have no effect on either variable.

Causes and consequences of patience

125

Appendix C Consideration of future consequences scale

The consideration of future consequences scale was introduced and validated by psychologists Stathman, Gleicher and Boninger (1994). They compared this scale with several other measures for occupation with future outcomes and report that this scale serves as a better predictor in a variety of settings, e.g., health or environmental behavior. Since this scale appeared to be useful in a variety of settings, it was used in this research as well. It was taken as a general measure for patience, because of its generality. Alinda van Bruggen helped to translate the text into Dutch. Subjects were presented with the following text (in Dutch): For each of the statements below, please indicate whether or not the statement is characteristic of you. If the statement is extremely uncharacteristic of you (not at all like you) please write a ”1” to the left of the question; if the statement is very characteristic of you (very much like you) please write a ”5” next to the question. And, of course, use the numbers in the middle if you fall between the extremes. Please keep the following scale in mind as you read each of the statements below. 1 2 extremely somewhat uncharacteristic uncharacteristic

3 uncertain

4 5 somewhat extremely characteristic characteristic

_______ 1. I consider how things might be in the future, and try to influence those things with my day to day behavior. _______ 2. Often I engage in a particular behavior in order to achieve outcomes that may not result for many years. _______ 3. I only act to satisfy immediate concerns, figuring the future will take care of itself. _______ 4. My behavior is only influenced by the immediate (i.e., a matter of days or weeks) outcomes of my actions. _______ 5. My convenience is a big factor in the decisions I make or the actions I take. _______ 6. I am willing to sacrifice my immediate happiness or well-being in order to achieve future outcomes. _______ 7. I think it is important to take warnings about negative consequences seriously, even if the negative outcome will not occur for many years. _______ 8. I think it is more important to perform a behavior with important distant consequences than a behavior with less-important immediate consequences. _______ 9. I generally ignore warnings about possible future problems because I think the problems will be resolved before they reach crisis level. _______ 10. I think that sacrificing now is usually unnecessary since future outcomes can be dealt with at a later time. _______ 11. I only act to satisfy immediate concerns, figuring that I will take care of future problems that may occur at a later date. _______ 12. Since my day to day work has specific outcomes, it is more important to me than behavior that has distant outcomes.

126

Chapter 5

Appendix D The Money-scale

Calculating discount rates from delay charges for hypothetical amounts of money is a common procedure in economics (e.g., Loewenstein & Thaler 1989, Cairns & van der Pol 2000). In this research a text from HiN95 (see Bruins & Weesie 1996 for a closer description) was adapted. Subjects were presented with the following text (in Dutch) Suppose you won a certain amount in a lottery. You can choose between two methods of payment. (1) You get the amount immediately, or (2) a somewhat higher amount will be paid later. Below we give you always the amount you receive immediately. State always how high the second amount has to be, in order for you to make it acceptable to wait the specified time. (1) ƒ 1000 immediately (2) ƒ 500 immediately (3) ƒ 25 immediately (4) ƒ 100 immediately (5) ƒ 10 immediately (6) ƒ10000 immediately

at least ƒ __________ to wait three months at least ƒ ___________ to wait one year at least ƒ __________ to wait one month at least ƒ __________ to wait three months at least ƒ __________ to wait half a year at least ƒ __________ to wait two years

Chapter 6

Parallels between 'anomalies' in risky and intertemporal choice

Summary*

In this chapter the close resemblance between the axiom systems for risky and intertemporal choice is demonstrated and the similarity between their regular behavioral violations is shown. It is argued that these 'anomalies' are due to a cognitive tendency to compare more or less ‘distant’ outcomes with a multi-dimensional reference point. Due to diminishing sensitivity, deviations occurring close to the reference point have a larger impact on preference than identical deviations occurring further away from it. Due to loss aversion, negative deviations from the reference point have a larger impact than positive ones. Finally generalizations of the empirical regularities to interspatial and interpersonal choice are suggested. *

This chapter benefited from discussions with Marcel van Assen, Laurie Hendrickx, and Christian Steglich. None of them is responsible for any remaining errors.

128

Chapter 6

6.1. Introduction

Most people have a clear preference for how they want to receive benefits: they want them as large as possible, certain, now, here, and for themselves. Any deviation from one of these ‘ideal’ aspects inevitably leads to a devaluation of the outcome. Correspondingly, most people also have clear preferences as to how they want losses to occur: they want them as small as possible, not to occur to themselves, far away, as late as possible and to be uncertain. In other words: any distance of a positive outcome from the state ‘large, certain, now, here and for myself’ leads to a discounting of the object’s value, and vice versa, for negative outcomes. ‘Distance’ thus refers to several preference dimensions: the outcome may be small (i.e., distant from ‘large’), risky (distant from ‘certain’), temporally distant (distant from ‘now’), spatially distant (distant from ‘here’) or socially distant (distant from ‘myself’). However, up to now these dimensions of distance and their consequences for choice behavior have been treated to a varying extent in separate lines of research. Sociologists and psychologists have developed several conceptions of social distance (see already Bogardus 1959) and social comparison but provided few links to choice behavior. Research on social orientation and its consequences for choices in social dilemmas (e.g., Liebrand, Messick & Wilke 1992) as well as on the effect of altruism on choices (e.g., Becker 1996) has been carried out rather extensively by psychologists and economists alike, but these findings are generally not related to concepts of distance. However, some links have been made (mainly by psychometrists studying risk perception) between the discounting of spatially distant outcomes and decision theory (see Vlek & Keren 1992 for an overview). Although Vlek and Keren suggest that the utility of spatially distributed outcomes could be assessed in a way similar to the assessment of temporally distributed outcomes they (1992: 267) state: ‘Decision-theoretic research on spatial dilemmas and preferences is highly underdeveloped.’ Recently, interest with respect to this area has risen and problems like the not-in-my-back-yard (NIMBY) phenomenon have received considerable attention, e.g., Frey, Oberholzer-Gee and Eichenberger (1996). Risky, uncertain1 and intertemporal choices have been investigated extensively by decision theorists. Several axiomatic theories were developed: Expected Utility (EU) theory for risky choice (e.g., Von Neumann & Morgenstern 1982 [1944], Luce & Raiffa 1957), Subjective Expected Utility (SEU) theory for choice under uncertainty (e.g., Savage 1954), and Discounted Utility (DU) theory for intertemporal choice (e.g., Fishburn & Rubinstein 1982). To the extent that the preferences of a decision maker satisfy the relevant axioms, these theories allow one to represent preferences by numerical outcome utilities. In the last decades several systematic deviations from the respective normative models have been detected (see Kahneman & Tversky 1979 and Loewenstein & Thaler 1989 for overviews). According to standard economic theory these deviations are called anomalies. In response to these empirical challenges to ‘rational’ choice theory, several attempts have been undertaken to develop descriptive theories accounting for these anomalies. However, although the nature of the underlying axiomatic systems is very similar, only few authors (Prelec & Loewenstein 1991 - in the following: PL 91; Quiggin & Horowitz 1995) have taken into account the similar nature of the anomalies, generalized empirical results to other preference dimensions (Björkman 1

Economists often distinguish risk and uncertainty by whether probabilities of events are objectively given (risk) or have to be estimated by the decision maker (uncertainty).

Parallels between 'anomalies' in risky and intertemporal choice 129 1984; Rachlin & Rainieri 1992), or have imported insights from other domains to their field of interest (Laibson [1997] models temporal discounting by adopting a formula originally created for representing intergenerational altruism). In this book it is argued that observed results in the domains of risky choice and intertemporal choice are in fact due to a general behavioral tendency to discount ‘distant’ outcomes of decision alternatives. Hence, they can be generalized to other preference dimensions where discounting takes place, e.g., to interspatial or interpersonal choice (see the next chapter for empirical demonstrations). It is further argued that this tendency can be expressed in several predictable shifts in the salience of decision attributes such as outcome size, riskiness, time delay, spatial distance and social distance, as their value, or their combination is changed. This and the following chapter are devoted to a closer examination of this thesis. The chapter is structured as follows: Section 2 reviews the normative axiomatic approaches to choice under risk and uncertainty as well as intertemporal choice. Further, it clarifies the close relation between them. Because research on spatial and social distance of decision outcomes is generally not yet related to choice behavior and research, concepts and findings from these fields will not be reported in this section. Section 3 compares empirical results about risky choice and intertemporal choice. Similarities in these results are briefly presented in section 4. I am well aware that there is some overlap between these three sections and Chapter 1 introducing the DU-model and it's accompanying anomalies. However, for demonstrating the parallelism between choice under risk and intertemporal choice this overlap is inevitable here. Section 5 summarizes the regularities and suggests their application to spatial and social discounting. These suggestions are investigated empirically in Chapter 7. The research presented in this and the following chapter builds on and extends the following previous research. Björkman (1984) speculates that human cognitive capacities with regard to distant events are limited and that therefore outcomes occurring here and now naturally get more weight. Vlek and Keren (1992) argue that the trade-offs between size and certainty of benefits, size and temporal delay of benefits, size and spatial distance of benefits and size and other-directedness of benefits are biased towards ‘myself, here, now and for sure’ in a similar way. In a related paper Rachlin and Rainieri (1992) generalize intertemporal hyperbolic discounting functions to other preference dimensions and predict similar preference reversals when discounting risk, sound or social distance. A systematic and thorough examination of common anomalies to EU theory and DU theory has been provided by Prelec and Loewenstein (PL 91). They state that similarities between choice under risk and choice over time reflect ‘certain fundamental properties of multidimensional prospect evaluation’ (PL 91: 770/771). Nevertheless, PL conclude their paper by mentioning ‘the possibility that the similarity of choice behavior is peculiar to time and uncertainty - that it reflects their intimate connection rather than fundamental processes underlying multiattribute choices’ (PL 91: 784). They argue that time and uncertainty both pertain to the choice object’s delivery that causes the similar behavioral patterns. But obviously the number of attributes pertaining to the choice object’s delivery is not restricted to time and uncertainty alone: whether you or somebody else receives an object may affect your evaluation of the object even more than whether its delivery is immediate or delayed. So, the thesis advanced in the present and tested in the following chapter is that the devaluation or discounting of decision outcomes applies to all major preference

130

Chapter 6

dimensions where an outcome may either be ‘close’ to or more ‘distant’’ from the decision maker. 6.2. Axiomatic models for EU, SEU and DU

This section begins with a paragraph clarifying formal notation: An immediate lottery (or gamble or risky prospect) L is denoted by (x1, p1; x2, p2; ...; xn, pn), including outcomes (x1, x1, ..xn) and their corresponding and known probabilities (p1, p2, ...pn) associated with each outcome. A riskless temporal prospect is denoted by (x1, t1; x2, t2; ...; xn, tn), including outcomes (x1, x2, ..., xn) and corresponding temporal attributes (t1, t2, ..., tn) expressing the temporal distance associated with an outcome. ‘~’ expresses preferential indifference between lotteries, temporal prospects or outcomes. ‘˜’ or ‘š‘ expresses ‘preference or indifference’, while ‘p‘ or ‘f‘ expresses strict preference between lotteries, temporal prospects or outcomes. Preferential symbols may be indexed, e.g., ~r, to describe preference from a specific reference state r. Preferential symbols with respect to a specific response mode are again indexed, e.g., ~r for the response mode rating, ~p for the response mode pricing etc. A lottery that yields $50 with probability .5, $20 with probability .2, and nothing with a probability of .3 is described as: (50, .5; 20, .2; 0, .3). A temporal prospect yielding $50 delayed three units of time (might this be days, months or years) and $20 delayed 20 time units is described by: (50, 3; 20, 20). Note that a ‘zero’ time attribute means that the outcome would be received immediately. Note further that in EU theory and SEU theory the outcomes are mutually exclusive: a decision maker receives only one outcome xi from a risky or uncertain prospect (x1, p1; x2, p2; ...; xn, pn). In DU theory this is not the case: a decision maker receives every outcome specified in a prospect, which thus is a time-stream of outcomes. Gambles yielding time-dependent outcomes are denoted by to (x1, p1, t1; x2, p2, t2; ...; xn, pn, tn). Such gambles are played immediately, but the outcome might be realized later in time. Additionally, λ and γ are probability distributions on the time dimension. That is, an outcome x might occur with a certain probability at time t1, with a certain probability at time t2 and so on. The format above may be generalized to (x1, p1, t1, s1, l1; x2, p2, t2, s2, l2; ...; xn, pn, tn, sn, ln), describing an outcome to be obtained with a corresponding probability, a corresponding time delay, for a certain subject s at a particular location l. In the following, zero outcomes and distances will be omitted. Expected Utility Theory

The first axiomatization underlying rational choice under risk was undertaken by Von Neumann and Morgenstern in their by now classic ‘Theory of Games and Economic Behavior’ (Von Neumann & Morgenstern 1982 [1944]). Further axiomatic formulations were developed by Herstein and Milnor (1953) and by Luce and Raiffa (1957). Despite some formal differences between these axiom sets the general result is the same: whenever the preferences of a decision maker satisfy the axioms, it is possible to represent preferences by a linear utility index, i.e. utility can be expressed by numbers. The system of Luce and Raiffa, stated in ‘Games and Decisions’, is relatively straightforward and is presented below (Luce & Raiffa 1957: 23-31):

Parallels between 'anomalies' in risky and intertemporal choice 131 1. Ordering of alternatives. The ordering ‘preference or indifference’ is: (a) complete, i.e. it holds for any two outcomes. That is, for any xi and xj either xi f xj, xi p xj or xi ~ xj; and (b) it is transitive. That is, if xi f xj and xj f xk then xi f xk. 2. Reduction of compound lotteries. Compound lotteries can be decomposed into more basic outcomes (if included in the compound lottery) by use of the probability calculus: (x1, p1; [x1, p3; x2, p4], p2; x2, p5) ~ (x1, [p1 + p2*p3]; x2, [p5 + p2* p4]). 3. Continuity. For each outcome xi there is a lottery involving xj and xk, where xj f xi f xk, such that there is a probability p so that xi ~ (xj, p; xk, [1-p]). 4. Substitutability. If a subject is indifferent between two gambles (A and B), he is indifferent between any compound lotteries that are identical except for these two gambles. If A ~ B then (A, p1; x2, p2;.....; xn, pn) ~ (B, p1; x2, p2; ....; xn, pn). Further, if A f B then the relation between the lotteries changes accordingly. 5. Transitivity. Preference and indifference between lotteries Li are transitive relations. If L1 f L2 and L2 f L3 then L1 f L3, and also if L1 ~ L2 and L2 ~ L3, then L1 ~ L3. Contrary to axiom 1 above here transitivity is defined across lotteries whereas in axiom 1 transitivity was defined over outcomes. 6. Monotonicity. If there are two lotteries yielding the same outcomes but having different probabilities, one should prefer the lottery in which the more preferred outcome has a higher probability, i.e. for x1 f x2: if and only if p1 f p2 then (x1, p1; x2, [1- p1]) f (x1, p2; x2, [1-p2]). If a decision maker accepts these six axioms as reasonable action principles, he should choose among a set of options the alternative A (compared to A’ etc.) with the highest expected utility: EU (A) = 3 pi u(xi), (1) where pi is the probability of outcome xi while u(xi) indicates its utility. Subjective Expected Utility Theory

For choices under uncertainty the first axiomatic formulation was developed by Savage (1954). This axiom set is a generalization of EU theory because if probabilities are known, EU theory and SEU theory converge. Savage’s axiom system will not be presented here in further detail for three reasons. First, he uses a somewhat different axiomatic system2 which hinders comparisons with the other axiomatic formulations. Second, empirical evidence has been gathered mainly with respect to EU theory, under probability-given conditions. Finally, Tversky and Fox (1995) provide evidence that people, when confronted with uncertainty, somehow estimate probabilities for these outcomes and then handle these options as in standard risky choice situations. Choice under uncertainty might thus be reduced to choice under risk. For completeness, it should be stated that SEU theory, too, involves the principle that the alternative A (compared to A’ etc.) with the highest subjective expected utility is to be chosen:

2

These axioms include an ordering axiom, a continuity axiom, several independence conditions and a dominance postulate.

132

Chapter 6

(2) SEU (A) = 3 ps(xi) u(xi), where ps(xi) denotes the subjective or personal probability that outcome xi will occur. Discounted Utility Theory

Discounted utility theory examines ‘the effect of the time of realisation of an outcome on the relative desirability of that outcome’ (Fishburn & Rubinstein 1982: 677) and specifies conditions for numerical utility representations of temporally distributed outcomes. This problem had been abandoned by EU and SEU theorists since Von Neumann and Morgenstern’s classical statement that ‘it would be an unnecessary complication, as far as our present objectives are concerned, to get entangled with the problems of the preferences between events in different periods of the future. It seems, however, that such difficulties can be obviated by locating all ”events” in which we are interested at one and the same, standardised moment, preferably in the immediate future’ (Von Neumann & Morgenstern 1982 [1944]: 3.3.3.). As in the case of EU theory, several different axiom systems for DU were developed (e.g., Koopmans 1960; Lancaster 1963; Fishburn & Rubinstein 1982) among which Fishburn and Rubinstein’s system will be investigated more closely, as follows. 13. Order. ˜ is a weak order on the Cartesian product X * T of outcomes and time periods. This simply states that all time-dependent outcomes are comparable amongst each other. Although formulated slightly differently this axiom is basically equivalent to Luce and Raiffa’s axiom 1 (see above). 2. Monotonicity. If x š y then (x, t) š (y, t). If an outcome is preferred to another immediately, it will also be preferred at any later point in time. 3. Impatience. If t1 < t2 then (x, t1) f (x, t2) for all x > 0; (x, t1) ~ (x, t2) for all x = 0; (x, t1) p (x, t2) for all x < 0. This axiom states that an individual prefers positive outcomes rather sooner than later, is time-indifferent as to receiving a zero outcome and prefers to defer negative outcomes. 4. Continuity. {(x, t): (x, t) š (y, s)} and {(x, t): (y, s) š (x, t)} are closed in the product topology on X and T. The set of points (x, s) weakly preferred to (y, s) is closed as is the set of points (y, s) weakly preferred to (x, t). This axiom ensures that the resulting function is continuous. 5. Stationarity. If (x, t1) š (y, t1 + r) then (x, t2) š (y, t2 + r). This axiom states that preference between two time-dependent outcomes depends only on the difference between time periods, not on the absolute delay of both events. Due to their reluctance to accept stationarity as a psychologically valid assumption, Fishburn and Rubinstein introduce a weaker ‘substitute’ for this axiom. In combination with the first four axioms this axiom suffices for proving the existence of an intertemporally separable utility function. 6. If (x, t1) ~ (y, t2) and (y, t3) ~ (z, t1) then (x, t3) ~ (z, t2), for outcomes z < x < y and delays t1 < t2 < t3. To illustrate this axiom, replace z, x, y with $25, $50, $100 and t1, t2, t3 with time delays 10 days, 20 days and 40 days, respectively. The indifference relations then become: If ($50, 10) ~ ($100, 20) and ($100, 40) ~ ($25, 10), then ($50, 40) ~ ($25, 3

Fishburn and Rubinstein actually include an additional axiom zero which describes the mathematical structure of the outcome-time structure. This axiom is of lesser interest here and it is therefore not reported.

Parallels between 'anomalies' in risky and intertemporal choice 133 20). This axiom is weaker than the stationarity axiom. It does no longer require that preference between all time-dependent outcomes depends on the difference between time periods. Instead it states that if preference between some time-dependent outcomes depends on the difference between time periods, than these outcomes and delays are interchangeable. Fishburn and Rubinstein (1982) further mention the two following separability axioms with respect to the effect of introducing probabilities (given that a utility function consistent with the necessary axioms exists). Similarly to axiom six above, these axioms serve as weaker substitutes for other axioms. 7. If (x1, p1, t1) š (x2, p2, t1) then (x1, p1, t2) š (x2, p2, t2). Preferring the gamble (x1, p1) over the gamble (x2, p2) when the outcome would be realized at time t1 implies that (x1, p1) is still preferred when the outcomes would occur at time t2. 8. If x, y > 0 and (x, λ) š (x, γ) then (y, λ) š (y, γ). For example, if one prefers ($100, .5, now; $100, .5, a year from now) to ($100, three months), then this preference holds for all outcomes of the same sign, e.g., if both outcomes are doubled ($200 in the numeric example above). Hence a decision maker should choose the alternative A (compared to A’ etc.) with the highest discounted utility: DU (A) = 3 δti u (xi),4 (3) where δ is the discount factor per time period, ti the time delay of outcome xi, and u(xi) indicates its ‘present’ or ‘immediate’ utility. As can be seen by comparing EU and DU theory, the underlying axioms are quite similar in nature: three of the axioms are identical (order, monotonicity and continuity), and both axiom systems result in similar additively separable formulas. In a previous analysis of intertemporal choice Lancaster (1963) simply added five axioms (independence from viewpoint, i.e. time separability; intertemporal transitivity; monotonicity of time preference; continuity of time preference; time compounding) to the Von Neumann and Morgenstern axioms, thus again highlighting the close relationship between EU and DU theory.

4

Two things are notable: first, this formula, contrary to the formulas for EU-theory and SEU-theory which are linear in probabilities, is exponential in time. Second, it describes sequences of outcomes. However, it follows from the additively separable form of this formula that sequences are simple aggregations of the single outcomes to which the axioms of Fishburn and Rubinstein apply.

134

Chapter 6

6.3. Behavioral regularities in choice experiments

In the previous section the axioms underlying rational choice under risk and choice over time were presented. If a decision maker accepts these axioms, he should act as implied by the respective formula – he may well behave differently in case he does not feel that these axioms describe his preferences correctly. Economists usually regard these axioms as reasonable action principles for all people. Hence, they model behavior by using the respective formulas. Behavior departing from these models then is called anomalous (anomalies are possible only when these – or any comparable – axioms are accepted). However, anomalies to EU-theory as well as DU-theory abound. Overviews of anomalies to EU theory (e.g., Tversky & Kahneman 1992) as well as overviews of anomalies to DU theory (e.g. Loewenstein & Thaler 1989) are available. Results have even been considered in comparison to each other (PL 91). This section (building on PL 91) intends to present a more comprehensive and systematic account of the empirical results than has been done before. Several effects will be explained and demonstrated on both the outcome and the time or probability attribute of choice alternatives. Some regularities will appear to be anomalous to one theory but consistent with the other. Consider, for example, the choice between two gambles ($4000, .8) and ($3000, 1). Most people prefer the smaller but sure outcome (Kahneman & Tversky 1979). However, dividing the probabilities in both gambles by four - ($4000, .2) vs. ($3000, .25) leads to a reverse preference pattern: most people now prefer the larger less probable outcome. This behavior violates the ‘reduction of compound lotteries’ axiom, thus revealing an anomaly (due to the 'certainty effect' see further below). Now consider a subject who is indifferent between the temporal prospects ($3000, one year) to ($4000, three years). According to DU-theory he should prefer ($4000, three weeks) to ($3000, one week), because the axioms laid out above require that only absolute time differences between options matter. Dividing time delays by 52 changes absolute time differences between options such that they are much smaller in the second choice and thus should alter preference. 6.3.1. Anomalies in risky choice

Below some empirical regularities in risky choice will be presented, viz. the common ratio effect, the common difference effect, framing effects and preference dependence on elicitation methods.5 a) Common ratio effect

An observed regularity in choice under risk with respect to the probability dimension is that multiplying probabilities in two options may change preferences between them:

5

The regularities presented here are used despite proposals about the existence of S-shaped probability weighting functions that are different for gains and losses (as, e.g., Tversky & Kahneman 1992). In fact, little evidence exists for the existence of such different functions. Further, also the specific curvature of the probability weighting function is controversial: Tversky and Kahneman (1992) and Fennema and Wakker (1997) provide evidence for S-shaped weighting functions whereas Prospect Theory’s original weighting function described the data better in Wu and Gonzalez (1996) and Bernstein, Chapman, Christensen and Elstein (1997).

Parallels between 'anomalies' in risky and intertemporal choice 135 If (x, p) ~ (y, p’) then (x, pk) f (y, p’k), pk’ > 0, and k >1.

for x < y, 1 > p > p’ > 0,

1 ≥ pk >

Example: If a person is indifferent between ($100, .45) and ($150, .3) he is likely to prefer ($100, .9) to ($150, .6). Although both expected values are the same, in the latter pair of gambles the probability difference is larger, making it more salient. Contrary to the ‘reduction of compound lotteries’ axiom the decision weight of this dimension becomes stronger and the option more favorable with respect to this dimension becomes relatively more attractive. Kahneman and Tversky (1979) named this the ‘common ratio effect’. An analogous regularity with respect to the outcome dimension would be:6 If (x, p) ~ (y, p’) then (xk, p) f (yk, p’),

for x > y, 1 > p’ > p > 0 and k > 1.

Example: If a person is indifferent between ($150, .2) and ($100, .3) he is likely to prefer ($300, .2) to ($200, .3). In the second pair of gambles the outcome difference is more salient than in the first and hence, the decision weight of the outcome dimension is stronger.7 A specific common ratio regularity for the probability dimension is the certainty effect: If (x, p) ~ (y, p’) then (x, pk) f, (y, p’k),

for x < y, p > p’ > 0 and pk ≈ 1.

Example: If a decision maker is indifferent between ($4000, .2) and ($3000, .25), he will prefer ($3000, 1) to (4000, .8). The common ratio effect is especially pronounced when options are multiplied such that one option is or appears certain. In risky choice behavior is affected qualitatively differently when outcomes are perceived as almost certain or impossible (Kahneman & Tversky 1979). Weighting functions for probabilities are not well-behaved at the end-points. That is, very high probabilities are likely to be either underestimated (when encoded as containing some risk) or overweighed (when encoded as certain). The same holds for very low probabilities. They are either overweighed (when encoded as improbable but likely) or completely neglected (when encoded as impossible). b) Common difference effect

The term ‘common difference effect’ generally describes an intertemporal choice anomaly, i.e., a violation of the stationarity axiom (see above). In intertemporal choice, preference between options is affected by adding an identical time delay to both options (see below). An analogous effect also occurs in choice under risk. Because of the two end 6 Transformations on the outcome dimension are not anomalous to either EU theory or DU theory: because the respective axioms remain silent about value functions for money, every value function is consistent with them. Manipulations of the outcome dimension are included nevertheless because of the correspondence between manipulations of the distance and the outcome dimension. 7 PL 91 (without giving references) argue that this does not hold: ‘Magnifying the ”stakes” of risky prospects increases the importance of the probability dimension, resulting in more risk-averse behavior.’ (PL 91: 782) In other words, the salience of the probability dimension is enhanced and the option being more likely is getting more attractive.

136

Chapter 6

points for the probability scale (certainty and impossibility), this effect is presented for the probability dimension twice. First, probability changes are considered for options with probabilities close to zero, then probability changes for probabilities close to one will be looked at. Regularity for the probability dimension (moving away from impossibility): If (x, p) ~ (y, p’) then (x, p + k) p (y, p’ + k), for x < y, .9 > p + k > p’ + k > 0. Example: If a person is indifferent between ($100, .03) and ($150, .02) he prefers ($150, .06) to ($100, .07). By absolutely enlarging both probabilities, their difference gets less salient and the salience of the probability dimension decreases. Because expected values also change in favor of the larger outcome option this is in accordance with EU theory as well as with theories that assume an S-shaped probability weighting function (concave below .35 and convex above it). See Wu and Gonzalez (1996) for evidence. When probabilities in two options are altered such that they are moving close to certainty, the regularity for the probability dimension becomes: If (x, p) ~ (y, p’) then (x, p + k) f (y, p’ + k), + k > p’ + k > 0

for x < y, 1> p > p’ > 0 and 1 ≈ p

Example: If a person is indifferent between ($100, .95) and ($105, .9) she is likely to prefer ($100, .98) to ($105, .93). Absolutely enlarging both probabilities brings them closer to certainty. The probability difference becomes more salient and the salience of the probability dimension rises. The more likely option may even appear to be certain resulting in an even stronger preference for it. Note that this runs counter to changes in expected value and hence EU theory. Again, this effect is in accordance with theories assuming an S-shaped probability weighting functions (for evidence see Wu & Gonzalez 1996). The analogous regularity for the outcome dimension would be: If (x, p) ~ (y, p’) then (x + k, p) f (y + k, p’),

for x < y, 1> p > p’ > 0 and k > 1.

Example: If a person is indifferent between ($100, .03) and ($150, .02) he is likely to prefer ($150, .03) to ($200, .02). By absolutely enlarging both outcomes their difference gets less salient and the decision weight of the outcome dimension decreases. Here again, expected values change in favor of the smaller-outcome option, thus providing (also) a rational explanation for this effect. According to PL 91 a common difference effect for the outcome dimension ‘seems perfectly compelling’ (PL 91: 782), yet no empirical evidence exists. With respect to the probability dimension the common difference effect due to changes in expected value appears to be trivial for probability changes away from impossibility. However, the preference pattern suggested for probability changes towards certainty runs counter to predictions based on expected value theory.

Parallels between 'anomalies' in risky and intertemporal choice 137 c) Framing effects

Under framing effects in this chapter shifts in preference due to manipulations of the reference point against which options are evaluated are subsumed. The normative theories described in section 2 require that options are evaluated as final states of wealth, i.e., independently from the decision maker’s status quo. Contrary to this assumption, it has been regularly demonstrated that initial entitlements do matter and it therefore matters how outcomes are related to some reference point (e.g., Tversky & Kahneman 1991). Kahneman and Tversky (1984) further demonstrated that this reference point can be manipulated via problem formulation. Two types of manipulations are possible: first, shifting the reference point such that the distinction between gains and losses is reversed. Effects due to this manipulation are called ‘sign effects’. For example, outcomes initially perceived as gains after such a reference point manipulation may appear as losses. Second, reference points may be shifted in 'distance' without affecting the distinction between gains and losses. Outcomes or distances thus may appear to be closer to or further away from the reference point. In combination with the common ratio effect and common difference effect such a manipulation may induce changes in preference. Effects due to this manipulation are referred to as ‘range effects’. First, the regularities with respect to the outcome dimension will be discussed (indices for the reference point thus refer to the outcome dimension). a) Sign effect (preference between options changes when the polarity of outcomes is changed after a reference point manipulation) If (x, p) ~r (y, p’) then (x, p) ps (y, p’)

for r < x < y < s and p > p’.

Example: If a person is indifferent between ($100, .3) and ($150, .2) he is likely to prefer (-150, .2) to (-$100, .3). Changing the polarity of the outcome from positive to negative raises the salience of the outcome dimension. In fact even the following relation holds (Kahneman and Tversky’s [1979] ‘reflection effect’): If (x, p) fr (y, p’) then (x, p) ps (y, p’)

for r < x < y < s and p > p’.

Example: If a person prefers ($100, .3) to ($150, .2) he prefers (-$150, .2) to (-$100, .3). b) Range effect (preference between options changes if outcomes are perceived to be larger or smaller after a reference point manipulation which due to the common difference effect changes preference): If (x, p) ~r (y, p’) then (x, p) ps (y, p’)

for 0 < r < s < x < y and p > p’

Example: Suppose a person is indifferent between ($2000, .75) and ($1500, 1). Now suppose the person just has been given $1000. When confronted with the choice between ($1000, .75) and (500, 1), he will prefer the former option, because the relative outcome distance between options is enlarged and the outcome dimension more salient.

138

Chapter 6

Regularities with respect to the probability dimension are (indices for the reference point now refer to the probability dimension): a) Sign effect on the probability dimension (probability losses weigh heavier than gains in probability): If (x, p) ~r (y, p’) then (x, p) fs (-y, p’)

for x < y , and s > p > p’ > r

Example: Suppose a decision maker currently owns a gamble ($1200, .5) and is indifferent between ($1500, .65) and ($1333, .75). He will prefer the latter option if instead he currently owns a gamble ($1200, .8). Because losses weigh heavier than gains, the probability dimension becomes more salient in the second pair of options. b) Range effect on the probability dimension: due to shifts of reference point, outcomes appear more or less likely. If (x, p) ~r (y, p’) then (x, p) fs (y, p’)

for 0 < x < y and 0 < p’ < p # s < r

Example: If a person is indifferent between ($4000, .2) and ($3000, .25) he will prefer the latter option, if the choice is presented in terms of the following two-stage gamble: in the first stage there is a probability of .75 to end the game and winning nothing and a probability of .25 of entering the second stage. In the second stage one has the choice between ($4000, .8) and (3000, 1). The choice for the second stage has to be made in advance, i.e., before the outcome of the first stage is known. Although, the two-stage gamble is logically equivalent to the first choice problem above, most subjects choose the option with the smaller but 'certain' outcome for the second stage (Kahneman & Tversky 1979). This option appears certain due to the reference point manipulation whereby perceived distance between the two probabilities has been enlarged. d) Violations of independence of elicitation methods

Although every manipulation reviewed in the previous subsections leads to changes in preference between options, the term ‘preference reversal’ initially involved an effect of the elicitation method on preferences between gambles. Slovic and Lichtenstein (1968) found that buying and selling prices of gambles were primarily determined by payoffs whereas rating the attractiveness of gambles depended more strongly on the probability of winning. In a subsequent study they were able to demonstrate these preference reversals even for experienced gamblers (Lichtenstein & Slovic 1973). This phenomenon has been replicated several times (e.g., Grether & Plott 1979, Tversky, Slovic & Kahneman 1990). Tversky, Slovic and Kahneman (1990) argue that people overprice the option with the lower probability of winning but a higher payoff. Subjects stating preference between options in terms of money, give more decision weight to the money attribute of the option and the option which is more attractive in this respect is preferred. This effect is interpreted (and thus included in this overview) as an influence of the elicitation method on the perceived distance of outcomes and probabilities. Making a dimension more salient has the same effect as (multiplicatively) enlarging the distance of

Parallels between 'anomalies' in risky and intertemporal choice 139 this dimension, because even small distances from the reference point are now perceived to be relatively large. Regularity for the outcome dimension (salience of a dimension is enhanced by the response mode): If (x, p) ~r (y, p’) then (x, p) fp (y, p’)

for x > y, 0 < p < p’ < 1.

Example: If a person rates ($100, .5) as equally attractive as ($200, .25), he will state a higher selling price for the latter option. Asking for selling prices raises the salience of the outcome dimension. Regularity for the probability dimension (salience of a dimension is enhanced by the response mode): If (x, p) ~r (y, p’) then (x, p) fp (y, p’)

for x < y, 0 < p’ < p ≤ 1.

Example: If a person rates ($100, .5) as equally attractive as ($200, .25), he will state a probability higher than .25 when asked to specify a probability p which makes ($200, p) equally attractive as ($100, .5). Asking for probabilities raises the salience of the probability dimension. e) Reactions to anomalies in risky choice

Decision theorists initially responded to empirical findings anomalous to EU theory by developing theories based on other axiom(s) than the ones questioned by experimental evidence (e.g., Machina 1982). That is, another set of axioms was designed against which no contrary evidence currently exists and the existence of a utility function was derived from them. Given the abundance of violations of the axioms underlying the normative models (only the violations common to EU theory and DU theory have been mentioned in the previous sections) not many rational axioms could be left unchanged. Subsequently, purely descriptive theories were developed, among which Kahneman and Tversky’s (1979) Prospect Theory is probably the most important. Here decision makers are assumed to replace probabilities by decision weights. These weights are allowed to differ from the probabilities explicitly given. Prospect Theory also contains statements about reference point and framing effects and explains most of the findings that deviate from EU theory. Recently, both response strategies (selecting other axiom sets, and developing descriptive theories) converged. EU theory has been generalized to rank dependent utility theory (RDU theory, e.g., Quiggin 1982, Yaari 1987). These theories, including an updated version of Prospect Theory: Cumulative Prospect Theory (CPT, Tversky & Kahneman 1992), have become very popular in the recent past and they have also been axiomatized (e.g., Wakker & Tversky 1993). In RDU theory weighting functions for probabilities have been combined with the notion that the weighting of outcomes and probabilities is rank-dependent, i.e., more attention is given to extreme outcomes and probabilities. Consider two prospects with n > 2 outcomes, their respective probabilities and one outcome included in both prospects. Preference between options is allowed to change if this outcome or its respective probability is altered in both options in an

140

Chapter 6

identical way such that the ordering of outcomes within an option changes because of the alteration. Weber and Krisner (1997) give the following example: suppose a person is indifferent between ($3.50, .55; $6, .25; $7, .2) and ($3.50, .55; $4.50, .25; $10.50, .2). Now change the common outcome in both options ($3.50, .55) to another common outcome in both options ($6.50, .55). According to RDU theory this change may well change preference between the options because the order of outcomes within the options has changed: $6 and $4.50 are now the lowest outcomes within a prospect. 6.3.2. Anomalies in intertemporal choice

We now present empirical regularities in intertemporal choice analogous to the ones observed in risky choice. a) Common ratio effect

In choice under risk, the term common ratio effect describes the decision maker’s sensitivity to multiplying probabilities in pairs of gambles. In intertemporal choice the term common ratio effect has not been used previously. However, several studies report a similar effect with respect to the money dimension (the so-called ‘magnitude effect’): Discount rates are lower for higher-valued objects (see Thaler 1981 and Benzion et al. 1989 for hypothetical questions; Holocomb & Nelson 1992 for outcomes with real payoffs). Yet analogous results for the time dimension are missing, but they are intuitively compelling and in accordance with DU theory. Regularity for the distance dimension: If (x, t) ~ (y, t’) then (x, tk) f (y, t'k ),

for x < y, t < t’ and k > 1.

Example: If an individual is indifferent between ($1000, one year) and ($1500, two years) she is likely to prefer (1000, four years) to (1500, eight years). The time difference in the second pair of options is more salient.8 Regularity for the outcome dimension: If (x, t) ~ (y, t’) then (xk, t) p (yk, t’ ),

for x < y, t < t’ and k > 1.

Example: If an individual is indifferent between ($1000, one year) and ($1500, two years) he is likely to prefer (3000, two years) to (2000, one year). The outcome difference in the latter pair of options is more salient. b) Common difference effect:

The best documented anomaly in intertemporal choice is that discount rates decline with time delay (e.g. Ainslie 1975; Thaler 1981; Benzion et al. 1989). Suppose that a decision maker is indifferent between two options that differ in outcome and time delay. By adding a common constant to the time delay in both options preference changes in favor 8

Because absolute time differences between options have increased, DU theory assumes the more immediate option to become more favorable.

Parallels between 'anomalies' in risky and intertemporal choice 141 of the later one, constituting a violation of the stationarity axiom. Instead of being exponential as assumed by economic theory, the discount function inferred from experiments rather has a hyperbolic form. Several names were given to this phenomenon. Here, as in PL 91, it will be labeled the common difference effect. The same effect (adding a common component to a given dimension in both options lowers the decision weight of that dimension) should also apply to the outcome dimension. Although it has not been tested explicitly in an intertemporal choice context, it simply follows from decreasing marginal utility. Regularity for the time dimension: If (x, t) ~ (y, t’) then (x, t + k) p (y, t' + k),

for x < y, k > 0 and 0 < t < t’.

Example: If an individual is indifferent between ($1000, one year) and ($1500, two years) he is likely to prefer ($1500, five years) to ($1000, four years). Absolutely enlarging both time delays decreases their relative difference. The time difference in the latter pair of options thus becomes less salient. Regularity for the outcome dimension: If (x, t) ~ (y, t’) then (x + k, t) f (y + k, t'),

for x < y, k > 0 and 0 < t < t’.

Example: If an individual is indifferent between ($1000, one year) and (1500, two years) he is likely to prefer ($3000, one year) to ($3500, two years). Absolutely enlarging both outcomes decreases their relative differences. The outcome difference in the second pair of options becomes less salient. A special case of the common difference effect is the so-called ‘immediacy effect’. Though formally identical to the common difference effect, several authors suggest that changes from no delay to any delay are not only a quantitative change but constitute a qualitatively distinct manipulation (Benzion et al. 1989, PL 91, Keren & Roelofsma 1995). Here, further experimentation is needed. c) Framing effects

Again sign effects and range effects and range effects are distinguished. For the time dimension sign effects create temporal gains or losses (speed-ups or delays) while range effects make outcomes appear more or less temporally distant. We begin with presenting the regularity for the time dimension (indices for the reference point refer to the time dimension): a) Sign effect on the time dimension (delaying is more aversive than speeding-up preferable) If (x, t) ~q (y, t’), then (x, t) pr (y, t’)

for x < y, and q < t < t’ < r.

Example: If someone is indifferent between (delaying $100 by one year) and (delaying $150 by two years), he will prefer (speeding-up $150 by two years) to (speeding-up $100 by one year). Because temporal losses weigh heavier than gains, changing the polarity

142

Chapter 6

from negative (delaying) to positive (speeding-up) lowers the weight of the time dimension, thereby making the outcome dimension more salient. A gain/loss asymmetry for the time dimension in intertemporal choice has been introduced by Loewenstein (1988) and was replicated in studies by Shelley (1993, 1994). They argue that decision makers adapt to receiving an outcome at a specified and known time point, which constitutes the temporal reference point. Then either a premium to speed-up receiving the outcome, or a charge to delay the receipt of the outcome can be elicited. Speeding-up the receipt of a positive outcome generally is perceived as an immediate gain (and future loss), while postponing such an outcome is perceived as an immediate loss (and future gain). Because losses loom larger than gains and future outcomes are discounted the asymmetry above occurs. Likewise, for negative outcomes delay would be seen as an immediate gain, while speeding-up would be an immediate loss. b) Range effect on the time dimension (delays appear more or less distant after a reference point manipulation If (x, t) ~r (y, t’) then (x, t) fs (y, t’)

for 0 < x < y and r < s # t < t’

Example: If a subject is indifferent between ($100, 26 weeks) and ($110, 30 weeks) he will prefer the first option when confronted with the following two-stage game. In the second stage one has the choice between receiving $100 immediately and $110 four weeks later. In the first stage the subject randomly has to wait either 6 or 26 weeks for a letter. However, only in case of waiting 26 weeks for the letter the subject enters the second stage. Thus the choice the subject has to make for the second stage is (the choice between $100 immediately and $110 after four weeks also has to be made immediately) in fact whether he prefers $100 in 26 weeks or $110 in 30 weeks (see Roelofsma 1994). The similarity with the corresponding Kahneman and Tversky (1979) experiment for choice under risk (see distance effect in section 3.1.3) was created on purpose to demonstrate the analogy between risky and intertemporal choice. The corresponding regularities for the outcome dimension are (indices for the reference point refer to the outcome dimension): a) Sign effect on the outcome dimension (gains are discounted stronger than losses) If (x, t) ~r (y, t’) then (x, t) fs (-y, t’)

for r < x < y < s, and t < t’.

Example: If a person is indifferent between ($100, one year) and ($150, two years) he prefers (-$100, one year) to (-$150, two years). Changing the polarity of an outcome from positive to negative raises the salience of the outcome dimension. The less aversive outcome becomes relatively more preferable. In overviews of anomalies in intertemporal choice (Loewenstein & Thaler 1989; PL 91; Loewenstein & Prelec 1992; Read & Loewenstein 2000) it is commonly stated that future gains are discounted more strongly than losses. Therefore losses, in fact, seem to loom larger and longer. However, Shelley (1993) argued that this assessment, based on two studies by Thaler (1981) and Benzion et al. (1989), is due to specific experimental designs. Subjects in these studies were told that they had received a receipt or fine.

Parallels between 'anomalies' in risky and intertemporal choice 143 Hence, they were induced to adjust to receive the outcome ‘now’, and were asked subsequently how much they would be willing to pay to delay the payment or charge to postpone the receipt. Thus what is observed in theses studies is an interaction effect of outcome sign (positive or negative) and time-frame (delay or speed-up). In an experiment conducted by Shelley (1993) she was unable to detect a gain/loss asymmetry under a neutral time frame where no reference point was suggested. Despite this mixed evidence, the sign effect is presented in the ‘standard’ way for two reasons. First, the neutral time frame used in the studies above consisted of a question format ‘You owe a debt of $x in time t to a public institute. What is the (negative) value of -$x, of that debt to you now?’ This question does explicitly not invoke a reference point while our regularities refer to situations with a manipulable reference point and therefore are consistent with the evidence. Second, the neutral time frame induces subjects to perform computations instead of asking for assessments of gains and losses.9 b) Range effect on the outcome dimension (gains or losses may appear more or less large) If (x, t) ~r (y, t’) then (x, t) ps (y, t’)

for 0 < x < y and s < r # t < t’

Example: If a person is indifferent between ($1000, 1 week) and ($1500, 52 weeks), he will prefer the latter option when confronted with the following choice problem: In addition to whatever you own you have been given $500. You are now asked to choose between ($500, 1 week) and (1000, 52 weeks). Due to an (apparent ) absolute reduction in size, the outcome distance between options has been relatively enlarged and thus the dimension has become more salient. d) Violations of independence of elicitation methods

Preference reversals (option A rated higher than option B, but higher pricing of option B) were replicated with intertemporal options by Tversky, Slovic and Kahneman (1990). They compared the influence of different elicitation methods on preference between sooner but smaller outcomes and later but larger outcomes. In these experiments on average 74% of the subjects preferred the short-term option but only 25% stated a higher selling price for this option. Tversky et al. explicitly ascribed this finding to scale compatibility of one of the option's attributes with the response mode and ruled out alternative explanations, e.g., violations of transitivity. Roelofsma (1994) reported an analogous effect with respect to the time dimension: subject’s charge $125 to accept a delay of six months for a CD-player worth $450 but are only willing to wait three weeks when offered the same amount of money.10 9

In one of the most famous anomalies in choice under risk Kahneman and Tversky’s Asian disease problem the anomalous preference pattern too is removed when subjects are asked to make statistical computations (Bless, Betsch & Franzen 1998). 10 One may argue that speed-up premiums and delay charges (treated as framing effects on the time dimension in this chapter) are different response modes for the evaluation of options too (Loewenstein 1988). These effects were separated here because the mechanisms causing the anomalous effect differ. In response mode effects the salience of a dimension is raised due to its compatibility with the response mode. The delay/speed-up asymmetry occurs because of different value functions for (relative) gains and losses. Subsuming the latter effect under response mode effect would also allow to call, for example, the common

144

Chapter 6

Regularity for the outcome dimension (response mode may enhance the salience of a dimension): If (x, t) ~r (y, t’) then (x, t) fp (y, t’) response modes rating and pricing respectively.

for x > y, t > t’, r and p describe

Example: If a person rates ($100, 1 year) as equally attractive as ($200, two years) he will state a higher selling price for the latter option. The outcome dimension is more salient in the response mode ‘pricing’ than in the response mode ‘choice’. Regularity for the time dimension (response mode may enhance the salience of a dimension): If (x, t) ~r (y, t’) then (x, t) fd (y, t’) for x < y, t’ < t, r and d describe response modes rating and pricing(of temporal delays) respectively. Example: If a person rates ($100, 1 year) as equally attractive as ($200, two years), he will state a shorter time delay when asked to specify a time delay t’ which makes ($200, t’) equally attractive as ($100, 1 year). Asking for time delays raises the salience of the time dimension, this has the same effect as enlarging temporal distances. e) Reactions to anomalies in intertemporal choice

In intertemporal choice researchers reacted similarly as in choice under risk: in response to violations of certain axioms competing models were developed and subsequently included in axiomatic formulations. For example, the evidence presented in section 3.2 demonstrates that people's choices are inconsistent with the stationarity axiom. Intertemporal choices then were modeled by hyperbolic discount functions instead of exponential ones (e.g., Ainslie 1975) and finally in more recent axiomatic systems the stationarity axiom was replaced by an axiom allowing for hyperbolic discounting (Loewenstein & Prelec 1992, Ahlbrecht & Weber 1995). Up to now models accounting for rank-dependency in intertemporal choice, though suggested by Quiggin and Horowitz (1995), have not been tested. 6.4. Similarities between regularities in risky and intertemporal choice

Above it was demonstrated that very similar regularities occur in risky and intertemporal choice. Common ratio manipulation, common difference manipulation, framing, and variation of elicitation methods induce changes in preference in both domains, independent of whether they are carried out on the time or risk dimension or on the outcome dimension. Consequently, models of risky and intertemporal choice make similar specifications on the discounting function for risk or time and the value function (evaluation of outcomes as a function of their distance from 'zero'). With respect to the value function reference dependence, different slopes of the value function for gains and losses and diminishing sensitivity from the reference point are assumed for choice under risk and intertemporal choice (Tversky & Kahneman 1992, difference effect in intertemporal choice a response mode effect, viz. asking for discount rates for smaller or larger time delays.

Parallels between 'anomalies' in risky and intertemporal choice 145 Loewenstein & Prelec 1992). Thus, the resulting value function apparently has the wellknown S-shaped form.11 With respect to the discounting function diminishing sensitivity is also invoked. In risky choice changes in probability have a larger impact on behavior when they occur near the (two) endpoints of the probability scale. The probability weighting function is concave below approximately .4 and convex above it. Hence, a probability change from 0 to .1 has a larger impact than a change from .3 to .4 and a probability change from .5 to .6 has a smaller impact than a change from .8 to .9. This is consistent with the interpretation that decision weights of a dimension decrease with increasing distance from the reference point. Hyperbolic discounting is compatible with diminishing sensitivity as well, because it implies that changes occurring close to the reference point, ‘immediacy’, have a larger impact than changes occurring further away. In intertemporal choice there exists only one natural reference point (immediacy) as opposed to risk (certainty and impossibility). Contrary to the certainty effect in choice under risk an immediacy effect in intertemporal choice cannot be demonstrated by referring to an analogous not wellbehaved weighting function for temporal delay. Hyperbolic discounting implies disproportionally strong overweighting of immediate as compared to delayed benefits. The discounting function for delayed gains is - though anomalous with regard to the stationarity axiom - well behaved: the closer to immediacy the greater the dislike of delay. However, in some experiments the data was described best by assuming that subjects charged a one-time premium for outcomes of any delay, however small and then discounted them exponentially (Stevenson 1986; Benzion et al. 1989; Shelley 1994). This is interpreted as a risk premium because subjects are assumed to be uncertain about actually receiving the delayed outcome including the utility derived from it. However, Chapman and Elstein (1995) failed to provide evidence for a qualitatively different behavior when outcomes are immediate. In their experiments subjects overweighed immediate benefits but also displayed hyperbolic discounting when only distant, i.e., not immediate outcomes were investigated. There is an analogous discussion in choice under risk, namely, whether gambles are discounted according to normative theory, i.e., linearly, except for irregularities at the endpoints of the probability scale (see Wu & Gonzalez 1996 for a discussion and rejection of this argument). The interpretation that subjects interpret delayed outcomes to be uncertain is consistent with an experiment by Keren and Roelofsma (1995). There, subjects preference for a smaller but immediate, over a larger but more delayed reward was reversed after subsequently introducing different levels of risk. Assume, for example, that a decision maker is indifferent between ($100, now) and ($110, one week), if outcomes are certain. Making both outcomes risky, say with a probability of .9, leads to a preference of ($110, .9, one week) over ($100, .9, now). This preference becomes even stronger when uncertainty increases, e.g., to .5. Keren and Roelofsma (1995) argue that this reveals the incorporation of uncertainty in delay.12 Because only immediate outcomes 11

Tversky and Kahneman (1992) assume a value function of the power family: v (x) = xa for x > 0 and -λ (-xb) for x < 0 , 1 < λ. Such a function cannot account for preference changes due to multiplication of outcomes in two options with a common constant (the common ratio effect with respect to the outcome dimension). To capture this effect Loewenstein and Prelec (1992) assume that the value function is steeper for losses than for gains, more elastic in the negative domain, i.e., it bends over more strongly, and more elastic in the higher regions. 12 Stevenson (1992) obtained similar results (introducing risk leads to lower time discounting), but argued that this effect is due to a distraction of subjects' attention from the temporal dimension. Note, that

146

Chapter 6

are certain, any delay seems to result in time discounting as well as risk discounting. The matter even complicates when losses are introduced. Björkman (1984) speculated that risky losses are discounted heavier than risky gains. Shelley (1994) actually found evidence for this, but remained cautious as to the generality of her results.13 With a different subject pool Ahlbrecht and Weber (1997) under a neutral time frame were unable to find evidence for a gain-loss asymmetry both in cases of risk and certainty. 6.5. Conclusions and discussion

This chapter demonstrated the close relationship between the axiomatic formulations for risky choice and intertemporal choice as well as the very similar behavioral regularities observed in choice experiments in these two domains. It was argued that these, i.e., risky and intertemporal, regularities are due to a general behavioral tendency to discount decision outcomes to the extent that they differ from the decision maker's reference point. This would imply the use of a multi-dimensional reference point, viz. ‘current assets or 'zero', with certainty, now, here, and for myself’. Recurring themes in the abovediscussed empirical choice regularities on both the time/risk and the outcome dimension are: (a) reference dependence, (b) diminishing sensitivity with increasing distance from the reference point, and (c) greater sensitivity to losses. Generalizing the behavioral regularities observed in risky and intertemporal choice to other preference dimensions where distance from a reference point is involved, viz. interspatial and interpersonal choice has several implications. In the following paragraphs, these implications are briefly outlined. An extensive discussion and empirical tests are presented in the following chapter. First, discount functions should be similar across preference-relevant 'distance' dimensions. Small deviations from the reference state ‘here’ or ‘for myself’ should have a relatively higher impact than small changes in options that are already distant. In addition, changing the magnitude or sign of outcomes should also affect preferences in the way as demonstrated above for risky and intertemporal choice. Note, that spatial and social distance, just like temporal distance, are unbound dimensions with a single reference point: 'here' and 'myself', respectively. Their discounting functions thus should mimic intertemporal discounting functions. On the other hand, bounded dimensions, such as perceived controllability (if subjects adapt to either of the reference points 'no control whatsoever' and 'absolute control'), should be discounted similarly to risk. A second implication is that discounting is not restricted to the above-mentioned dimensions. Rachlin and Raineri (1992) provide evidence that discount functions for sound (perceived loudness) and used cars also follow the same pattern. For example, small deviations from the state ‘new car’ have a larger impact on the car's value than some additional use of already used cars. Cars (with a single reference point: 'new car') thus seem to be discounted hyperbolically with increasing distance from the reference point, too. uncertainty about the receipt of delayed outcomes or the future utility provided by them does not imply time discounting. Besides an attitude like ‘the future is uncertain: eat dessert first!’ ‘saving for your rainy days’ is equally justifiable on normative grounds (Olson & Baily 1981). Whether uncertainty leads to patience (negative time preference) or impatience (positive time preference) is an empirical question. 13 Shelley used managers as subjects. Managers are overconfident, if they think that they have some control over the outcome. Shelley hypothesized that managers might discount future losses stronger, because they argue that during the delay a solution will be found.

Parallels between 'anomalies' in risky and intertemporal choice 147 Third, these considerations also hold if the reference point is shifted. Owning a used car that is two years old then is analytically similar to having adjusted to receiving an inheritance with a delay of two years, to having a nuclear power plant created at 200 kilometers distance away or to receiving a video recorder for your neighbor, rather than yourself. Fourth, comparisons between presented and reference options are multiattributive. Therefore, contrary to normative theories, adding an identical third distance dimension to each of two options comprising two attributes, may alter preference between them. Such a third dimension, for example a common risk component added to two intertemporal options (x, t) and (x', t'), changes the overall distance of both objects. More specifically, adding such a third distance dimension has the strongest impact on options that are otherwise close to the reference point. In the example above, adding risk should affect the attractiveness of options that are of little temporal delay more than the attractiveness of options whose consequences are rather delayed. Experiments by Keren and Roelofsma (1995) where preference for a smaller/immediate reward over a larger/delayed reward reversed, when both options were made risky, provide evidence for this. However, Keren and Roelofsma concluded that this effect was due to the close relationship between temporal delay and uncertainty. It is suggested that similar results may be obtained by replacing the risk component with a spatial or social decision component. In general, our preferences vary as a function of the perceived distance of possible outcomes from our manipulable reference points on evaluation dimensions such as outcome sign and size, risk, temporal, spatial, or social proximity. Distant decision consequences are discounted, and we are more sensitive to differences that are close to us than to remote differences.

Chapter 7 Generalizing ‘anomalies’ in choice under risk and intertemporal choice to other preference dimensions

*

Summary

In this chapter some well established ‘anomalies’ observed in risky and intertemporal choice, viz., the common ratio effect, the common difference effect and the sign effect are replicated. In addition, these effects are generalized to interpersonal and interspatial choice as well. In interspatial choice, subjects chose between different locations for shops or supermarkets varying with respect to the number of items-in-store. In interpersonal choice, subjects allocated different amounts of money to several fictive characters being ‘socially distant’ to the decision maker to varying degrees. For choice under risk and intertemporal choice results replicate previous research. For interspatial choice, the respective effects occurred on both the outcome and the spatial dimension except for the common difference effect on the spatial dimension, suggesting that the discount function for spatial distance is exponential instead of hyperbolic. These results demonstrate that it is possible to generalize anomalies observed in choice under risk and intertemporal choice to other preference dimensions. With respect to interpersonal choice, results are much less consistent. While for females none of the predicted effects occurred, for males, after slightly adjusting the stimulus set, common ratio and common distance effects were found on the social dimension, suggesting an hyperbolic discount function. For the outcome dimension only a common difference effect but neither a common ratio effect nor a sign effect could be demonstrated. It remains unclear whether this discrepancy between males and females in interpersonal choice is due to differences in discounting of social distance or whether it is due to differences in the stimulus set, which was not easy to construct. *

Planning of experiments and data collection has been done jointly with Maarten Reinink who also reported in Dutch (see Reinink 2000). I am indebted to Tamas Bartus for introducing me to the statistical program STATA that is used for data analysis.

150

Chapter 7

7.1. Introduction

In the previous chapter, normative concepts and empirical research on choice under risk and intertemporal choice have been summarized and compared. It was demonstrated that normative arguments and empirical violations in both preference domains are highly similar. We argued that this is due to a general cognitive tendency to evaluate the distance of attributes, such as probability or temporal delay, against a certain reference point. This chapter extends the reference point against which options are evaluated. More specifically, it is assumed that people do not only discount options that are risky or delayed, thus being distant from 'for sure' and 'now', but also options whose consequences take place elsewhere and options which occur to others, thus being distant from 'here' and 'myself', respectively. This tendency should lead to identical behavioral regularities in all four preference dimensions, i.e., choice under risk, intertemporal choice, interspatial choice, and interpersonal choice. To support this thesis, in this chapter first the most prominent previous results in risky and intertemporal choice, viz. the common ratio effect, common difference effect and sign effect, are replicated and then these effects are investigated also for interpersonal and interspatial choice. We study the common ratio and common difference effects to the outcome, probability, temporal, personal and spatial attributes, respectively. The sign effect is investigated only for the outcome dimension, because experimental tests of this effect for the respective distance dimension require somewhat more complicated manipulations of the reference point. For instance, Loewenstein (1988) experimentally induced a reference point for temporal delay which is not 'now' but 'in one year' by telling subjects to imagine that they bought a video recorder which could be received only after a certain time delay. Inducing subjects to adjust to a probability level of .7 is even more difficult. The effects are investigated for both positive and negative outcomes, since perceptions of gains and losses might differ. For example, the weighting function for probabilities may depend on whether positive or negative outcomes are involved (e.g., Tversky & Kahneman 1992). In the following, first previous research on spatial and social discounting is briefly reviewed and the effect of the above-mentioned manipulations on preference will be recapitulated from Chapter 6. Then, applications of these effects to spatial and social discounting will be discussed. Then, the subjects and procedure of data gathering are explained and the stimuli presented to subjects are introduced. Here, special emphasis is given to the creation of the stimuli for the interpersonal choice problems, since a quantitative measure for social distance had to be constructed. Results are presented and compared among the various preference dimensions. Finally, we turn again to similarities in choice behavior between different preference domains and discuss implications. 7.2. Spatial and social discounting

Although spatial and social discounting refer to the devaluation of distant outcomes, contrary to the normative models for choice under risk and intertemporal choice no axiomatic formulation exists for them. Hence, it is not possible to provide ‘anomalous’ results with regard to an axiomatic system. However, standard rationality assumptions imply properties such as weakly ordering, transitivity and monotonicity for any set of preferences. An equivalent to the stationarity and reduction of compound lotteries axioms would also be necessary to prevent inconsistencies. A normative formal representation of

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

151

‘spatial preferences’ or spatial utility (SU) for an option A (including several possible outcomes at different locations) has been proposed by Vlek and Keren (1992) in analogy to temporally distributed outcomes: SU (A) = ∑ ∂ s-1 u(xi),

(1)

where ∂ represents a discounting factor for each unit of spatial distance, u(xi) the outcome's local utility ('here') and s the number of spatial distance units. This representation fulfils the normative requirements above, if ∂ is a constant factor over spatial distance units. As was the case for interspatial choice, for interpersonal choice there exist no normative concepts or axiomatic formulations. In addition, social distance is not as easily quantified as the other distance dimensions. While the regularities mentioned above are directly applicable to the outcome dimension of social discounting, applying the regularities to the distance dimension, e.g., doubling social distance, is difficult. Several solutions to the problem of quantifying social distance have been suggested. First, sociology has a long history of social distance scales (Bogardus 1959). These scales measure social acceptance, which is assumed to be the inverse of social distance, of social groups by individuals. To produce such a social distance scale, respondents typically have to rank other individuals or groups on several attributes. In a similar manner, psychologists have developed social comparison matrices. In such matrices scores on several social dimensions in terms of liking or the propensity to share certain activities are weighted and aggregated into an overall score, resulting in a numerical representation of social distance. Economics deals with social distance mainly in the context of intergenerational altruism, frequently modeled by exponential discount functions (e.g., Becker 1991). Interestingly, Laibson (1997) modeled temporal (hyperbolic) discounting by using a formula developed by Phelps and Polak (1968) for intergenerational altruism. Alternatively, Becker (1996) and Akerlof (1997) recently interpreted social distance as either status (raising distance between a group and an individual) or conformism (lowering distance between a group and an individual), but did not suggest quantitative measures When investigating social dilemma situations, psychologists Rachlin and Raineri (1992) suggested another measure for social distance. They argued that outcomes are discounted not only when they occur to others but also when they have to be shared with others. Analogous to intertemporal choice they propose a hyperbolic discount function: a = A/(1 + sN),

(2)

a being the discounted value of the outcome, A its absolute, i.e., undiscounted, value ('for myself'), s a measure for selfishness and N the number of people with whom the outcome is to be shared. In this approach social distance enters the equation in two ways: in the measure for selfishness (i.e., the distance to the people with whom the outcome has to be shared) and in the number of people with whom the outcome has to be shared. This approach is

152

Chapter 7

reasonable when focusing on social dilemmas where the number of players often is critical for determining equilibria but it is less appropriate when investigating how varying social distance of recipients affects the discounting of outcomes. To apply formula (2) to such decision problems, the parameter s should describe the steepness of an individual's discount function for social distance, while N should specify the 'absolute' social distance - equivalent to absolute temporal delay or probability - of the outcome's recipient. In this case, formula (2) describes hyperbolic social discounting. A final possibility for quantifying social distance is provided by allocation problems where preference for outcomes to oneself and to others is elicited. These are problems of the kind ‘what do you prefer: (x for yourself, y for other) or (x’ for yourself, y’ for other)?’ Liebrand and McClintock (1988) suggest the following utility function for such choice problems: U = w1 (outcome for self) + w2 (outcome for others), where both weights are allowed to vary between -1 and 1. (3) Applying the regularities outlined in Chapter 6 then yields predictions about changes in the relative weights depending on the social distance between 'self' and 'other'. In fact, Loewenstein, Thompson and Bazerman (1989) estimated a social utility function for such mere allocation problems. The problem with such allocation problems is twofold: first, subjects place heavy weight on fairness considerations which makes results from e.g., Loewenstein et al. (1989) difficult to interpret. Although in their study changing outcome signs from positive to negative resulted in a 'selfish shift' (raising the salience of the outcome dimension: subjects prefer losses to occur to others) most subjects chose to distribute gains equally, irrespective of the outcome size or the social distance of the other person who receives part of the outcome. A second difficulty of such allocation problems in our context is that the social distance of the outcome's recipients is usually either not varied at all, defined in terms of individual contribution (investment) to the distributed outcome, or in terms of individual need. The approach chosen in this chapter differs from the concepts reviewed above. We argue that people discount objects occurring not to themselves but to others in a way similarly to the discounting of risk or temporal delay. We consider only outcomes which are due to fictive characters varying with respect to their social distance to the decision maker and expect changes in preference, e.g., the common ratio effect or the common difference effect, when the social distance of the outcome's recipients or the size of the outcomes are altered. To do so we created a quantitative measure for social distance that is introduced below. If the problem of quantifying social distance is resolved, a normative representation of social or interpersonal utility (PU) of an option A (including several possible outcomes due to people varying with respect to social distance) analogous to the one for spatial utility can be proposed (Vlek & Keren 1992): PU (A) = ∑ F s-1 u(xi)

(4)

where F represents the discounting factor for social distance, s the number of social distance units and u(xi) the utility of the outcome i for myself. Analogously to the other preferential discounting functions for distance, normatively F should be constant across

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

153

different social relations.105 That is, preference about who receives an outcome should be independent of whether the potential recipients are socially close or distant to the decision maker as long as the social distance between the recipients remains constant across choice situations. 7.3. Effects investigated

In Chapter 6 a summary was given of several parallel regularities in choice under risk and intertemporal choice. In this chapter the focus lies on the common ratio effect, the common difference effect, and the sign effect. The common ratio effect implies that preference between choice options is sensitive to the multiplication of attribute values. Multiplying either the outcome or the distance dimension in two options with a common positive factor raises the salience of this dimension because of increased discriminability of attributes (compare sections 6.3.1 and 6.3.2, first example):106 If (x, x') ~ (y, y') then (x, px') ™ (y, py'), for outcomes: 0 < x < y, and risky (indicating probabilities) or temporal attributes (indicating temporal closeness)107 x' > y', p > 1. If (x, x') ~ (y, y') then (px, x') — (py, y'), for outcomes: 0 < x < y, and risky (indicating probabilities) or temporal attributes (indicating temporal closeness) x' > y', p > 1. The common difference effect implies that preference between choice options is sensitive to additive changes in attributes. Adding a common constant to two options decreases the salience of this dimension (compare sections 6.3.1 and 6.3.2, second example): If (x, x') ~ (y, y') then (x, p + x') — (y, p + y'), for outcomes: 0 < x < y, risky (indicating probabilities) or temporal attributes (indicating temporal closeness) y', p > 0. If (x, x') ~ (y, y') then (x + p, x') ™ (y + p, y'), for outcomes: 0 < x < y, risky (indicating probabilities) or temporal attributes (indicating temporal closeness) y', p > 0.

and x' > and x' >

As to the common difference effect on the probability dimension in choice under risk there is a peculiarity. Since the probability dimension has two reference points, impossibility and certainty, moving away from impossibility when adding a common 105

As Rachlin and Raineri (1992), Vlek and Keren (1992), when suggesting this discount function for interpersonal choice, focused on social dilemmas. They investigated the influence of factors like group size, relation between individual and collective payoffs, the possibility of communication among players etc., on cooperation in dilemma situations. Though these factors can be interpreted as varying social distance, no numerical representations are available. 106 The notation for preference relations between risky or intertemporal options generally follows the one introduced in chapter 6 except that in chapter 6 the temporal attribute equivalent to x' below indicated temporal delay while here it indicates temporal closeness. This is done in order to compare the common ratio and common difference effect for choice under risk and intertemporal choice more easily. In the description of the sign effect as in chapter 6 the temporal attributes t and t' denote temporal delay. 107 Higher values for the temporal attribute therefore do not indicate larger delays as is standard in describing intertemporal options.

154

Chapter 7

constant to two risky options implies approaching certainty. Hence, we expect that adding a common constant to two gambles lowers the salience of the probability dimension when options are evaluated against the reference point 'impossibility' and raises the salience of the probability dimension when options are evaluated against the reference point 'certainty'. This pattern is in accordance with Prospect Theory's weighting function (Kahneman & Tversky 1979) as well as the more recent S-shaped probability weighting function (Tversky & Kahneman 1992). Below the common difference effect will be analyzed separately depending on the reference point against which options are evaluated. This peculiarity is less problematic in case of the common ratio effect on the probability dimension. Multiplying probabilities in both options by a common constant alters the difference between probabilities. Hence, when probabilities close to 'impossibility' are multiplied, this manipulation on the one hand raises the disciminability of the probability attribute because probability differences between options become larger. On the other hand this manipulation lowers the discriminability of the probability attribute because options are shifted away from the reference point. However, if the former effect outweighed the latter, the data has not to be analyzed separately for the two reference points. This point is discussed further in the results section below. The sign effect implies that preference between choice options is sensitive to whether outcomes are perceived as positive or negative. Changing the sign of outcomes in a choice option from positive to negative leads to risk-seeking behavior in choice under risk and to less discounting in intertemporal choice (compare sections 6.3.1 and 6.3.2, third example). If (x, p) ~r (y, p') then (x, p) —s (y, p') If (x, t) ~r (y, t') then (x, t) ™s (y, t')

for outcomes r < x < y < s and probabilities 0 < p' < p ≤ 1. for outcomes r < x < y < s, and temporal delays t < t’.

While for interspatial and interpersonal choice analogous common ratio and common difference effects are expected, with respect to the sign effect predictions are more difficult. Changing the polarity of outcomes is expected to change preference between options, but it is unclear whether loss aversion manifests itself by preferring the smaller but less distant loss (as in intertemporal choice) or by putting losses as far away as possible even at the expense of these losses being larger (as in choice under risk). In the following, the three effects are applied to choice under risk and intertemporal choice in order to replicate results from previous research. In addition, these effects are tested for interpersonal and interspatial choice as well. For all four preference dimensions, subjects are given a selection of a total of 138 binary choice problems. Options consist of an outcome attribute specifying the respective amounts and an attribute specifying the respective probability, temporal delay, spatial distance, or social distance. Both outcome and distance dimensions are varied across choice problems such that the effects mentioned above may manifest themselves. Because interspatial and interpersonal choice have been neglected by decision theorists, the design of these stimuli will be explained in some detail in section 7.5. We will focus especially on the stimuli for interpersonal choice - in section 7.5.1 - since metric measurement units for social distance necessary to investigate the effects mentioned above are not easily quantifiable.

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

155

7.4. Subjects and procedure108 a) Subjects

Eighty subjects participated in the experiment. Subjects were students from various disciplines at the University of Groningen recruited via e-mail and by flyers. Mean age was 22 years, 59% of the subjects were females, and their average monthly income was ƒ 1100. The main fields of study were psychology (27%), management science (9.6%), humanities (10%), sociology (7.5%), and economics/econometrics (7.4%). b) Procedure

The experiment took place via a computer. Subjects gathered in groups ranging from 6 to 15 and were given an introductory text stating that this was an experiment concerning decisions in a variety of settings and that there were no right or wrong answers. Then followed a practice trial for each of the four preference domains before the experiment started. Between presentations of these choice problems were irregular breaks where subjects turned to several other decision scenarios and a questionnaire. The decision scenarios concerned control questions about the perception of the characters used for interpersonal choice as well as some intertemporal choices between consumer goods reported in Chapters 3 and 4. The questionnaire included the personal background of subjects and some questions concerning their health and lifestyle as reported in Chapter 5. These questions as well as the other decision scenarios were given on paper. The experimenter was present during the whole experiment in order to answer questions concerning the tasks. Completing the whole experiment took between 50 and 110 minutes, most people needed between 80 and 90 minutes. Subjects were paid ƒ25 (at that time approximately $12) for their participation. The intermittent questions on paper were introduced to avoid fatigue among the subjects and in order to force subjects not to employ simple decision strategies such as basing the decision on only one attribute. To additionally avoid fatigue, each subject answered only half of the whole set of choice problems (69 out of 138 choice problems). To test for consistency in subjects' answers, 8 of these 69 choice problems were repeated at later times in the experiment. This was also announced in the introductory text. Hence, subjects in total had to make 77 binary decisions in their experimental session. The set of stimuli was varied such that each choice problem was always followed by a problem using a different amount and a different preference domain (i.e., distance variable). For example, a choice between gambles worth ƒ75 and ƒ100 was followed by a choice between intertemporal options offering ƒ150 and ƒ160. To additionally control for order effects, the order of choice problems was randomly varied as well.

108

The experiment was designed and conducted jointly with Maarten Reinink and Charles Vlek (see Reinink 2000).

156

Chapter 7

7.5. Stimuli and method

Stimuli were binary choice problems where subjects had to state their preference for either option A or B. All options consisted of an outcome attribute and an attribute specifying the distance from the respective reference point for each preference dimension (‘for sure’, ‘now’, ‘here’, and ‘for myself’). For each preference dimension, both attributes were varied between stimuli such that the effects mentioned above could occur for the outcome and the distance attribute. For choice under risk 42 choice problems were created including seven different pairs of probabilities (.04 vs. .05; .1 vs. .2; .24 vs. .25; .4 vs. .5; .8 vs. .9; .87 vs. .97; and .8 vs. 1) and six different pairs of amounts (3 positive and 3 negative: ƒ100 vs. ƒ75; ƒ125 vs. ƒ150; ƒ150 vs. ƒ200; same values for the negative amounts). All pairs of amounts were paired with all pairs of probabilities as summarized in Table 7.1. Table 7.1. Overview of choice problems for choice under risk, negative gambles having numerically identical amounts Amounts ƒ100 vs. ƒ75 ƒ125 vs. ƒ150 ƒ150 vs. ƒ200

Probabilities .04 vs. .05

.1 vs. .2

24 vs. .25

.4 vs. .5

.8 vs. .9 .87 vs. .97

.8 vs. 1

Total number of choice problems: 42 (21 positive, 21 negative option pairs) For intertemporal choice 30 choice problems were created with five different pairs of time delays (now vs. 4 weeks; 8 weeks vs. 12 weeks; 16 weeks vs. 24 weeks; 26 weeks vs. 30 weeks; 32 weeks vs. 48 weeks) and six different pairs of amounts: ƒ100 vs. f110; ƒ150 vs. ƒ160; ƒ200 vs. ƒ220, same values for negative amounts. Table 7.2 summarizes choice problems for intertemporal choice. Table 7.2. Overview of choice problems for intertemporal choice, negative gambles having numerically identical amounts. Amounts ƒ100 vs. ƒ110 ƒ150 vs. ƒ160 ƒ200 vs. ƒ220

Temporal delays Now vs. 4 8 weeks vs. 16 weeks vs. 26 weeks vs. 32 weeks vs. 48 weeks 12 weeks 24 weeks 30 weeks weeks

Total number of choice problems: 30 (15 positive, 15 negative option pairs) While in these two domains ample previous research exists, for interspatial choice this is not the case. Because monetary amounts to be obtained at varying spatial distance did not seem to be a particularly realistic scenario, choices between supermarkets and grocery stores - varying in the number of offered products - to be built or removed at different locations were used. Thirty interspatial choice problems were created with 5 different pairs of spatial distances (here vs. 2 streets away; 6 streets away vs. 8 streets away; 12 streets away vs. 16 streets away, 16 streets away vs. 18 streets away, 24 streets away vs. 32 streets away) and three different pairs of items-in-store (a small grocery store with 150 items vs. a small shop with 500 items; a medium shop with 1150 items vs. a

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

157

somewhat larger shop with 1500 items; a large shop with 1500 items vs. a supermarket with 5000 items; the same number of items for removed stores). Table 7.3 summarizes choice problems for interspatial choice. Table 7.3. Overview of choice problems for interspatial choice, negative gambles having numerically identical amounts. Amounts Spatial distances here vs. 6 streets vs. 12 streets vs. 16 streets vs. 24 streets vs. 150 items vs. 500 items streets streets 32 streets 18 streets 16 1150 items vs. 1500 items 2 streets 8 away away away away 1500 items vs. 5000 items away Total number of choice problems: 30 (15 positive, 15 negative option pairs)

For interpersonal choice 36 choice problems were created. Again six pairs of monetary amounts in the positive and negative domain were used (ƒ100 vs. ƒ150; ƒ150 vs. ƒ200; ƒ200 vs. ƒ300; same values for negative amounts) and six combinations of persons varying in social distance who could either gain or lose these amounts (see below for a detailed account of how social distance and these combinations were created). This design allows one to compare the manipulations occurring at each ‘distance’ dimension (probability, temporal delay, spatial distance, and social distance) between six different pairs of monetary amounts, since manipulations are identical for each pair of amount. It further allows one to compare the manipulations occurring at the outcome dimension between several values on the distance dimension, since for each value pair on the ‘distance’ dimension there are six pairs of amounts available. For example, in intertemporal choice a common difference effect on the temporal dimension is expected when holding the outcome dimension constant, e.g., ƒ100 vs. ƒ110, and comparing temporal delays, e.g., ‘now vs. 4 weeks’ and ‘8 weeks vs. 12 weeks’. Preference for the later choice is expected to be stronger in the second problem. A common ratio effect on the temporal dimension is expected when the outcome dimension is held constant and choice problems with temporal delays ‘8 weeks vs. 12 weeks’ and ’16 weeks vs. 24 weeks’ are compared. Preference for the later reward is to be expected to be higher in the first problem. For the common difference effect on the outcome dimension, temporal delays between choice problems are held constant, e.g., ‘now vs. 4 weeks’. Preference for the later option is expected to be weaker when the involved amounts change from ‘ƒ100 vs. ƒ110’ to ‘ƒ150 to ƒ160’. Likewise, a common ratio effect on the outcome dimension involves choices between amounts ‘ƒ100 and ƒ110’ and ‘ƒ200 vs. ƒ220’, the temporal attribute being kept constant. Preference for the option with the larger outcome is expected to be stronger in the second problem. The sign effect involves choices between amounts ‘ƒ100 and ƒ110’ and ‘-ƒ100 and -ƒ110’, again the temporal attributes being kept constant. The proposed effects for the remaining preference domains are investigated analogously. Note, that for each effect there are several comparisons possible within a stimulus set. For example, in intertemporal choice a sign effect is expected to occur when comparing amounts in choice problems ‘ƒ100 and ƒ110’ and ‘-ƒ100 and -ƒ110’, ‘ƒ150 and ƒ160’ and ‘-ƒ150 and -ƒ160’, and ‘ƒ200 and ƒ220’ and ‘-ƒ200 and –ƒ220’ with temporal delays being held

158

Chapter 7

constant between choice problems. Table 7.4 exemplifies for intertemporal choice how effects are investigated but note that the list of comparisons is not exhaustive. Table 7.4. Overview of effects for intertemporal choice (examples) Amounts

Delays

1) 100 vs. 110

Now vs. 4 weeks

2) 100 vs. 110

8 weeks vs. 12 weeks

3) 100 vs. 110

16 weeks vs. 24 weeks

4) 150 vs. 160 5) 150 vs. 160 6) 150 vs. 160 7) 200 vs. 220 8) 200 vs. 220 9) 200 vs. 220 10) -100 vs. –110 11) –100 vs. –110 12) -100 vs. -110

Now vs. 4 weeks 8 weeks vs. 12 weeks 16 weeks vs. 24 weeks Now vs. 4 weeks 8 weeks vs. 12 weeks 16 weeks vs. 24 weeks Now vs. 4 weeks 8 weeks vs. 12 weeks 16 weeks vs. 24 weeks

Comparisons and effects With 2): Common difference on time dimension With 4): Common difference on outcome dimension With 7): Common ratio on outcome dimension With 10) Sign effect With 3): Common ratio on time dimension With 5): Common difference on outcome dimension With 8): Common ratio on outcome dimension With 11) Sign effect With 6): Common difference on outcome dimension With 9): Common ratio on outcome dimension With 12) Sign effect With 5): Common difference on time dimension With 6): Common ratio on time dimension With 8) Common difference on time dimension With 9): Common difference on time dimension With 11) Common difference on time dimension With 12) Common ratio effect on time dimension

In total thus 138 choice problems were created for all four preference dimensions of which subjects had to answer 69 in order to avoid fatigue. Choice problems were presented to subjects on a computer displaying the attribute values and in most cases also an additional text to make the scenarios more vivid. For example, in intertemporal choice for negative outcomes, subjects faced the following screen (in Dutch):

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

159

Imagine you have to choose between the following two options: A) A loss of ƒ100 in 8 weeks. You have to pay ƒ100 in 8 weeks B) A loss of ƒ110 in 12 weeks. You have to pay ƒ110 in 12 weeks Which option do you prefer _______ Confirm:

_______

For positive outcomes, the respective text was 'you receive x guilders in time t'. Because this situation seemed to be more natural, no additional explanation was given. For choice under risk, the respective screen was (in Dutch, negative outcomes): Imagine you have to choose between the following two gambles where you can only lose money: A) You have a probability of .8 (80%) to lose ƒ150. You lose ƒ150 if a blue marble is drawn from an urn containing 80 blue and 20 white marbles. B) You have a probability of 1.0 to lose ƒ125. You lose ƒ125 for sure. Which gamble do you prefer

_______

Confirm:

_______

For positive gambles text and explanation were given analogously. The example with the marbles was given to make the choice more vivid.109 For interspatial choice, subjects had to answer the following problem (in Dutch, positive outcomes):

109

Providing subjects with such an additional text is by no means trivial. Generally, subjects' evaluation of probabilities is much better, i.e., less anomalous, when choices are presented using frequencies (Gigerenzer & Hoffrage 1995). Hence, when attempting to demonstrate similar anomalies across preference dimensions such an additional example may reduce the strength of effects in choice under risk and lead to more conservative results.

160

Chapter 7

'Imagine in your neighborhood there are no stores. Now there are two plans to build a store in your neighborhood: A) A small grocery store with 150 items-in-store 12 streets away from your home. B) A small store with 500 items-in-store 16 streets away from your home. Which plan do you prefer

_______

Confirm:

_______

For negative outcomes the respective initial text was: 'Imagine there are two stores in your neighborhood. Now there are two plans to close one of the two stores. A) A small grocery store with 150 items-in-store 12 streets away from your home will be closed. B) A small store with 500 items-in-store 16 streets away from your home will be closed. Which plan do you prefer?' For each choice problem subjects had to confirm their preference. In case the two answers did not match, the subjects had to redo the whole choice problem, i.e., they faced the same text on the screen again. For interpersonal choice, the following section explains the creation of stimuli in detail. 7.5.1. The interpersonal stimuli

To demonstrate the effects on the social dimension in interpersonal choice a metric measure for social distance is necessary. Since there exist only few measures for social distance which, moreover, use only ordinal scales (e.g., Bogardus 1959), fictive characters were created defined by five dimensions: age, social-economic status (SES), hobbies, nationality, and political orientation. These dimensions were chosen because they were considered to be important, though not exhaustive, for defining social distance. Then weights were given to each of these dimensions and overall social distance scores were created depending on how much each character differed from the ‘average subject’ for all five characteristics. As 'average subjects' we considered 24-year old, Dutch students who are politically oriented left-wing and whose favorite activity is reading (see below for empirical validity of this assessment). To determine the relative importance of these five dimensions a pretest was conducted. Twenty-five subjects (students and Ph.D. students) rank-ordered the five dimensions according to their importance for three hypothetical situations: 'considering a new neighbor', 'considering a new colleague', and 'considering a friendship relation'. In addition to rank-ordering the five dimensions, subjects also had to distribute 100 points between the dimensions, i.e., to weight them, according to their importance. Three situations were used since social distance and the importance of the five dimensions is likely to vary across situations. Using several situations then should result in a more robust measure. In fact the rank-ordering and the weights of the five dimensions differed between the three situations. A Friedman-test revealed differences in rank-ordering between the three situations for SES (χ2: 18.2; p < .001), hobbies (χ2: 15.4, p < .001), and political

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

161

orientation (χ2: 8.9; p < .05). For age and nationality differences in rank-ordering were marginally significant (χ2: 5.7; p = .057 and χ2: 4.77; p = .092). The weights between the three situations differed for nationality (F: 5.9; p < .01), SES (F: 20.0; p < .01), hobbies (F: 4.5; p < .05), and political orientation (F: 3.7; p < .05). For age no differences in the weights between the three situations occurred (F: .96; n.s.). On average across the situations, hobbies were ranked the most important dimension, age and SES were jointly ranked second and third, political orientation was ranked fourth and nationality was ranked as least important. As to weights, hobbies received the strongest weight (30 out of 100 points), age the second strongest (25 points), SES third (20), political orientation fourth (15), and nationality least (10). Thus, weights and ranking resulted in almost identical orders. Now these estimates were used to create six fictive characters more or less different from our ‘average subject’. For each of the five dimensions the fictive characters could fall in one of the following three categories (c): ‘identical’ to our ‘average subject’, ‘somewhat distant’, or ‘very distant’. Each of these categories has a certain weight (w) with which it contributes to the social distance of a character. If the fictive character was defined as being 'identical' it received the full weight of this dimension, if it was defined as ‘somewhat distant' it received half the weight of the respective dimension, and in case it was defined as ‘very distant’ it received a weight of zero. For example, for the dimension ‘age’ there are three categories: 24 years, 40 years, and 55 years. Age received an average weight of 25 points in the pretest. Hence, a fictive character received 25 points (full weight) when he or she was described as 24 years old, 12 points (half weight) when described as 40 years old, and zero points (zero weight) when described as 55 years old. For the other dimensions the procedure was carried out accordingly, resulting in several degrees of social distance varying from 0100, the latter defining the closest character and the former defining the most distant character. The social distance (SD) of each character i then is defined as the sum of the weighed social distance scores (w) for each of the five dimensions (d): 3

SDi = ∑ wd (cd ) 1

for categories (c): 'identical' ('full weight': 1), 'somewhat distant' (0.5), and 'very distant' (0); and d: Age, SES, nationality, hobbies, and political orientation Characters differed for male and female subjects, male subjects being presented with male characters and female subjects being presented with female characters. This was done in order to avoid the possibility that men and women evaluate the same characters differently. There is some evidence concerning interaction effects on social distance of sex and nationality as well as of sex and age (Johnson and Marini 1998). In the description of the characters it was explicitly stated that income is identical for all of the characters to avoid that subjects base their decision on considerations concerning the utility of the respective amounts for the characters. Table 7.5 summarizes dimensions, categories and weights for both sexes. In case of different categories for both sexes, males (m) and females (f) are indicated between brackets.

162

Chapter 7

Table 7.5 Summary of dimensions (ordered from least to most important), categories (most distant to least distant), and weights (least to most important) Dimension 1 Nationality 2

Political orientation

3

SES

1 2 1 2 3 1

Age

2 3 1

4

5

Hobbies

Category French Dutch Right Middle Left Worker in a steel factory (m) Cleaning lady (f) Employee at an administration Student 55 years

2 40 years 3 24 years 1 Playing bingo in the neighborhood 2 Soccer (m) Shopping with friends (f) 3 Reading

Weight 0 10 0 7 15 0 0 10 20 0 12 25 0 15 15 30

For each character, a score of the sum of points can be calculated ranging from 0 to 100, 0 being the most distinct character (55 years old, working in a steel factory or as a cleaning lady, French, likes to play bingo in the neighborhood and politically right-wing oriented) and 100 (24 years old, student, Dutch, likes reading, and politically left-wing oriented) being the socially closest character, i.e., the average subject. Six characters for each sex were created with scores of 100, 77, 39, 25, 10, and 0 respectively. The scores of 77, 39, and 25 were rounded to 80, 40, and 20 respectively, in order to create the necessary common ratio and common difference effects on the social dimension. Table 7.6 summarizes characters with the accompanying categories and scores.

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

163

Table 7.6 Category weights and social distance scores for each fictive character Character (males first) A Jean-Marie Marie-Claire B Pièrre Louise C Ben Hennie

D Leo Hanneke E Willem Anne

F Pim Merel

Weight per category 55 Years 0 Worker in a steel factory (m) 0 Cleaning lady (f) 0 French 0 Playing bingo 0 Right-wing 0 55 Years 0 Employee at an administration 10 French 0 Playing bingo 0 Right-wing 0 55 Years 0 Worker in a steel factory (m) 0 Cleaning lady (f) 0 Dutch 10 Soccer (m) 15 Shopping with friends (f) Right-wing 0 40 Years 12 Employee at an administration 10 Dutch 10 Playing bingo 0 Middle 7 24 Years 25 Student 20 Dutch 10 Soccer (m) 15 Shopping with friends (f) 15 Middle 7 24 Years 25 Student 20 Dutch 10 Reading 30 Left-wing 15

Category 1 Age 2 SES 3 4 5 1 2 3 4 5 1 2

Nationality Hobbies Pol. orientation Age SES Nationality Hobbies Pol. orientation Age SES

3 Nationality 4 Hobbies 5 1 2 3 4 5 1 2 3 4

Pol. orientation Age SES Nationality Hobbies Pol. Orientation Age SES Nationality Hobbies

5 1 2 3 4 5

Pol. Orientation Age SES Nationality Hobbies Pol. Orientation

Social distance score

0 Points

10 Points

25 ⇒ 20 Points

39 ⇒ 40 Points

77 ⇒ 80 Points

100 Points

For example, the fictive character D presented to males (’Leo’) is defined as ’40 years old’ (half the weight of the category ‘age’ = 12 points), ‘employed at an administration (half the weight of the category ‘SES’ 0 = 10 points), Dutch (full weight of category ‘age’ = 10 points), ‘playing bingo (zero weight for category ‘hobbies’ = 0 points), and ‘middle’ (half the weight for category ‘political orientation’ = 7 points), resulting in an overall social distance score of 39 points, rounded to 40 in order to arrive at values suited for the common ratio and common difference effects under study. In the choice problems presented, subjects were given the following pairs of characters presented below in Table 7.7. To make the decisions between different

164

Chapter 7

amounts of money due to different characters more vivid, each character's description was supplemented by a photograph. Table 7.7. Pairs of characters and effects for the social dimension. Higher values indicate less social distance. Effects for the outcome dimension are created analogous to the other preference dimensions, with negative outcome having identical amounts Amounts

Characters 1) Pim/Merel vs. Willem/Anne

(ƒ100 vs. ƒ150)

Leo/Hanneke vs. Ben/Hennie 3) Willem/Anne vs. Leo/Hanneke (ƒ200 vs. ƒ300) 4) Ben/Hennie vs. Jean-Marie/Marie-Claire 5) Ben/Hennie vs. Pièrre/Louise (ƒ150 vs. ƒ200)

2)

Social distance Comparisons and effects for the scores social dimension 100 vs. 80 With 2): Common difference effect With 4): Common difference effect With 3): Common ratio effect 40 vs.20 With 4): Common difference effect With 5): Common ratio effect 80 vs. 40 With 5): Common ratio effect 20 vs. 0 20 vs.10

Subjects answered the following problem on the computer screen (in Dutch, positive outcomes, males): 'Please read the descriptions of Willem and Leo given in the appendix carefully. Imagine the following two events: A) Willem loses ƒ150. C) Leo loses ƒ200. Which event would you regard to be less harmful? _______ Confirm:

_______

For females, positive outcomes and other amounts the text was varied correspondingly. As for the other preference dimensions, subjects had to confirm their answers. Because the numerical creation of social distance is far from perfect and in order to control for possible distortions due to added information in the photographs, several control questions in the paper-and-pencil part of the experiments were included. First, subjects were asked about their age, nationality, hobbies, and political orientation to check whether they matched the 'average subject'. For example, subjects whose main hobby is to play bingo in the neighborhood are likely to perceive the social distance of the characters differently than intended. SES of the respondents was not elicited, since only students served as subjects. Second, respondents were asked to rate the social distance of each of the characters on a scale. The question was ‘could you please indicate

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

165

on the following scale how similar each of the following characters is to yourself.’ The aim of this question is twofold. First, to have a direct test whether the created social distances are in fact perceived as intended. Second, in case social distance was misspecified, the average empirical values can still be used to create new effects for the social dimension.110 7.6. Data analysis and control variables

a) Data analysis Eighty subjects each completed 69 choice problems, i.e., half of the complete stimulus set and 8 consistency checks not used for further analysis (see Section 7.4). Recall that for each preference dimension there was a certain number of choice problems, 42 for choice under risk, 30 for intertemporal choice, 30 for interspatial choice, and 36 for interpersonal choice. Each choice problem now is defined by a certain pair of monetary amounts, a certain pair of distances: probabilities, temporal delays, spatial distances or social distances, and a sign: positive or negative, of the outcome. To analyze the data all binary choice responses of the 80 subjects for each preference dimension were aggregated. That is, each subject contributes 69 real choices, 69 'missing values', and 8 consistency checks. For each specific choice problem approximately forty valid responses exist. This procedure resulted in 3360 responses for choice under risk, 2400 responses for intertemporal choice, 2400 responses for interspatial choice, and 2880 responses for interpersonal choice. Because each subject answered only half of the stimulus set, half of these responses could be analyzed and the other half were missing values. Since the dependent variable was dichotomous (choice of either option A or B) the data were analyzed by running logistic regressions analyses (see Appendix B at the end of this book for details) for all preference dimensions and effects. To analyze positive and negative outcomes jointly, preference among the negative-outcome options had to be recoded. For example, in an intertemporal choice between gains subjects can either weight the temporal attribute or the outcome attribute more strongly. Hence, they either prefer the smaller but sooner option ('A') or the larger and later option ('B'). For negative outcomes the same weighting of attributes results in a numerically different preference between options. If the temporal attribute is more salient, subjects will prefer the later but larger loss (B). Therefore, preference for negative options was recoded such that 'A' consistently described a larger salience of one attribute in case of both positive and negative outcomes. As explained above, within each preference domain only part of the choice problems was used to create the respective effects (see Tables 7.1-7.3 and 7.7) and therefore was selected for the respective analysis. For example, to investigate whether in intertemporal choice there is a common difference effect on the temporal dimension, only choice problems were selected where the values for temporal delay where shifted into the future by a common constant (see Table 7.2). If the variable 'temporal delay' contributes to the fit of the model, then preference between options offering two different outcomes 110

In addition, several other questions on the perceptions of the six characters were asked, e.g., how subjects rate the characters on the 'Big Five' personality inventory. Since these questions are of minor importance here, results are not reported in more detail. Reinink (2000) provides a detailed report of them.

166

Chapter 7

either 2 weeks sooner or later is sensitive to whether the sooner option is available now or in 16 weeks. Likewise, when investigating the common ratio effect for the outcome dimension only choice problems are selected where the available outcomes are related multiplicatively, i.e., either (ƒ100 vs. ƒ110) or (ƒ200 vs. ƒ220). If the variable 'outcome' improves model fit, then there is evidence for a common ratio effect, because preference is sensitive to whether options available at the same time offer smaller or larger outcomes. The tests for predicted effects with respect to the other effects and preference dimensions work analogously. Note, that by aggregating data in such a way observations are treated as being independent while in fact observations are clustered: each subject responded to several choice problems and contributed several observations for analysis, thus forming a special cluster. This aggregation is problematic, because it artificially raises the number of observations. While only eighty subjects contributed data for the analysis, the logistic regression model assumes that, say, all 3360 responses for choice under risk are independent. In other words, between-subjects variance in responses, e.g., due to the respective experimental manipulations, is confounded with within-subjects variance due to individual differences in risk-aversion or time preference. Consequently, such an aggregation of observations reduces standard error estimates and in turn affects significance levels based on these standard errors. Hence, effects might appear unduly strong.111 To control for this possible bias, a robust variance estimator (White 1980) available in the statistical program STATA 6.1 was used, thus controlling for clustered observations as is the case when each individual contributes several observations to the analysis. This robust variance estimator can be applied even in case error variances differ between clusters, i.e., when individual i is consistently more risk-averse, more patient etc. than individual j. The estimator adjusts the standard errors on which significance levels are based, thereby leading to more conservative estimates.112 For the analysis of the common difference effect on the distance dimension in choice under risk the data were split in order to analyze preferences depending on the reference point against which options are evaluated. If the safer option had a probability of .5 or less, options are assumed to be evaluated against the reference point 'impossibility', if the riskier option had probability of more than .6, options are assumed 111

One might also argue that each person has an individual discount rate for risk, temporal delay etc. Therefore an individual who prefers (ƒ100, now) to (ƒ110, 4 weeks) is more likely to prefer (ƒ100, 8 weeks) to (ƒ110, 12 weeks) than an independent randomly chosen other subject. Hence, to be successful, the manipulation, e.g., the common difference effect on the temporal dimension, has to overcome this natural ‘inertia’ of subjects, thereby making results more conservative. 112 Of course, the problems could also be tackled by using multi-level analysis. Such analyses sort out the variance for each level, e.g., between and within subjects as in our context. Because our interest lies not in how much variance there is at the individual level (within subjects), but only in exclusion of interference of individual dispositions with the respective effects, multi-level logistic regression analysis is an unnecessary complex tool. Suppose for example, a multi-level logistic regression analysis would reveal that individual dispositions have an equally strong effect to predict preference as the respective manipulations. This relation between within-subjects and between-subjects variance could easily be altered by using stronger experimental manipulations. For example, to demonstrate the common ratio effect on the outcome dimension in intertemporal choice in this study the respective monetary amounts are doubled. By multiplying the initial amounts by four or eight one would also alter the relation between within-subjects and between-subjects variance. Hence, using the robust variance estimator provides a much more simple means to analyze the data.

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

167

to be evaluated against the reference point 'certainty'. All options were created such that they clearly fell in either category (see above). b) Control variables Several control variables were included in the analysis to ensure that possible effects would not be distorted by intervening variables. These control variables were sex, the number of inconsistencies occurring at the specific preference dimension under study, the number of inconsistencies occurring for all four preference dimensions, the different monetary amounts in each preference dimension in the analysis for the effects on the distance dimension, the respective distances in the analysis of the effects on the outcome dimension, and interaction effects for amount * distance, amount * sign, and sign * distance.113 All of these variables were included for all four preference dimensions. Sex was included because men and women faced different stimuli in interpersonal choice and sexes sometimes have been found to behave differently in intertemporal choice and choice under risk (e.g., Loewenstein & Sicherman 1991. In addition, women might be more altruistic than men (see Eckel & Grossman for evidence concerning dictator games). The two variables concerning inconsistencies were included to control for subjects' different motivation or experience with such decision tasks. They were added as control variables instead of removing the relevant subjects from the sample, because we are only interested in possible distortions on the respective manipulation. In case there are no differences between consistent and inconsistent subjects it seemed unnecessary to exclude subjects from further analysis.114 The variables concerning the interactions were included to ensure that the effects of the respective manipulation on preference are insensitive to the specific amount, distance, or sign. By aggregating all preference data, it was argued that the predicted effects would occur irrespective of the specific amount or sign. In addition, controlling for these variables served as an additional manipulation check for interspatial choice. In the other three preference dimensions, simple monetary outcomes were used while in interspatial choice outcomes consisted of shops of different size (see above). While there is ample research on utility functions for money (e.g., Fennema & van Assen 2000), analogous research for utility functions for shops varying with respect to items-in-store does not exist. Effects of shop size or its interactions in interspatial choice would suggest that the utility function for shop size differs from the utility function for money. Further, two variables were created measuring the respective differences between the options in a choice problem with respect to outcome and distance. For example, for the risky choice problem between the options (100; .04) and (75; .05) the variable 'difference between amounts' receives the value '25' and the variable 'difference between distances (probabilities)' receives the value '.01'. This was necessary to be able to select

113 In fact, most analyses presented below were carried out twice, with and without a robust variance estimator. Results for the respective effects are by and large similar but are somewhat more in accordance with the predictions using the robust variance estimator. However, using the robust variance estimator the control variables reach significance less often. This makes sense, since, say the sex of a subject, in a conventional logistic regression analysis contributes several times independently to the prediction of the subject's preference, thus overestimating its influence. 114 Several analyses were also run after excluding subjects revealing one or more inconsistencies. This did not affect results.

168

Chapter 7

the appropriate choice problems for analysis.115 Finally, in choice under risk a dummy variable was included indicating whether the certainty effect might occur or not. Whenever the probability of the safer gamble was greater than or equal to .97 the dummy was coded 1 (0 otherwise). Below, several logistic regression models are presented. First, all control variables were entered together with the effect under study ('complete model'). Then control variables that did not improve model fit where subsequently removed. Removal continued until the effect under study plus the control variables improving model fit were left ('intermediate model'). Finally, a 'simple model' with only the manipulation under study is presented. All three models are presented in order to better compare the explanatory value of the respective manipulation as compared to the 'complete model'. In addition, for some analyses the sign of the respective coefficients is sensitive to the covariates added. This can be reported most easily by presenting all three models below. In the text below only logistic regression coefficients will be reported, preference distributions for choice problems will be presented in detail only if the respective regression coefficient differs from the prediction, otherwise only selected preference distributions will be presented in footnotes to make the reading more smoothly. A complete preference distribution between choice problems for each choice problem and preference dimension is presented in Appendix A at the end of this chapter. Table 7.8 recapitulates and summarizes predictions for the respective regression coefficients. These coefficients express whether a variable raises or lowers the likelihood to prefer the option coded as 'A' (negative coefficient) or the option coded as 'B' (positive coefficient). For the common difference effect on the probability dimension in choice under risk the effect of this manipulation depends on whether it is carried out close to the reference point 'impossibility' or to the reference point 'certainty' (see above). In the former case, a common difference manipulation results in a larger distance of both options from the reference point and salience of the probability dimension decreases. In the latter case, the common difference manipulation results in a move towards the reference point 'certainty' and increases the weight of the probability dimension.

115

These variables turned out to be very good predictors for choice behavior. In choice under risk assuming that subjects evaluate only differences in risk accounts for more than 21.5% of the variance (χ2: 261.46; z: 16.17). For intertemporal choice a model with 'temporal delay' and 'differences in temporal delay' as independent variables accounts for 1.2% of the variance (χ2: 17.44; z: 3.85 and -4.84). For interspatial choice a model with 'differences in spatial distance' as the only independent variable accounted for 3.2% of the variance (χ2: 38.69, z: -6.22). For interpersonal choice a model with only 'differences in social distance' as an independent variable accounted for 1.8% of the variance (χ2: 18.86, z: -4.34). All these values are significant at p < .001. This is interesting since there is an ongoing discussion (e.g., Roelofsma & Read 2000) of whether multiattribute choice are alternative-based - subjects evaluate all attributes and integrate them into a single value as in the DU-model - or attribute-based - subjects compare options attribute by attribute. If a single variable accounts for such a large proportion of variance, this seems to favor the latter interpretation.

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

169

Predicted logistic regression coefficients for all four preference dimensions. For each preference dimension the less distant option is underlined. For interpersonal choice larger values mean less distance while for intertemporal and interspatial choice this is reversed. Hence, positive regression coefficients can imply both increasing and decreasing salience of the distance dimension. Note further that in choice under risk loss aversion results in risk-seeking behavior reducing the salience of the outcome dimension, while for intertemporal choice this is reversed.

Table 7.8 Preference dimension: Basic choice problem:

Choice under risk A: (100, .04) vs. B: (75, .05)

Intertemporal choice A: 100 now vs. B: 110 in 4 weeks

Effects: Negative coefficient: Positive coefficient: Common difference on Salienceoutcome ↑ Salienceoutcome ↑ distance Preference A ↑116 Preference A ↑

Interspatial choice

Interpersonal

A: 150 items here vs. choice B: 500 items 2 streets A: 100 for 80 vs. away B: 150 for 40 Positive coefficient: Salienceoutcome ↑ Preference B ↑

Negative coefficient: Salienceoutcome ↑ Preference A ↑

Common Positive coefficient: difference on Salienceoutcome ↓ outcome Preference B ↑

Negative coefficient: Negative coefficient: Salienceoutcome ↓ Salienceoutcome ↓ Preference A ↑ Preference A ↑

Positive coefficient: Salienceoutcome ↓ Preference B ↑

Common ratio on distance

Negative coefficient: Negative coefficient: Salienceoutcome ↓ Salienceoutcome ↓ Preference A ↑ Preference A ↑

Positive coefficient: Salienceoutcome ↓ Preference A ↑

Positive coefficient: Salienceoutcome ↓ Preference B ↑

Negative coefficient: Positive coefficient: Common ratio Salienceoutcome ↑ Salienceoutcome ↑ on outcome Preference A ↑ Preference B ↑ Positive coefficient: Sign effect on Salienceoutcome ↓ outcome 117 (losses vs. gains) Preference A ↑ Risk-seeking

Positive coefficient: Salienceoutcome ↑ Preference A ↑

Positive coefficient: Salienceoutcome ↑ Preference B ↑

Negative coefficient: Salienceoutcome ↑ Preference B ↑

Coefficient: unclear Salienceoutcome: unclear Preference A: unclear (see section 7.3.)

Coefficient: unclear Salienceoutcome: unclear Preference A: unclear (see section 7.3.)

7.7. Results

In this section, first, results with respect to individual consistency are reported. Then we turn to the effectiveness of manipulations for social distance, i.e., whether social distance of the characters was perceived as intended, and finally, results for the respective outcome and distance effects are reported.

116

This prediction refers to changes occurring close to 'impossibility'. For manipulations occurring close to 'certainty' the coefficient reverses, since the probability dimension becomes more salient. 117 Note that values have been recoded. Without recoding one would expect a negative coefficient for choice under risk: risk-seeking for losses implies increased preference for the larger loss. Since negative options have been recoded an increase in the preference for option A here results in a positive coefficient. Likewise in intertemporal choice changing the polarity of options should increase preference for option A (in intertemporal choice the smaller option is defined as 'A') and hence the expected coefficient is positive.

170

Chapter 7

7.7.1 Consistency check for individual answers

On the whole, results of the consistency check were mixed. Since for each of the four distance dimensions there were two control questions, subjects in total could reveal between 0 and 8 inconsistencies. Most subjects (95%) revealed between 0 and 3 mistakes.118 Subjects behaved most inconsistently in interpersonal and interspatial choice: here only 59% and 58%, respectively, responded to both problems consistently. For intertemporal choice and choice under risk, percentages of subjects who gave identical answers were much higher: 77.5% and 69% respectively. The percentage of subjects who made two inconsistent choices was roughly similar across all four preference dimensions, varying between 4% for intertemporal and interpersonal choice and 9% for interspatial choice.119 7.7.2 Manipulation check for social distance

Our prior ideas concerning the ‘average subject’ were by and large correct. Age and SES were correct given that it was university students who served as subjects. Nationality was correct as well, since 95% of the subjects were Dutch. For both sexes a relative majority (females 50%, males 44%) were politically oriented ‘left’ or ‘slightly left’. With respect to hobbies 41% (males) and 44% (females) stated ‘reading’ as one of their favorite hobbies. However, ‘music’ (both making and listening) is more popular among males (50%) and ‘sport’ is more popular among females (61%). These results do not affect social distance of the characters, because neither 'sport' nor 'music' was used to create the fictive characters (see table 7.5.). As to the rating of the characters by the respondents on a ten-point scale for social distance it turned out that women rated the female characters as intended: all six characters had the predicted rank-order. A paired t-test revealed differences in social distance for five of the six intended comparisons, only the comparison in social distance between Hennie and Louise with intended social distances of 20 and 10 respectively was insignificant (t: .77; n.s; all other comparisons significant at p < .001). Hence, the manipulations appear to have worked reasonably well. The closest three male characters, Pim, Willem, and Leo with respective social distances of 100, 80 and 40, respectively, had the predicted position, while the latter three characters were all rated differently from what we intended. Moreover, in two comparisons, stimulus 1 (Pim vs. Willem; t: .46; n.s.) and stimulus 5 (Ben vs. JeanMarie; t: .81; n.s.) there are no differences in perceived social distance. For all other comparisons, perception of social distance differed between characters at p < .05. Because it was argued that raising social distance close to the reference point, ‘myself`, has a stronger impact on preference than introducing social distance at later stages, stimulus 1 did not work as intended. Since the characters were perceived as being identically distant, stimulus 1 can not be used for the demonstration of common ratio and common difference effects on the distance dimension and was removed from all analyses. Stimulus 5 was kept, because the absence of differences in social distance between these

118

Exact values are: 0 mistakes: 23.8%, 1 mistake: 27.5%, 2 mistakes: 22.5%, 3 mistakes: 21.3%. Some of the analyses below were replicated without the subjects giving two or more wrong answers for each consistency check. Results were absolutely not affected by the removal of these subjects. 119

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

171

two characters might be due to insensitivity to social distance at large distances comparable to perceived differences between temporal delays 335 and 345 days. Since stimuli were different for males and females in addition to the joint analysis for both sexes presented below, all analyses were carried out for males and females apart. Whenever results of these separate analyses differed from the results for men and women combined, this is mentioned in the text. 7.7.3. Common difference effects

a) Model fit Generally model fit of the complete models was very good for all four preference dimensions (see Table 7.9 below where all results and coefficients concerning the common difference effect are summarized jointly). In each choice domain the complete models are significant at p < .001, the only exception being the common difference effect on the outcome dimension in interpersonal choice (significant at p < .01). This is taken to be an indicator that the selected variables in fact do measure relevant aspects of choice behavior for these preference dimensions. The respective Pseudo R2 varies from 4.1% for the common difference effect on the outcome dimension in interpersonal choice to 28% in choice under risk for the common difference manipulation when the reference point is 'impossibility'. However, these differences in model fit between preference dimensions strongly decrease for the intermediate and simple models. Generally the control variables measuring personal characteristics - sex, number of inconsistencies for the respective preference dimension, number of inconsistencies for all preference dimensions - do not affect choices strongly, the only exception being interpersonal choice with respect to sex and interspatial choice with respect to the number of inconsistencies in this preference dimension (see Table 7.9 below). b) Choice under risk In choice under risk there is clear evidence for a common difference effect on the probability dimension for both reference points. When options are evaluated against the reference point 'impossibility', as predicted, the coefficient for the common difference manipulation is negative and significant at p < .001 in both the intermediate and simple model.120 Adding a common constant to the probability attribute in both options - moving them further away from the reference point - lowers the salience of the probability dimension and raises the salience of the outcome dimension. Therefore, preference for the larger and riskier outcome increases.121 In addition, preference between gambles strongly depends on the probability level on which the common difference manipulation occurs (z: 11.97; p < .001). That is, for small probability levels used to create a common difference effect (.04 vs. 05 and .24 vs. .25), there is a strong preference for the riskier option offering the larger amount. For larger probability levels (.1 vs. .2 and .4 vs. .5), irrespective of the options' amounts, a majority of subjects preferred the safer but smaller 120

Coefficients of all variables in the respective complete models are only rarely significant and therefore not reported here in detail. 121 For example, in choices between amounts ƒ100 and ƒ75 changing the respective probabilities of options from .04 and .05 to .24 and .25 raises preference for the larger but riskier reward from 86.8% to 97.6%.

172

Chapter 7

option. Finally, in the intermediate model larger amounts induce risk-seeking (z: -2.37; p < .05). When options are evaluated against the reference point 'certainty', adding a common constant to the probability attribute in both options reduces the options' distance to the reference point. Hence, salience of the distance dimension and preference for the safer option increase as predicted.122 Therefore, the coefficient for the common difference manipulation is positive and significant for both the intermediate (p < .001) and the simple model (p < .01). As to the outcome dimension there is strong evidence for a common difference effect as well. The coefficient for the common difference manipulation is positive and significant at p < .01 in the intermediate and simple model. Adding a common amount to both gambles reduces the salience of the outcome dimension and raises preference for the safer gain and the less likely loss.123 These results are also in accordance with the respective changes in the expected value of options. Both common difference manipulations alter expected values of options in favor of the safer options. c) Intertemporal choice In intertemporal choice there is clear evidence for a common difference effect on the temporal dimension. In both the intermediate and the simple model, the coefficient for the common difference effect is positive and significant at p < .001. This implies a decreased salience of the temporal and an increased salience of the outcome dimension when a common constant is added to both temporal distances. After a common difference manipulation, consistent with hyperbolic discounting, subjects prefer the larger gain and the smaller loss more often, thus confirming existing results (see e.g., Loewenstein & Prelec 1992).124 As to the common difference effect on the outcome dimension, results are less straightforward. In the intermediate model, as predicted, the coefficient for 'amount' is negative and significant at p < .001. Adding a common amount to the outcomes of both options decreases the salience of the outcome dimension and increases the salience of the temporal dimension. These changes in salience result in a stronger preference for the sooner but (relatively) smaller gain and for the larger but more distant loss.125 In addition, in this model the coefficient for the interaction between outcome sign and amount is also positive and significant at p < .001. This indicates that for larger losses the salience of the outcome dimension increases as well. Hence, when losses become large, subjects are more likely to opt for the smaller but sooner loss.126 However, removing this interaction

122

For example, in choices between amounts ƒ100 and ƒ75 changing the respective probabilities of options from .8 and .9 to .87 and .97 reduces preference for the larger but riskier option from 46.2% to 20.5%. 123 For example, in choices between probability levels .8 and 1.00 changing the respective outcomes of options from ƒ100 and ƒ75 to ƒ150 and ƒ125 reduces preference for the larger but riskier option from 12.2% to 2.6%. 124 For example, in choices between amounts ƒ100 and ƒ110 changing the respective delays of options from 'now' and '4 weeks' to '8 weeks' and '12 weeks' reduces preference for the sooner but smaller option from 63.2% to 46.3%. 125 For example, in choices between delays '8 weeks' and '12 weeks' changing the respective amounts of options from ƒ100 and ƒ110 to ƒ150 and ƒ160 raises preference for the sooner reward from 46.3% to 51.2%. 126 For example, in a choice between delays '26 weeks' and '30 weeks' changing amounts ƒ100 and ƒ110 to

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

173

term from the analysis results in insignificance of the common difference manipulation that was pronounced in most of the estimated models (not reported here in full detail). Likewise including only the interaction term into a simple model results in insignificance of this effect as well (analysis not reported here). Since the two coefficients have different polarities, the effects work in different directions and appear to be linked to each other. Similar consequences of including or removing this interaction term can be found below in the analyses of the common ratio effect and the sign effect in intertemporal choices. d) Interspatial choice In interspatial choice there is no evidence for a common difference effect on the spatial dimension. Apparently, shifting the spatial distances of two shops to be built or removed, by a common constant does neither affect salience of attributes in the intermediate or in the simple model. For the outcome dimension there is a common difference effect. Adding a common number of items-in-store to both shops as predicted lowers the salience of the outcome dimension and raises the salience of the spatial dimension - in both the intermediate and simple model the coefficient for 'amount' is negative and significant at p < .001. Hence, after a common difference manipulation on the outcome dimension subjects prefer the (relatively) smaller but closer shop to be built and the larger but more distant shop to be removed.127 e) Interpersonal choice In interpersonal choice there is neither a common difference effect on the outcome nor on the interpersonal dimension. However, in both cases in the intermediate model sex influences preference for which character receives an outcome. Separating the sample into males and females reveals a common difference effect on the social dimension for males (z: -2.16; p < .05) but not for females (z: .68; n.s.). For males adding a common constant to the social distance of characters lowers the salience of the social dimension and subjects prefer the option offering the larger gain or smaller loss for the more distant character. The coefficient is negative, because for interpersonal choice larger numerical values for social distance mean smaller social distance. Since the coefficient is negative, smaller social distance raises the salience of the social dimension and subjects prefer the smaller gain for the socially closer character.128 A similar difference between sexes occurs in the analysis for the common difference effect on the outcome dimension. The coefficient for males is significant (z: 3.3; p < .001) but for females it is not (z: -.52, n.s.). However, note that the coefficient for males is positive, indicating that - contrary to the prediction - adding a common constant to the amounts in both options raises the salience

-ƒ200 and -ƒ220 alters preference for the sooner option from 56.4% to 87.2%. Note, that unchangesd salience of outcome and temporal dimension would imply a preference for the sooner option in the second problem of 43.6 (the complement of 56.4 in the first problem). 127 For example, in choices between shops at 16 and 18 streets distance changing the respective items-instock from 150 and 500 to 1150 and 1500 raises preference for the closer shop to be built from 7.7% to 42.1%. 128 For example, in choices between amounts ƒ100 and ƒ150 changing the respective social distance of options from Leo (40) and Ben (20) to Ben 0) and Jean-Marie (0) reduces preference for the socially closer option from 50.0% to 16.7%.

174

Chapter 7

of the outcome dimension, making it more likely that the larger gain or the smaller loss is preferred.129 Table 7.9 summarizes results for the analyses of the common difference effect on both the distance and the outcome dimension for all four preference dimensions. Values for the coefficients are z-values, results for all three models - complete, intermediate, and simple - are presented, the values for the respective common difference manipulation on the distance or outcome dimension are printed in boldface. Empty cells in the respective intermediate model indicate that, due to insignificance, the specific variable was not included in the 'intermediate model'. For the distance dimension in choice under risk results are presented twice: first for options that are evaluated against the reference point 'impossibility' and second, in italics, for options that are evaluated against the reference point 'certainty'.

129

For example, in choices between Leo (40) and Ben (20) changing the respective amounts from ƒ100 and ƒ150 to ƒ150 and ƒ200 lowers preference for the socially closer but smaller option from 50.0% to 28.6%.

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

175

Table 7.9 Results for the logistic regression analyses investigating common difference effects on the respective outcome and distance dimensions for all four preference dimensions. Choice under risk

Intertemporal choice

Risk

Time

Outcome

Space

Outcome

Personal

Outcome

(df=10)

(df=10)

(df=10)

(df=10)

(df=10)

(df=10)

55.03***

31.0***

49.73***

81.58***

24.47***

29.1***

.083

.055

.068

.13

.043

.041

(df=3)

(df=2)

(df=3)

(df=3)

(df=3)

(df=3)

-0.67

-5.17***

0.27

-1.89!

-3.83***

-5.83***

2.48*

1.21

-3-33***

-2.47*

Outcome

(df=12) (df=11) (df=10)130 225.97*** 305.96*** 42.03*** .282 .253 .074 (df=3) (df=3) (df=2) -9.93*** 10.03*** 3.35*** -2.37* 3.34** n.s.

Complete model χ2 Pseudo R2 Intermediate model Distance Amount Interaction Sign *Amount Differences in amounts Certainty-effect Sex Errors1 Constant χ2 Pseudo R2

3.77*** 1.7!

-4.87***

Interpersonal choice

5.0*** 4.07*** -3.86*** n.s. 2.62** 6.32*** -6.76*** -3.12** 153.23*** 140.09*** 15.94*** .243 .119 .046

-1.13

1.20

6.93***

-2.71*** 8.39***

2.70**

1.82!

15.24***

25.02***

17.64***

54.37***

15.01**

14.93**

.02

.04

0.29

.105

.035

.024

(df=1)

Simple model CD-effect Constant χ2 Pseudo R2 Observations

Interspatial choice

-3.46*** 3.17** 0.56 -1.96* 11.96*** 10.05** .076 .015 963 480

3.12**

3.75***

-0.37

-0.57

-5.94***

-1.09

1.10

-2.5*

-0.04

1.39

6.26***

7.73***

3.01**

-.02

9.72**

14.0***

.013

0.32

35.29***

1.19

1.22

.005

.002

.000

.000

.057

.003

.001

1117

720

805

717

798

615

895

! p < .1; *p < .05; **p < .01; *** p < .001, two-tailed tests

130

The number of choice problems and consequently observations is lower for options evaluated against the reference point 'certainty'. Hence, two variables were dropped for the complete model due to multicollinearity, resulting in fewer degrees of freedom.

176

Chapter 7

7.7.4. Common ratio effects

a) Model fit As in the analysis for the common difference effects, generally model fit of the complete models was significant at p < .001, the only exception again being interpersonal choice (see table 7.10 below where all results and coefficients concerning the common ratio effects are summarized jointly). However, the intermediate models in interpersonal choice were both significant at p < .01. Hence, again for all four preference dimensions, the selected variables do measure relevant aspects of choice behavior. The respective Pseudo R2 varies from 2.5% for the common ratio effect on the social dimension to 24.3% for the common ratio effect on the probability dimension. Thus, again preference could be predicted best in choice under risk and worst for interpersonal choice. Again these differences in model fit between preference dimensions drastically decreased for the intermediate and simple models. As in the analyses for the common difference effects generally the control variables measuring personal characteristics - sex, number of inconsistencies for the respective preference dimension, number of inconsistencies for all preference dimensions - do not affect choices strongly, the only exception being choice under risk and interspatial choice, both with respect to the number of inconsistencies in this preference dimension. b) Choice under risk In choice under risk there is clear evidence for a common ratio effect on the probability dimension. The coefficient for 'distance' in both the intermediate and simple model is positive, indicating that higher values raise the salience of the probability dimension, and significant at p < .001.131 With respect to the outcome dimension in the intermediate model the coefficient is negative and significant at p < .001. Doubling amounts in two risky options raises the salience of the outcome dimension and therefore preference for the larger but riskier outcome.132 However, this effect disappears in the simple model after removing the interaction effect between probability and amount. In the intermediate model this coefficient was positive and strongly significant (p < .001), indicating that subjects strongly opt for the safer option when probabilities and stakes are both high. Since this effect runs counter to the common ratio manipulation, removing this variable from the analysis, reduces explanatory power of 'amount'. c) Intertemporal choice In intertemporal choice there is evidence for a common ratio effect on the temporal dimension. In both the intermediate and simple model the coefficient for 'delay' is negative and significant at p < .05. Doubling temporal distance of two options increases

131

For example, in choices between amounts ƒ100 and ƒ75 changing the respective probabilities of options from .04 and .05 to .4 and .5 lowers preference for the larger but riskier reward from 86.8% to 45.2%. 132 For example, in choices between probabilities .04 and .05 changing amounts ƒ100 and ƒ75 to ƒ200 and ƒ150 raises preference for the larger but riskier reward from 86.8% to 95.2%.

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

177

the salience of the temporal dimension and thus preference for the sooner option.133 As in the analysis for the common difference effect, again the interaction between sign and amount is significant at p < .001. The coefficient is positive: larger losses raise the salience of the outcome dimension and raise preference for the smaller but sooner loss. As to the common ratio effect on the outcome dimension in the intermediate model, the coefficient for 'amount' is negative and significant at p < .001. This implies that doubling amounts - contrary to the prediction and contrary to ample previous evidence in intertemporal choice - raises the salience of the temporal dimension and increases preference for the sooner reward. However, in the intermediate model again there is a strong positive coefficient (p < .001) for the interaction between sign and amount. Removing this interaction term in the 'simple model' results in a strongly reduced - positive - coefficient for 'amount' that is no longer significant. To understand theses results consider the following preference distributions (numbers in the respective cells denote preference for the sooner gain or loss): Amounts Delays Now vs. 4 weeks 8 weeks vs. 12 weeks 16 weeks vs. 24 weeks 26 weeks vs. 30 weeks 32 weeks vs. 48 weeks

Gains ƒ100 vs. ƒ110 ƒ200 vs. ƒ220 63.2 56.4 46.3 46.3 66.7 43.9 56.4 24.4 53.3 64.1

Losses

ƒ-100 vs. ƒ-110 ƒ-200 vs. ƒ-220 63.4 81.6 66.7 82.1 59.5

46.2 82.9 81.6 87.2 71.8

Preference changes in opposition to the prediction in italics A common ratio manipulation on the outcome dimension was expected to increase the salience of the outcome dimension: for gains (columns 2 and 3) preference for the sooner reward is expected to decrease between choice problems, for losses (columns 4 and 5) preference for the sooner but smaller loss is expected to increase. For three out of the five choice problems this holds, for one problem ('8 weeks vs.12 weeks') preference remains virtually unchanged. For the most remote gains and the closest losses, preference changes in opposition to the prediction. Removing these problems from the analysis results in a strong positive coefficient for 'amount in the simple model (z: 2.95; p < .01; significant for gains and losses analyzed separately at p < .05 each). Given the relatively small manipulation in the present study and the ample evidence for a common ratio effect on the outcome dimension ('magnitude effect' - see Chapters 1 and 6), the failure to replicate this effect in the present study is probably due to random effects. d) Interspatial choice In interspatial choice there is strong evidence for a common ratio effect on the spatial dimension. The coefficient for 'spatial distance' is negative and significant at p < .001, Doubling spatial distance raises the salience of the spatial dimension and increases

133

For example, in choices between amounts ƒ100 and ƒ75 changing the respective delays of options from 16 weeks and 24 weeks to 32 weeks and 48 weeks lowers preference for the sooner option from 66.7% to 53.3%.

178

Chapter 7

preference for the closer shop to be built and the larger shop to be removed.134 As to the outcome dimension in the intermediate model the coefficient for 'amount' is positive and significant at p < .001. As predicted, doubling amounts raises the salience of the outcome dimension and increases preference for the larger shop to be built and the smaller shop to be removed.135 However, in the simple model the effect of 'amount' is only marginally significant (p < .1) and the coefficient is negative. Again in the intermediate model there is a strong - negative - interaction effect of sign and amount (p < .001), indicating that higher losses raise the salience of the spatial dimension and lead to a preference to remove the more distant store. e) Interpersonal choice For interpersonal choice there is strong evidence for a common ratio effect on the social dimension. In both the intermediate and simple model the coefficient for 'social distance' is negative and significant at p < .001. Doubling social distance raises the salience of the social dimension and increases preference for the smaller gain and the larger loss both due to the more distant character. Again, 'sex' is significant in the intermediate model. Separate analyses for males and females revealed that there is a common ratio effect on the social dimension for males (z: -4.29; p < .001) but not for females (-1.04; n.s.).136 For the outcome dimension in interpersonal choice, there is no common ratio effect present in either the intermediate or the simple model. This holds also when analyses are carried out for males and females apart. Table 7.10 summarizes results for the analyses of the common ratio effect on both the distance and the outcome dimension for all four preference dimensions.

134

For example, in choices between shops with 1500 and 5000 items-in-store respectively changing the respective distances from 6 and 8 streets to 24 to 32 streets increases preference for the closer but smaller shop to be built from 21.4% to 35.5%. 135 For example, in choices between shops at 12 and 16 streets distance changing the respective items-instore from 150 and 500 to 1500 and 5000 lowers preference for the smaller but shop closer to be built from 23.7% to 14.6%. 136 For example, in choices between amounts ƒ150 and ƒ200 changing the respective social distance of options from .Leo (40) and Ben (20) to Willem (80) and Leo (40) raises preference for the socially closer option from 28.6% to 59.1%

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

179

Table 7.10. Results for the logistic regression investigating common ratio effects on the respective outcome and distance dimensions for all four preference dimensions. Choice under risk Risk Outcome Complete model χ2 Pseudo R2

(df=11)

(df=11)

Intertemporal choice Time Outcome

Interspatial choice Space Outcome

(df=10)

(df=10)

(df=10)

(df=10)

(df=10)

(df=10)

58.11*** .093

55.21*** .098

70.17*** .077

15.00 .025

21.92* .032

(df=3)

(df=2)

(df=2)

(df=4)

(df=3)

(df=3)

-2.34* -5.42***

-4.99*** -5.69***

-4.16*** 3.53***

-3-43*** 0.39

-2.62** 1.28

5.77***

5.93***

-3.93***

103.34*** 277.21*** 50.79*** .243 .231 .082 (df=3)

(df=3)

5.63*** 2.13* -4.65***

-5.61*** -3.13**

Intermediate model Distance Amount (ƒ) Certainty-effect Interaction Sign*ƒ Interaction Distance*ƒ Errors1 Sex Constant χ2 Pseudo R2

-4.54*** 33.44*** .091

.026 88.17*** .089

1.08 41.09*** .074

Simple model

(df=1)

(df=1)

5.45*** -2.60**

CR-effect Constant χ2 Pseudo R2 Observations

Interpersonal choice Personal Outcome

8.75*** -2.31*

-2.53**

0.16 35.5*** .05

6.7*** 29.3*** .044

7.42*** 40.49*** .058

-1.12 2.67** 14.49** .021

-1.52 2.43* 13.32** .019

(df=1)

(df=1)

(df=1)

(df=1)

(df=1)

(df=1)

.351 -.517

-2.11* 2.87**

1.44 .39

-5.09*** 6.31***

-1.79! 7.50***

-3.4*** 4.23***

1.19 1.03

29.66*** .03

.012 .000

4.44* .004

2.06 .001

25.9*** .031

3.22! .004

11.58*** .018

1.43 .001

479

1121

724

798

725

799

722

890

! p < .1; *p < .05; **p < .01; *** p < .001; two-tailed tests 7.7.5. Sign effects

a) Model fit Again, model fit of the complete models generally is very good. For all four preference dimensions, the models are significant at p < .001 (see Table 7.11 below). Pseudo R2 again was highest for choice under risk and lowest for interpersonal choice and varies from 24.5% to 4%. Again, for the intermediate and simple models these differences disappeared completely. b) Choice under risk In choice under risk there is clear evidence for a sign effect in the intermediate model. The coefficient for 'sign' is negative and significant at p < .001. Recall, that responses for negative had been recoded to ensure that '0' and '1' consistently denote a larger salience of either the outcome or the distance dimension. A negative coefficient then implies that

180

Chapter 7

salience of the outcome dimension in choices between negative options is increased. That is, for negative options subjects are more likely to base their choice on the outcome dimension and hence choose the smaller but more likely loss. Thus, changing the polarity of outcomes from positive to negative induces risk-aversion.137 This result runs counter to many previous findings (e.g., Tversky & Kahneman 1992). In addition, in the intermediate model a strong certainty effect is observed (p < .001), indicating that, again replicating ample previous results (see Chapter 6) preference for the safer reward increases strongly when the probability for this reward approached certainty. However, in the simple model this negative coefficient is only marginally significant (p = .076). c) Intertemporal choice In intertemporal choice there is clear evidence for a sign effect. In both the intermediate and the simple model, the coefficient for 'sign' is positive and significant at p < .001. Changing the polarity of outcomes from positive to negative raises the salience of the outcome dimension and raises preference for the smaller but sooner loss.138 d) Interspatial choice In interspatial choice the coefficient for 'sign' is negative and significant at p < .01 in both the intermediate and simple model. This indicates that in choices about which store to remove subjects prefer the larger and more distant store to be removed. Changing the polarity of options from positive to negative thus lowers the salience of the outcome dimension and raises the salience of the spatial dimension: subjects prefer to shift losses away as far as possible, even at the expense of these losses being substantially higher.139 Hence, there is a clear sign effect: subjects want to avoid losses even at the expense of higher losses - contrary to the sign effect manifesting itself in choice over time, where subjects are afraid of losses and thus prefer the sooner but smaller loss. e) Interpersonal choice In interpersonal choice there is no evidence for a sign effect in either the intermediate or the simple model. This also holds when analyses are carried out for males and females separately. Table 7.11 summarizes results for the analyses concerning the sign effect in all four preference dimensions.

137

For example, in choices between probabilities .24 and .25 changing amounts ƒ100 and ƒ75 to -ƒ100 and -ƒ75 reduces preference for the larger but less likely loss from 97.6% to 88.1% (in Appendix A the complement of the latter value - 11.9% - is indicated). 138 For example, in choices between delays 'now' and '4 weeks' changing amounts ƒ100 and ƒ110 to -ƒ100 and -ƒ110 does not affect preference for the sooner option (63.2% vs. 63.4%.). Unchanged salience of the temporal and the outcome dimension implies a preference for the sooner loss of 36.8% (the complement in the first problem) in the second problem. The increase in preference for the smaller loss then indicates an increased salience of the outcome dimension. 139 For example, in choices between shops at 6 and 8 streets distance removing shops with 150 and 500 items-in-store instead of building them alters preference for the closer store to be built or removed from 9.8% to 66.7%. Unchanged salience of the spatial and outcome dimension would imply a preference of 91.2% for the closer store to be removed. Hence, the actual reduction in preference indicates a reduced salience of the outcome dimension and a tendency to shift losses away.

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

181

Table 7.11. Results for the logistic regression investigating sign effects for all four preference dimensions Sign effects Complete model χ2 Pseudo R2 Intermediate model Distance Amount Errors1 Sign Certainty-effect Sex Constant χ2 Pseudo R2

Choice under risk Outcome

Intertemporal choice Outcome

Interspatial choice Outcome

Interpersonal choice Outcome

(df=12)

(df=11)

(df=11)

(df=11)

358.65*** .245

55.2*** .073

88.7*** .106

33.43*** .04

(df=2)

(df=3)

(df=2)

(df=4)

0.77 1.63

-2.26* 1.14

-3.58*** 6.06***

5.93***

-2.74** -2.74**

.41

1.96 36.93*** .021

0.16 39.61*** .047

6.94*** 16.89*** .02

-2.38* 3.22*** 15.73** .023

Simple model

(df=1)

(df=1)

(df=1)

(df=1)

Sign Constant

-1.77! 1.96!

5.91*** -1.16

-2.71** 6.90***

0.39 2.94**

3.15! .002

34.87*** .046

7.36** .006

.02 .000

1680

1203

1198

1335

χ2 Pseudo R2 Observations

! p < .1; *p < .05; **p < .01; *** p < .001, two-tailed tests 7.8. Overall summary and discussion

This chapter demonstrated that discount functions for risk, time, and spatial distance are indeed similar with respect to several features. In all these domains distant outcomes are discounted and discounting increases with increasing distance from the respective reference point, 'for sure', 'now', and 'here'. We were able to replicate several existing results from studies on choice under risk and intertemporal choice. Effects such as the common difference effect, common ratio effect, and sign effect showed up in our experiment as well, confirming that our experimental setup did work. In choice under risk and intertemporal choice the discounting function seems to be hyperbolic, if one keeps in mind that for risky choice with its two reference points, certainty and impossibility, this results in an S-shaped probability weighting function (see Chapter 6). For intertemporal choice, in addition, often the interaction between sign and amount is significant, indicating that subjects prefer the smaller but sooner loss when stakes are higher. For interspatial choice, results yield a consistent picture, too. There was no common difference effect on the spatial dimension, but all other effects are significant at p < .001, except for the sign effect which is significant at p < .01. These results suggest first, that the discounting function for spatial distance is better modeled by an exponential function

182

Chapter 7

instead of an S-shaped or hyperbolic function and second, that the utility function for shops varying in size to be built or removed does not differ from the utility function for monetary amounts. Interestingly, the sign effect in interspatial choice raises the salience of the spatial dimension and increases preference to remove the more distant shops. In analogy to the sign effect often reported in choice under risk where people are riskseeking for losses despite of these losses being larger, in interspatial choice people appear to be interested to shift losses away as far as possible, despite of these distant losses being larger. While for choice under risk and intertemporal choice stimuli could be used that had been tested in ample previous research, for interspatial and especially interpersonal choice the stimuli had to be created by ourselves, since there exists no previous comparable research in these domains. After the manipulation checks for interpersonal choice had been carried out and the stimulus set had been adapted, results for the discounting processes in interpersonal choice were much less consistent with the predictions. Analyzing the data jointly for both sexes reveals only a (very strong) common ratio effect on the social dimension suggesting - as in interspatial choice - an exponential discounting function for social distance. All other predicted effects could not be demonstrated. Because the manipulation check revealed that perception of the stimulus set differed between men and women and because 'sex' in several analyses influenced preference, the data were analyzed separately for men and women. For females, in all analyses no single effect even approached significance. For men, common difference effects on the social and the outcome dimension occurred. In addition, the common ratio effect was observed on the social dimension. Table 7.12 summarizes the most important results for the four preference dimensions and the respective effects. Values (z-values) are taken from the intermediate models. For the common difference effect on the probability dimension, first the value for options evaluated against the reference point 'impossibility' and then (in italics) the value for options evaluated against the reference point 'certainty' is given. For interpersonal choice, results are presented for males and females separately as well as for both sexes combined. Effects that run counter to the prediction are presented in italics.

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

183

Table 7.12 Comparisons of significant results for all four preference domains. Risky Choice

Intertemporal Choice

Interspatial Choice

Interpersonal Choice

Males

Effects:

Females

Combined

3.77***

n.s.

-2.16*

n.s.

n.s.

CDOutcome

-9.93*** 3.35*** 3.34**

-4.87***

-5.83***

3.3***

n.s.

n.s.

CRDistance

5.63***

-2.34*

-4.99***

-4.29***

n.s.

CROutcome

-5.61***

-5.69***

3.53***

n.s.

n.s.

n.s.

Sign

-3.58***

5.93***

-2.74**

n.s.

n.s.

n.s.

CDDistance

-3.43***

*p < .05; **p < .01; *** p < .001; n.s. = no effect; two-tailed tests Given these results, what are possible directions for future research? Besides this general similarity in discounting functions for risk, temporal delay and spatial distance, the question whether these preference dimensions resemble each other seems appropriate. That is, are risk, time, and spatial distance interchangeable or do they only have similar discount functions? If they were interchangeable, subjects who are more risk-averse for positive gambles should also be more eager to receive benefits here and now. This could be checked by seeing how well individual choices in one domain predict individual choices in another domain. Another interesting task would be to investigate how several effects combined do influence preference. For example, is it possible to compare preference in a risky choice problem with another risky choice problem where manipulations of sign, common difference effect on the outcome dimension, and common ratio effect on the probability dimension are carried out simultaneously? Further, the effects investigated in this chapter do not cover all similarities previously observed in choice under risk and intertemporal choice. Chapter 6 mentions effects of framing, due to shifts of the reference point completely left out in the experiments reported here. Application of such effects to interspatial choice seems highly warranted. Further, the effect of adding attributes to simple options consisting of two attributes should be investigated. For instance, what is the effect of adding spatial distance to a temporally delayed outcome? It is well known that adding risk to intertemporal options leads to a lower salience of the temporal dimension and to less temporal discounting. Do similar effects occur when the added attribute is an interspatial one? In our interpretation the addition of a third or fourth attribute to already distant outcomes has less effect on preference than introducing any form of distance to outcomes. Suppose, the reference point against which outcomes are compared is multidimensional and distances for each attribute are integrated into an overall distance. Then, making an option that is identical to the reference point distant on one attribute affects overall discounting more strongly, than making an already distant option more distant. Hence, the addition of a third or fourth attribute generally should favor the initially more distant option offering the larger outcome. Finally, it should be explored why results for interpersonal choice do differ so strongly for men and women as was observed in this

184

Chapter 7

study. Here it seems useful to replicate the results for the male sample with an adapted stimulus set.

Generalizing 'anomalies' in choice under and intertemporal choice to other preference dimensions

185

Appendix A: Preference distributions over choice problems Choice under risk (preference percentages for the riskier reward indicated, outcome dimension more salient: risk-aversion for gains and risk-seeking for losses) Gains Amounts Probabilities ƒ100 vs. ƒ75 ƒ150 vs. ƒ125 ƒ200 vs. ƒ150 .04 vs. .05 86.6 65.9 95.2 .1 vs. .2 2.6 10.2 10.3 .24 vs. .25 97.6 100 97.6 .4 vs. .5 45.2 11.9 48.8 .8 vs. .9 46.2 28.2 30.8 .87 vs. .97 20.5 17.1 30.8 .8 vs. 1.00 12.2 2.6 12.2

Losses -ƒ100 vs. -ƒ75 -ƒ150 vs. ƒ-125 -ƒ200 vs. -ƒ150 19.5 19.0 7.7 65.8 79.5 74.4 11.9 23.1 17.1 51.3 66.7 22.0 63.4 61.0 53.7 71.8 80.5 64.3 90.2 86.6 86.8

Intertemporal Choice (preference percentage for the sooner option indicated, temporal dimension more salient for gains, outcome dimension more salient for losses) Gains Losses Amounts ƒ100 vs. ƒ110 ƒ150 vs. ƒ160 ƒ200 vs. ƒ220 -ƒ100 vs. -ƒ110 -ƒ150 vs. ƒ-160 -ƒ200 vs. -ƒ220 Delays Now vs. 4 63.2 54.8 56.4 63.4 56.1 46.2 weeks 8 weeks vs. 46.3 51.2 46.3 81.6 64.1 82.9 12 weeks 16 weeks vs. 66.7 60.5 43.9 66.7 70.7 81.6 24 weeks 26 weeks vs. 56.4 46.2 24.4 82.1 66.7 87.2 30 weeks 32 weeks vs. 53.3 61.0 64.1 59.5 73.2 71.8 48 weeks

Interspatial choice (preference for the closer option indicated, spatial dimension more salient for gains, outcome dimension - items in-store of shops to be built - more salient for losses) Shops Distance Here vs. 2 streets away 6 streets away vs. 8 streets away 12 streets away vs. 16 streets away 16 streets away vs. 18 streets away 24 streets away vs. 32 streets away

Gains (shops to be built) 150 vs. 500 1500 vs. 5000 1150 vs. 1500

Losses (shops to be removed) -150 vs. -500 -1500 vs. -5000 -1150 vs. -1500

7.7

10.3

28.2

75.6

71.1

59.5

9.8

21.4

28.2

66.7

75.6

71.4

23.7

14.6

63.4

82.9

53.7

46.3

7.7

7.7

42.1

78.0

69.2

56.4

3.9

39.5

71.8

71.8

54.8

33.3

186

Chapter 7

Interpersonal choice (preference percentage for the socially closer option indicated, social dimension more salient for gains, outcome dimension more salient for losses. Results are separately presented for both sexes, values between brackets indicate preference for the socially closer option in the whole sample. Differences in preference between the sexes appear quite large but keep in mind that each cell summarizes approximately 20 subjects. After each character its social distance score is also given between brackets) Males Gains Characters ƒ100 vs. ƒ150ƒ150 vs. ƒ160ƒ200 vs. ƒ300 Pim (100) vs. 25.0 (41.0) 22.2 (31.0) 31.3 (39.5) Willem (80) Leo (40) vs. 50.0 (55.0) 28.6 (35.1) 50.0 (40.0) Ben (20) Willem (80) vs. 55.6 (53.7) 59.1 (52.2) 68.8 (56.4) Leo (40) Pierre (10) vs. 22.2 (26.2) 6.3 (21.1) 12.5 (25.6) Jean-Marie (0) Ben (20) vs. 16.7 (41.5) 18.8 (35.9) 12.5 (28.9) Jean-Marie (0) Ben (20) vs. 22.2 (30.2) 27.8 (36.6) 27.8 (45.2) Pierre 1(0)

Losses -ƒ100 vs. -ƒ150 -ƒ150 vs. -ƒ160 -ƒ200 vs. -ƒ300 83.3 (62.8) 75.0 (66.7) 81.3 (71.1) 50.0 (47.4)

77.8 (58.5)

81.3 (61.5)

31.3 (55.3)

56.3 (42.1)

44.4 (43.9)

87.5 (73.4)

100 (65.7)

83.3 (68.3)

72.2 (63.4)

88.9 (69.0)

81.3 (69.2)

66.7 (63.4)

81.3 (66.7)

81.3 (76.3)

Females Gains Losses Characters ƒ100 vs. ƒ150ƒ150 vs. ƒ160ƒ200 vs. ƒ300 -ƒ100 vs. -ƒ150 -ƒ150 vs. -ƒ160 -ƒ200 vs. -ƒ300 Merel (100) vs. 52.2 (41.0) 37.5 (31.0) 45.5 (39.5) 48.0 (62.8) 60.9 (66.7) 63.6 (71.1) Anne (80) Hanneke (40) vs. 59.1 (55.0) 39.1 (35.1) 33.3 (40.0) 45.5 (47.4) 43.5 (58.5) 47.8 (61.5) Hennie (20) Anne (80) vs. 52.2 (53.7) 45.8 (52.2) 47.8 (56.4 72.7 (55.3) 31.8 (42.1) 43.5 (43.9) Hanneke (40) Louise (10) vs. 29.2 (26.2) 31.8 (21.1) 34.8 (25.6) 63.6 (73.4) 47.8 (65.7) 56.5 (68.3) Marie-Claire (0) Hennie (20) vs. 60.9 (41.5) 47.8 (35.9) 40.9 (28.9) 56.5 (63.4) 54.2 (69.0) 60.9 (69.2) Marie-Claire (0) Hennie (20) vs. 36.0 (30.2) 43.5 (36.6) 58.3 (45.2) 60.9 (63.4) 56.5 (66.7) 72.7 (76.3) Louise 1(0)

Chapter 8 Summary and discussion

Summary

In this chapter the research questions and the main findings from this dissertation are briefly summarized and recapitulated. In addition, some issues concerning the measurement of impatience are discussed (again). The chapter concludes with suggestions for further research and for potential application of the experimental results to real world phenomena.

188

Chapter 8

8.1. Introduction

Intertemporal choices abound. We all frequently make choices where our immediate actions have future consequences and where we have to take these consequences into account. However, we all vary with respect to the extent we are willing to take these delayed consequences into account. Moreover, we all vary with respect to what specific intertemporal tradeoffs are acceptable. Sometimes we are eager to receive outcomes immediately, while sometimes we show much more patience and are well aware of longterm consequences. This book so far has dealt with three questions concerning such intertemporal tradeoffs. It was first investigated when people are more likely to be driven by their short-term interests and hence neglect future consequences. Second, it was investigated which people are most likely to take long-term consequences of their immediate actions into account. Finally, the question was considered whether and how behavioral regularities observed in intertemporal choice can be generalized to other choice domains where people have to make similar tradeoffs. This chapter summarizes the content of each chapter, discusses the major empirical findings, and presents some suggestions for future research. Chapter 1 introduced the standard economic model as to how intertemporal tradeoffs between current and future benefits ought to be made. The Discounted Utility model (DU-model) states that rewards should be devalued (discounted) by a fixed proportion of their utility for every time interval that they are to be delayed. Further, according to economists, this devaluation should generally be equal to or at least closely related to the market interest rate. Empirical research has demonstrated several anomalies to this model. For example, people do not discount the future at a constant rate as prescribed by the model but weight time delays close to immediacy much heavier than identical delays occurring later in time. Similar problems for the DU-model have been reported with respect to the discounting of different amounts and forms of money. While small amounts are usually discounted heavily, i.e., people charge relatively high premiums for having the receipt of small payments delayed, people are rather patient, i.e., charge relatively low premiums, for having the receipt of larger amounts delayed. Further, people discount certain forms of money, such as windfall gains, much stronger than others, say permanent income. A final example for the DU-model’s descriptive shortcomings is that people are much more aversive to delay the receipt of (positive) outcomes than they are eager to speed-up the receipt of identical (delayed) outcomes. Subjects charge higher premiums for delay than they are willing to pay for a speed-up of the same outcome. Because for the DU-model all that matters is an object's utility and its (absolute) time delay, this behavior is incompatible with the model. Standard attempts to trivialize these anomalies do not hold up to scrutiny. For example the claims that 'anomalies do not transfer to real world behavior where people have incentives to behave rationally in the sense of the theory' or 'individual errors cancel out at an aggregate level' are ill-founded for both theoretical and empirical reasons (see Chapter 1 for an extensive discussion). Some prominent alternatives to the DU-model (e.g., Loewenstein & Prelec 1992) are also subject to criticism. They are insensitive to situational variations in individual discount rates. Objects of identical utility should according to these models - be discounted identically, an implication that appears implausible, in view of existing empirical evidence (see Chapter 1 for an extensive discussion). Subsequently the three above-mentioned research questions were introduced.

Summary and discussion

189

Yet another alternative, a framing theory linked to a model for choice behavior (the 'discrimination model') both developed by Lindenberg (e.g., Lindenberg 1980, 1986) was introduced. It was argued that this model can account for differences in discount rates for objects of identical present value, for individual differences in discount rates, i.e., some people discounting more strongly than others, and for similarities of anomalies in intertemporal choice, choice under risk and previously unrelated areas of choice behavior such as interpersonal or interspatial choice. The theory assumes that individuals are boundedly rational, that is, they have cognitive limitations. More specifically, though individuals have (unrealistically) large calculating abilities, they frame situations such that they concentrate on only one goal in a given situation and maximize utility given that frame. Possible goals relevant in an intertemporal context are 'immediacy - receive benefits as soon as possible' or 'gain - receive benefits as large as possible'. 8.2. When do people act myopically?

In Chapter 2 some of the discrimination model’s postulates and implications were tested. As to the model’s postulates, it was investigated whether preference depends on the salience of frames and whether this salience in turn depends on the frame's power to discriminate between choice options. That is, whether one frame clearly favors one option over the other. Simple intertemporal options, consisting of a temporal and an outcome (monetary) attribute, were presented to subjects who rated the importance of these attributes. For such simple options, attributes are equivalent to frames and these importance ratings therefore can be interpreted to measure salience. In two experiments salience served as a very good predictor for preference. Subjects being in a frame where ‘immediacy’ is the most important goal are more likely to prefer the sooner option and vice versa. Then, it was also investigated whether discriminatory power of frames also serves as a good predictor for preference. Evidence with respect to this postulate was mixed. In one choice problem where no frame discriminated between choice options, preference in fact was distributed evenly between options. However, in another choice problem where the temporal attribute discriminated much better than the outcome attribute - and the sooner option therefore should be preferred more often - preference between options was distributed evenly as well. In another experiment we tried to relate preference changes between choice problems to changes in the discriminatory power of attributes. That is, if in comparison between two choice problems A and B the later and larger option is preferred more strongly in A than in B, then, in comparison to the temporal attribute, the outcome attribute must discriminate more strongly in A than in B. Here, preference and discriminatory power of frames were correlated, as predicted, but this correlation did not reach significance and the predicted preference reversals did not occur. Thus, the question remains open whether the absence of effects is due to a failure of the experimental design or a descriptive shortcoming of the model. Because the other predictions of the model were by and large confirmed by the data, the first explanation appears to be more valid, but this issue has not been settled in this dissertation. To test several implications of the model’s postulates, discriminatory power and salience of frames were manipulated indirectly. To do so, some experiments from consumer research on the effect of the choice context on preference were adapted. Most concepts of rationality require that the relative attractiveness between two options remains unaffected by what other options are available. However, according to the

190

Chapter 8

discrimination model such an addition of other options might alter the discriminatory power of frames. Consequently, the salience of frames and preference between options are altered as well, a prediction compatible with results from consumer research on context effects. In a series of experiments it was demonstrated that the addition of options that alter the discriminatory power of frames in fact changes preference such that the frame discriminating best between options becomes decisive and the option preferable with respect to this frame becomes preferred more often. Chapter 3 turned to the more specific question which frames are associated with myopic behavior and under which frames long-term considerations are more important. It was proposed that hedonic aspects of consumption are more likely to lead to a frame where immediate receipt of the goods is salient than utilitarian considerations. Since hedonic usage of a good provides direct utility to the decision maker, it can be imagined better, resulting in a stronger feeling of deprivation when one does not possess an object. Consequently, delaying hedonic goods is more aversive than delaying utilitarian goods and people prefer to receive the former sooner. In three experiments, subjects in fact preferred to receive hedonic goods sooner than utilitarian goods. In one of these experiments the effect of hedonic usage on salience of goals favoring immediate receipt was so strong that it could even outweigh utility (i.e., monetary) differences. That is, though the utilitarian good was more attractive for subjects – because its selling price was higher – subjects nevertheless preferred to receive the hedonic good sooner, an effect running counter to the Discounted Utility model as well as its major descriptive rival, the hyperbolic discounting model. In addition, it was investigated whether the same preference ordering between hedonic and utilitarian goods also holds when subjects do not have to state which good they would like to receive sooner, but have to state adequate monetary charges that would compensate them for the aversive feeling of delaying the receipt of an attractive good. From a framing point of view, the latter measure, though often taken to be equivalent with the former in the literature, might lead to different results because asking for monetary amounts inclines subjects to think in terms of utilitarian aspects (money) even when hedonic goals are considered. For example, subjects may reflect about the appropriateness of the compensation charge or alternative usage of the money. Such utilitarian aspects should wipe out or even reverse the difference in impatience between hedonic and utilitarian goods. The results showed that, in choice situations involving compensation charges effects for hedonic goods were indistinguishable from the effects for utilitarian goods. Chapter 4 again investigated intertemporal preferences for hedonic and utilitarian goods but this time it was studied how sensitive preferences are to a vivid presentation of the goods. Though vividness, contrary to our intuitions, has little effect on preference and judgments in general, several researchers suggested effects of vividness in intertemporal choice. However, there is disagreement among different theorists as to whether vividness raises or lowers impatience for both and whether vividness affects both goods similarly or not. We argued that if vividness stresses the hedonic aspects of consumption experience, it should raise impatience for hedonic goods more strongly than for utilitarian goods. Further, in analogy to Chapter 3 we also expected impatience for both goods to depend on the way impatience is measured. First, an experiment was designed to test whether vividness affects hedonic and utilitarian goods equally in choices about which good to receive sooner. It turned out that for such choice situations, vividness raised the immediate attractiveness of hedonic goods

Summary and discussion

191

much more strongly than of utilitarian goods. As in Chapter 3 this effect disappeared when subjects had to determine adequate monetary compensation charges for delaying both goods. In fact, in this experiment compensation charges for utilitarian goods were even higher than for hedonic goods. 8.3. Who acts myopically?

While the former three chapters dealt with the question when people are likely to behave myopically, Chapter 5 investigated which people are most likely to behave in a shortsighted way. Since patience is advantageous in numerous situations and this kind of ability is learned to a large extent in the parental household, many parents will try to induce adequate preferences within their children. That is, they will try to raise their children such that they are able to overcome their natural tendency to behave shortsightedly. However, parents differ with respect to ability, incentives and restrictions to teach patience to their children. For example, parents who are held more responsible for the behavior of their children have higher incentives to socialize them properly. Having many siblings at home decreases parental time per child and thus decreases the opportunities to teach children self-control. Using data containing detailed retrospective information about the raising of the respondents and a psychological scale to measure how strongly respondents take future consequences into account, it was demonstrated that indeed higher incentives for parents to socialize children 'properly' lead to more patient respondents. Furthermore, parents who serve as a better model for patience, due to educational achievements and professional requirements, also produce more patient offspring. Little evidence was found for an effect of restrictions, such as family composition. There was evidence that parental socialization techniques, such as the strictness or parental stimulation of children, affect patience. Respondents reporting to be raised more strictly and to grow up with more stimulation appeared to be more concerned about future consequences. Investigating potential causes for patience is only valuable, if differences in patience have consequences for actual behavior. Hence, patience as measured by the psychological scale was also used to predict potential consequences of myopic behavior in a variety of settings, e.g., health behavior, engagement at school and university, and engagement in potentially risky behavior. In all of these domains, subjects who are classified as more patient report to behave more prudently. That is, they consumed less alcohol and fewer drugs, were less likely to smoke and more likely to engage in sporting activities, they did their homework at school more carefully, followed extra courses at the university, and finally were much less likely to show deviant behavior in a variety of settings. While these results were obtained by using a psychological scale measuring concern for future consequences, patience is usually measured in intertemporal choice experimentally by calculating discount rates from delaying hypothetical or actual monetary payoffs. Because Chapters 3 and 4 demonstrated that asking subjects to translate their feelings of impatience into terms of monetary payoffs does affect their framing and consequently impatience, it is questionable whether this measure actually deserves its current prominent position. To investigate this point further, it was also tested whether parental socialization also affects patience if it is measured by the delay of hypothetical amounts and whether this measure serves as an as good predictor for actual behavior as the psychological scale.

192

Chapter 8

The results confirm our suspicion. Patience, measured by the monetary measure turned to be much less predictable than when measured by the psychological measure and the former also predicted actual behavior much worse than the latter. Further, there was no relation between both measures of patience. Not only were there no correlations between the psychological scale and the scale for delaying monetary outcomes, but no single factor predicted patience on both scales. For example, higher parental status reduced children's patience measured by the psychological scale, but it did not affect patience as measured by the money-scale. These results corroborated results from Chapters 3 and 4 and underlined that different measures for impatience not only reflect different kinds of impatience, but also that calculating discount rates from the delay of monetary amounts omits important aspects of intertemporal choices. Framing effects thus appear to be important also for measurement instruments. 8.4. Generalization of effects

Chapter 6 compared normative concepts and empirical results in choice under risk and intertemporal choice and demonstrated the close analogies between the normative axiomatic formulations as well as their almost identical empirical violations. It was argued that these similarities reflect general behavioral tendencies: people evaluate outcomes against a multi-dimensional reference point and prefer them to be as large as possible, to be certain, to be obtained 'now', to occur to themselves and to occur 'here'. Deviations of any of these attributes from this ideal state lead to a devaluation of the object’s absolute value. Therefore, the empirical regularities observed in choice under risk and intertemporal choice are not restricted to these preference dimensions but can be generalized to interspatial and interpersonal choice. Moreover, this devaluation should be similar across preference dimensions and therefore result in similar empirical regularities and discount functions for all preference dimensions. Chapter 7 replicated the above-mentioned regularities for choice under risk and intertemporal choice and empirically tested their generalization to interspatial and interpersonal choice. To do so, first, decision-theoretic research on interpersonal and interspatial choice was reviewed briefly and then some space was devoted to the discussion of what might be adequate stimuli for both preference dimensions. For interspatial choice, shops and supermarkets (varying with respect to items-in-store) that could be opened or closed down at varying spatial distance were used as stimuli. For social distance, a metric measure was created, making this dimension roughly comparable to the easily quantified dimensions of risk, time and space. For this metric measure fictive characters defined by social-economic status, age, hobbies, political orientation, and nationality were created. These dimensions were weighted according to a pretest. Social distance for each character was determined by the weighted sum of distances for each dimension. For example, in the pretest for a variety of settings, hobbies were rated as three times as important for social closeness as nationality. A character being identical to what was considered to be the ‘average subject’ with respect to hobbies but different with respect to nationality thus had a lower social distance than a character being identical with respect to nationality and different with respect to hobbies. The social distance created in this way was compared with the actual perception of the character's social distance by the subjects. It turned out that many effects observed in intertemporal choice and choice under risk could also be observed in interspatial choice. Examples are the sign effect and the

Summary and discussion

193

common ratio effect. Therefore, there is evidence for a general tendency to discount outcomes for several preference dimensions similarly. Moreover, since for the spatial dimension common ratio effects were observed but common difference effects were not, the discount function for spatial distance appears to be exponential instead of hyperbolic as for intertemporal choice or S-shaped as for probabilities. For interpersonal choice results were much less clear. First, tests of the subjects' perception of the stimuli revealed that males perceived the stimuli different than intended. Therefore, the stimulus set was slightly adapted. For females, perception of stimuli was much more in accordance with the intentions. However, for females none of the predicted effects could be demonstrated while for males results were much more in accordance with our predictions. For example, we observed common difference and common ratio effects for the social dimension, which can adequately be explained by assuming a hyperbolic discount function for social distance. Future research should investigate whether this result is an artifact due to misspecification of the stimuli for males or whether there is in fact a discrepancy in the discounting of social distance between males and females. 8.5. Measuring impatience

A reoccurring theme in this dissertation is how to measure impatience and whether different measures of impatience lead to identical results. Standard measures of impatience, such as the delay of hypothetical amounts and stating adequate monetary compensation charges, turned out to be much less reliable and to lead to different results than simply asking subjects which of two outcomes they would like to receive sooner. Framing theory would predict that making the subject think in terms of money would direct attention from hedonic to utilitarian aspects and thus reduce impatience. Thus, measurement issues should be treated as theoretical issues and be covered by the same theory that is used to predict the measured effects. Besides these theoretical considerations, solely measuring impatience by calculating discount rates from compensation charges for delaying monetary amounts is unreliable, as we have seen. It may be very difficult for subjects to state adequate compensation charges for situations they have no experience with, as reported in a previous experiments on the delay/speed-up asymmetry. There, Loewenstein (1988: 208) commented on the striking result that on average subjects valued a delayed dinner slightly more than an immediate: 'This result again testifies to the unimportant role played by time delay when subjects are simply asked how much they would pay to consume at different points in time'. 8.6. Suggestions for future research

This section summarizes research problems not sufficiently dealt with or neglected thus far in this book, and it suggests applications for some of its results. We begin with some open questions. First, in Chapter 2 we tried to relate changes in preference between choice problems to changes in the power of frames to discriminate between choice options. It was not possible to demonstrate such an effect. This may be due to the fact that the choice problems under study were chosen such that, contrary to the intentions, no preference change occurred. That is, preference for the sooner option statistically was not

194

Chapter 8

stronger for the problem 'ƒ100 now vs. ƒ110 in 4 weeks' than for the problem 'ƒ 100 in 10 weeks vs. ƒ110 in 14 weeks'. It appears useful to again let subjects independently rate the difference in attractiveness for both the temporal and outcome attribute but shift both options in the second choice problem further into the future such that changes in preference occur. If the theory is correct, these preference changes should be correlated with changes in the discriminatory power of attributes. Further, also in Chapter 2, for a choice set consisting of two options, adding an option that is almost indistinguishable from one of the initial options, raised preference for the distinct option only marginally. It is therefore not clear, whether or not the hypothesis is confirmed. This experiment should be replicated to clarify this issue. In the following chapters it was investigated when people are likely to behave myopically and who is most likely to behave in such a way. A natural follow-up would be to relate these findings to each other. That is, to use the personal measures for impatience introduced in Chapter 5 as predictors for the intertemporal choices between consumer goods presented in Chapters 3 and 4. In Chapter 7 it remained unclear whether the discrepancy in results for social discounting between males and females - absence of results for females, predicted effects for males - was due to sexes' different discounting of social distance or due to a misspecification of the stimulus set. Further research with respect to this question seems warranted. In addition, Chapter 7 did not investigate all regularities specified in Chapter 6. For example, range effects were completely left out of the experimental design as were the anomalies equivalent to the delay/speed-up asymmetry in intertemporal choice. Here, further research is needed as well. Further, it is interesting to investigate whether the same personal correlates that affect patience do also affect risky and interspatial preferences. Since the arguments put forth in Chapter 5 refer to preference formation in general, one would expect similar personal characteristics to affect preference for all three dimensions. A related question not tackled in this book is whether the same subjects are impatient, risk-averse, and prefer benefits to occur here. That is, to investigate more closely whether not only discount functions for these dimensions are similar, but whether the temporal, probability and spatial attributes can in fact be substituted for each other without altering preference. Consider a subject who prefers to receive (ƒ100; now) over (ƒ110, in four weeks). Does this preference increase the likelihood to prefer the gamble (75; 1) over the gamble (100; .8)? A final question with respect to the similarities between these preference dimensions is the effect of presenting multi-attribute choices, consisting of three attributes, e.g., outcome, temporal delay, and spatial distance, to subjects. Previous research in intertemporal choice suggests that adding a third attribute, e.g., adding probabilities to options consisting of outcome and temporal delay, results in relatively less discounting than the discounting caused by each attribute individually. Similar effects could be expected when the probability attribute is replaced by a spatial attribute. Turning to applications to real world phenomena, it would be useful to replicate the results concerning impatience for hedonic and utilitarian goods using actual choices. Chapters 3 and 4 demonstrate that people are more likely to behave impatiently when hedonic consumption experience is important. Hence, myopic behavior should be more frequent in such situations, especially when goods are presented vividly and choices do not involve apparent financial trade-offs. For example, one would expect (excessive) credit card shopping to occur more often when hedonic goods are involved. Further, one would expect people to engage in activities with long-term negative consequences

Summary and discussion

195

especially when immediate hedonic consumption is traded for such long-term consequences.140 Another implication is that it is easier to stick to one’s long-term consumption plan when utilitarian considerations are involved than when hedonic aspects play a role. For example, it might be easier to save for utilitarian than for hedonic goods, because saving for the latter creates higher levels of deprivation. Hedonic goods therefore should be more likely to be bought on credit. While these considerations pertain to the situational characteristics of temporal discounting, it is of course also possible to apply the personal characteristics demonstrated in Chapter 5 to different areas. In many economic analyses, differences in time preference strongly affect behavior. Hence, the effects of patience on behavior in such settings should be investigated in more detail than it has been done in this book. For example, one would expect patient people to save more and to display different behavior with respect to fertility and deviant behavior. Although in this dissertation much emphasis has been put on the usefulness of farsighted behavior, it remains an open question as to whether farsightedness is always valuable. For example, hesitation in many respects seems to be opposed to impatience, likewise curiosity in some respects seems to be related to impatience. Future research might be devoted to these relations in more detail. A final noteworthy point is that although in this dissertation much emphasis has been put on 'anomalies' to rational choice theory, framing theory is compatible with the notion that much, but not all, behavior is rational and in fact adequately modeled by standard microeconomic theory. Because people maximize utility given their frame, in market situations where 'gain' is by far the most prominent goal, predictions from economics and framing theory converge. This possibly also accounts for the fact that in experiments investigating the functioning of markets, predictions of economic theory are regularly borne out, while in individual decision making 'irrational' behavior is observed much more often (see Kagel & Roth 1995 for overviews). Prices and quantities in such market experiments usually quickly converge to equilibrium solutions specified in advance. However, in such cases the situation is well structured for subjects: they have to maximize some experimental credits which in turn are converted to actual payoffs. Experimental credits and consequently payoffs clearly form the most salient goal for subjects in such situations. In experiments investigating the consistency of individual preferences, numerous deviations from rational models have been reported, some of which are summarized in this book. According to framing theory this is due to the fact that, through experimental manipulations, the salience of goals shifts and hence choices (or 'preferences') change. Utility-maximizing rational behavior is maintained given a certain frame but not necessarily between frames.

140

Addictions appear to clearly confirm this suggestion. However, since for additive goods consumption reinforces itself, addictions differ from choices between consumer goods as investigated in this book.

References

197

References: Ahlbrecht, Martin and Martin Weber. (1995) ‘Hyperbolic Discounting Models in Prescriptive Theory of Intertemporal Choice.’ Zeitschrift für Wirtschafts- und Sozialwissenschaften. 115: 535-568. Ahlbrecht, Martin and Martin Weber. (1997) ‘An Empirical Study on Intertemporal Decision Making under Risk.’ Management Science 43: 813-826. Akerlof, George. (1991) ‘Procrastination and Obedience.’ American Economic Review. 81 (2): 1-19. Akerlof, George. (1997) 'Social Distance and Social Decisions.' Econometrica. 65: 10051027. Allais, Maurice. (1953) 'Le comportement de l'homme rationnel devant le risque: Critiques des postulats et axiomes de l'école américaine.' Econometrica. 21: 503-546. Ainslie, George. (1975) ‘Specious Reward: A Behavioral Theory of Impulsiveness and Impulse Control.' Psychological Bulletin 82 (4): 463-496. Ainslie, George. (1991) 'Derivation of 'rational' economic behaviour from hyperbolic discount curves.' American Economic Review. 81 (May): 334-340. Ainslie, George. (1992) ‘Picoeconomics.’ Cambridge: Cambridge University Press. Ainslie, George and Nick Haslam. (1992). 'Hyperbolic Discounting'. In: George Loewenstein and Jon Elster (Eds.): Choice over Time. New York: Sage. pp. 57-92. Arkes, Hal, Cynthia Joyner, Mark Pezzo, Jane Nash, Karen Siegel-Jacobs, and Eric Stone. (1994) ‘The Psychology of Windfall Gains.’ Organizational Behavior and Human Decision Processes. 59: 331-354. Astone, Nan M. and Sara McLanahan. (1991) ‘Family Structure and High School Completion: The Role of Parental Practices.’ American Sociological Review. 56: 309320. Bandura, Albert, and Richard H. Walters. (1963) ‘Social Learning and Personal Development.’ New York, Chicago etc.: Holt, Rinehart, and Wolters. Barsky, Robert, Thomas Juster, Miles Kimball, and Mathew Shapiro. (1997) ‘Preference Parameters and Behavioral Heterogeneity: An Experimental Approach in the Health and Retirement Sector.’ Quarterly Journal of Economics. 62: 537-579. Beattie, Jane and Graham Loomes. (1997) 'The impact on incentives upon risky choice experiments.' Journal of Risk and Uncertainty. 14: 155-168. Becker, Gary S. (1976) ‘The Economic Approach to Human Behavior.’ Chicago: University of Chicago Press. Becker, Gary S. (1991) ‘A Treatise on the Family.’ Cambridge: Harvard University Press. Becker, Gary S. (1992) 'Habits, Addictions, and Traditions.' Kyklos. 45: 327-345. Becker, Gary S. (1996) ‘Accounting for Tastes.’ Cambridge: Harvard University Press. Becker, Gary S. and Kevin Murphy. (1988) ‘A Theory of Rational Addiction.’ Journal of Political Economy. 96: 675-700. Becker, Gary and Casey Mulligan. (1997) ‘The Endogenous Determinants of Time Preference.’ Quarterly Journal of Economics. 112: 729-758. Bellenger, D., D. Robertson, and E. Hirschman. 1978 'Impulse buying varies by product.' Journal of Advertising Research. 18: 15-18.

198

References

Benartzi, Shlomo and Richard Thaler. (1995) 'Myopic Loss Aversion and the Equity Premium Puzzle.' Quarterly Journal of Economics. 110: 75-92. Benzion, Uri, Amnon Rapoport, and Joseph Yagil. (1989) ‘Discount Rates Inferred from Decisions: An Experimental Study.' Management Science. 35 (3): 270-284. Bernstein, Lionel, Gretchen Chapman, Caryn Christensen, and Arthur Elstein. (1997) ‘Models of Choice Between Multioutcome Lotteries.’ Journal of Behavioral Decision Making. 10: 93-107. Björkman, Mats. (1984) 'Decision making, risk taking, and psychological time: review of empirical findings and psychological theory. ' Scandinavian Journal of Psychology. 25: 31-49 Bless, Herbert, Tilman Betsch, and Axel Frenzen. (1997) 'Framing the framing effect: The impact of context cues on solutions to the "Asian disease" problem. European Journal of Social Psychology. 28: 287-291. Blossfeld, Hans-Peter. 1996 ‘Macro-Sociology, rational choice theory, and time. A theoretical perspective on the empirical analyses of social processes.’ European Sociological Review. 12 (2): 181-206. Bogardus, Emory. (1959) 'Social Distance.' Yellow Springs, OH, Antioch. Braun, Norman and Pablo Vanini. (1998) 'On Habits and Addictions.' Universität Bern, Institut für empirische Soziologie and Universität Zürich, Institut für Empirische Wirtschaftsforschung. Braspenning, Joze. (1992) 'Framing: de prospecttheorie en het discriminatiemodel. Een empirische vergelijking en validiering. Ph.D. dissertation. Amsterdam: Thesis. Bruins, Joost and Jeroen Weesie. (1996) ‘Hin95Exp. A Set of Experiments connected with HiN95. Codebook of Hin95Exp. ISCORE Paper, Nr.77. Utrecht University’ Camerer, Colin. (1995) 'Individual Decision Making.' In: John Kagel and Alvin Roth: The Handbook of Experimental Economics. Princeton: Princeton University Press: 587703. Camerer, Colin, Linda Babcock, George Loewenstein, and Richard Thaler. (1997) 'Labor Supply of New York City Cabdrivers: One Day at a Time.' Quarterly Journal of Economics. 112 (2): 407-441. Cairns, John and Marjon van der Pol. (2000) 'Valuing future private and social benefits: The discounted utility model versus hyperbolic discounting models.' Journal of Economic Psychology. 21: 191-205. Chapman, Gretchen B and Arthur S. Elstein. (1995) ‘Valuing the future: temporal discount rates for health and money.’ Medical Decision Making. 15: 373-386. Chapman, Gretchen B. (1996) 'Expectations and Preferences for Sequences of Health and Money.' Organizational Behavior and Human Decision Processes. 67: 59-75. Chapman, Gretchen B. (1998) ‘Sooner or Later: The Psychology of Intertemporal Choice.’ In: D.L. Medin (Ed.): The Psychology of Learning and Motivation. Vol. 38. San Diego: Academic Press: 83-113. Coleman, James S. (1990) ‘Foundations of Social Theory.’ Harvard: Belknap Press. Coleman, James S. (1993) ‘The Rational Reconstruction of Society.’ American Sociological Review. 58: 1-15. Cropper, Maureen L., Sema Aydede, and Paul Portney. (1994) ‘Preferences for LifeSaving Programs: How the Public Discounts Time and Age.’ Journal of Risk and Uncertainty. 8: 243-265.

References

199

DeBondt, Walter and Richard Thaler. (1986) 'Does the Stock Market Overreact?' Journal of Finance. 40 (3): 793-807. DeBondt, Walter and Richard Thaler. (1987) 'Further Evidence on Investor Overreaction and Stock Market Seasonality.' Journal of Finance. 42 (3): 557-581. DeBondt, Walter and Richard Thaler. (1990) 'Do Security Analysts Overreact.' American Economic Review. 80 (2): 52-57. Dhar, Ravi and Rashi Glazer. (1996) 'Similarity in Context: Cognitive Representation and Violation of Preference and Perceptual Invariance in Consumer Choice.' Organizational Behavior and Human Decision Processes. 67 (3): 280-296. Dhar, Rhavi and Klaus Wertenbroch. (2000) ‘Consumer choice between hedonic and utilitarian goods.’ Journal of Marketing Research. 37 (February): 60-71. Dittmar, Helga,, Jane Beattie, and Susanne Friese. (1996) ‘Objects, decision considerations and self-image in men’s and women’s impulse purchases.’ Acta Psychologica. 93: 187-206. Downey, Douglas B. (1995) ‘When bigger is not better: Family Size, Parental Resources, and Children’s Educational Performance.’ American Sociological Review. 60: 746-761. Dronkers, Bas and Arthur van Soest. (1997. ‘Subjective Measures of Household Preferences and Financial Decisions.’ CentER Discussion Paper No. 9770. Eckel, Catherine, and Phillip Grossman. (1998) ‘Are women less selfish than men?: Evidence from dictator experiments.’ The Economic Journal. 108: 726-735. Elster, Jon. (1979) ‘Ulysses and the Sirens: Studies in rationality and Irrationality.’ Cambridge: Cambridge University Press. Esser, Hartmut. (1990) '"Habits, "Frames" und "Rational Choice". Die Reichweite von Theorien der rationalen Wahl am Beispiel des Befragtenverhaltens.' Zeitschrift für Soziologie. 19 (4): 231-247. Esser, Hartmut. (1991) 'The Rationality of Everyday Behavior. A Rational Reconstruction of the Theory of Action by Alfred Schütz.' Rationality and Society. 5: 731. Esser, Hartmut. (1996) 'Die Definition der Situation.' Kölner Zeitschrift für Soziologie und Sozialpsychologie. 48 (1): 1-34. Fennema, Hein and Peter Wakker. (1997) ‘Original and Cumulative Prospect Theory: A Discussion of Empirical Differences.’ Journal of Behavioral Decision Making. 10: 53-64. Fennema, Hein and Marcel van Assen. (1999) 'Measuring the utility of losses by means of the trade-off method.' Journal of Risk and Uncertainty. 17: 277-295. Fishburn, Peter and Ariel Rubinstein. (1982) ‘Time Preference.' International Economic Review. 22 (3): 677-694. Fisher, Irving. (1930) 'The Theory of Interest.' New York: Macmillan. Fischoff, Baruch. (1983) 'Predicting Frames.' Journal of Experimental Psychology: Learning, Memory, and Cognition. 9:103-116. Fiske, Susan T. and Shelley E. Taylor. (1984) ‘Social Cognition.’ Reading, Menlo Park etc.: Addison-Wesley. Frank, Robert H. (1992) 'Frames of Reference and the Intertemporal Wage Profile.' in: George Loewenstein and Jon Elster (Eds.): Choice over Time. New York: Sage. pp. 371382. Frey, Bruno and Reiner Eichenberger. (1989a) 'Should Social Scientists Care about Choice Anomalies?' Rationality and Society. 1: 101-122.

200

References

Frey, Bruno and Reiner Eichenberger. (1989b) 'Anomalies and Institutions.' Journal of Institutional and Theoretical Economics. 145: 423-437. Frey, Bruno and Reiner Eichenberger. (1996) 'Marriage Paradoxes.' Rationality and Society. 8 (2): 187-206. Fuchs, Victor. (1982) 'Time Preference and Health: An Explorative Study.' In: Victor Fuchs (Ed.): Economic Aspects of Health. Chicago: University of Chicago Press. Fuchs, Victor. (1996) ‘Economics, Values, and Health Care Reform.’ American Economic Review. 86 (1): 1-22. Gately, Dermot. (1980) ‘Individual Discount Rates and the Purchase and Utilization of Energy-Durables: Comment.’ The Bell Journal of Economics. 11: 373-374. Gigerenzer, Gerd and Ullrich Hoffrage. (1995) 'How to improve Baysian resoning without instructions: frequency formulas.' Psychological Review. 103: 650-669. Gneezy, Uri and Jan Potters. (1997) 'An Experiment on Risk Taking and Evaluation Periods.' Quarterly Journal of Economics. 112 (2): 631-645. Goffman, Erving. (1974) ‘Frame Analysis: An Essay on the Organization of Experience.’ Cambridge, MA: Harvard University Press. Gottfredson, Michael R. and Travis S. Hirschi. (1990) ‘A General Theory of Crime.’ Stanford: Stanford University Press. Green, Donald and Ian Shapiro. (1994) 'Pathologies of Rational Choice Theory: A Critique of Applications in Political Science.' New Haven: Yale University Press. Grether, David, and Charles Plott. (1979) ‘Economic theory of choice and the preference reversal phenomenon.’ American Economic Review. 69: 623-628. Harvey, C.M. (1986) 'Value functions for infinite period planning.' Management Science. 32: 1123-1139. Hausman, Jerry. (1979) ‘Individual Discount Rates and the Purchase and Utilization of Energy-Durables.’ The Bell Journal of Economics. 10: 33-54. Hechter, Micheal and Satoshi Kanazawa. (1997) 'Sociological Rational Choice Theory.' Annual Review of Sociology. 23: 191-214. Herstein, Israel and James Milnor. (1953) ‘An axiomatic approach to measurable utility.’ Econometrica. 21: 291-297. Herne, Kaisu. (1996) 'The role of decoys in choice: a review of research on contextdependent preferences.' Risk Decision and Policy. 1 (1): 105-119Herrnstein, Richard. (1990) 'Rational Choice Theory: Necessary but not Sufficient.' American Psychologist. 45: 356-367. Herrnstein, Richard, George Loewenstein, Drazen Prelec, and William Vaugham (1991) 'Utility Maximization and Melioration: Internalities in Individual Choice.' Journal of Behavioral Decision Making. Herrnstein, Richard and Drazen Prelec. (1992) 'Melioration.' In: George Loewenstein and Jon Elster (Eds.): Choice over Time. New York; Sage. pp. 235-266. Hirschi, Travis S. (1994) ‘Family.’ in: Travis S. Hirschi and Michael R. Gottfredson (Eds.): ‘The Generality of Deviance.’ New Brunswick: Transaction Publishers. pp. 47-69. Hirschi, Travis S. and Michael R. Gottfredson (Eds.). (1994) ‘The Generality of Deviance.’ New Brunswick: Transaction Publishers. Hoch, Stephen and George Loewenstein. (1991) ‘Time-inconsistent Preferences and Consumer Self-Control.’ Journal of Consumer Research. 17: 492-507.

References

201

Hogarth, Robin and Marvin Reder (Eds.). (1987) 'Rational Choice. The Contrast Between Economics and Psychology.' Chicago: University of Chicago Press. Holocomb, James and Paul Nelson. (1992) ‘Another Experimental Look at Individual Time Preference.' Rationality and Society. 4 (2): 199-220. Hosmer, David and Stanley Lemeshow. (1989) 'Applied Logistic Regression.' New York: Wiley. Huber, Joel, John Payne, and Chris Puto. (1982) 'Adding Asymmetrically Dominated Alternatives: Violations of Regularity and the Similarity Hypothesis.' Journal of Consumer Research. 9 (June): 90-98. Jenni, Karen and George Loewenstein. (1997) ‘Explaining the “Identifiable Victim Effect.’ Journal of Risk and Uncertainty. 14 (3): 235-257. Johnson, M.K. and M.M. Marini (1998) 'Bridging the racial divide in the United States: the effect of gender.' Social-Psychological-Quarterly. 61(3): 247-258. Junger, Marianne. (1994) ‘Accidents.’ in: Travis S. Hirschi and Michael R. Gottfredson (Eds.): ‘The Generality of Deviance.’ New Brunswick: Transaction Publishers. pp. 81112. Kachelmeier, Stephen and Mohamed Shehata. (1992) 'Examining Risk Preferences Under High Monetary Incentives: Experimental Evidence from the People's Republic of China.' American Economic Review. 70: 1120-1141. Kagel, John and Alvin Roth. (1995) 'The Handbook of Experimental Economics.' Princeton: Princeton University Press. Kalmijn, Matthijs, Wim Bernasco, and Jeroen Weesie. (1996) ‘Households in the Netherlands 1995 - Huishoudens in Nederland 1995. Codebook of HiN95.’ ISCORE Paper, Nr. 67. Utrecht University. Kahneman, Daniel and Amos Tversky. (1979) ‘Prospect Theory: An Analysis of Decisions under Risk.' Econometrica. 47 (2): 265-291. Kahneman, Daniel and Amos Tversky. (1984) ‘Choices, Values, and Frames.' American Psychologist. 39 (4): 341-350. Kahneman, Daniel, Peter Wakker, and Rakesh Sarin. (1997) 'Back to Bentham? Explorations of experienced utility.' Quarterly Journal of Economics. 112: 375-405. Kapteyn, Arie and T. Wansbeek. (1982) 'Empirical Evidence on Preference Formation.' Journal of Economic Psychology. 2: 132-154. Keren, Gideon and Peter Roelofsma. (1995) ‘Immediacy and Certainty in Intertemporal Choice.’ Organizational Behavior and Human Decision Processes. 63 (3): 287-297. Kirby, Kris. (1997) 'Bidding on the Future: Evidence against Normative Discounting of Delayed Rewards.' Journal of Experimental Psychology: General. 126: 54-70. Koopmans, Tjalling. (1960) ‘Stationary Ordinal Utility and Impatience.' Econometrica. 28 (2): 287-309. Laibson, David. (1997) ‘Golden Eggs and Hyperbolic Discounting.’ Quarterly Journal of Economics. 112: 443-477. Laibson, David. (1998) 'Life-cycle consumption and hyperbolic discount functions.' European Economic Review. 42: 861-871. Laibson, David and Zeckhauser. (1998) 'Amos Tversky and the Ascent of Behavioral Economics.' Journal of Risk and Uncertainty. 16: 7-47.

202

References

Lancaster, Kelvin. (1963) ‘An axiomatic theory of consumer time preference.' International Economic Review. 4: 221-231. Laub, John H. and Robert Sampson. (1994) ‘Unemployment, Marital Discord and Deviant Behavior: The Long-term Correlates of Childhood Misbehavior.’ in: Travis Hirschi and Michael Gottfredson (Eds.): ‘The Generality of Deviance.’ New Brunswick: Transaction Publishers. pp. 235-252. Lawrence, Emily. (1991) 'Poverty and the Rate of Time Preference: Evidence from Panel Data.' Journal of Political Economy. 99 (11): 54-79. Lea, Stephen, R. Tarpy, and Paul Webley. (1982) 'The Individual in the Economy.' Cambridge: Cambridge University Press. Leland, Jonathan. (1994) 'Generalized Similarity Judgements: an Alternative Explanation for Choice Anomalies.' Journal of Risk and Uncertainty. 9: 151-172. Leland, Jonathan. (1998) 'Similarity Judgements in Choice under Uncertainty: A reinterpretation of Regret Theory.' Management Science. Lichtenstein, Sarah and Paul Slovic. (1973) ‘Response induced reversals of preferences in gambling: an extended replication in Las Vegas.' Journal of Experimental Psychology. 101 (1): 16-20. Liebrand, Wim and Charles McClintock. (1988) ‘The ring measure of social values: a computerized procedure for assessing individual differences in information processing and social value orientation.’ European Journal of Personality. 2: 217-230. Liebrand, Wim, David Messick, and Henk Wilke. (Eds.). (1992) ‘Social Dilemmas. Theoretical Issues and Research Findings.’ Oxford, New York: Pergamon Press. Ligthart, Paul. (1995) ‘Solidarity in Economic Transactions.’ Ph.D. Dissertation. Amsterdam: Thesis. Lindenberg, Siegwart. (1981) Rational, repetitive choice: the discrimination model versus the Camilleri-Berger model. Social Psychology Quarterly. 44 (4): 312-330. Lindenberg, Siegwart. (1986) 'The paradox of privatization in consumption.' In: Andreas Diekmann and P. Mitter (Eds.): Paradoxial Effects of Social Behavior. Essays in Honor of Anatol Rappoport. Heidelberg, Wien: Physica Verlag. pp.297-310. Lindenberg, Siegwart. (1989) 'Choice and Culture: The behavioral basis of cultural impact on transactions.' in: Hans Haferkamp (ed.): Social Structure and Culture. Berlin: de Gruyter. pp. 175-200. Lindenberg, Siegwart M. (1992) 'The method of decreasing abstraction.' in: James Coleman and Thomas Fararo: Rational Choice Theory: Advocacy and Critique. Newbury Park: Sage. pp. 3-20. Lindenberg, Siegwart M. (1993) ‘Club hierarchy, social metering and context instruction: governance structures in response to varying self-command-capital.' in: Siegwart Lindenberg and H. Schroeder (Eds.): Interdisciplinary Perspectives on Organization Studies. Oxford: Pergamon Press. pp. 195-220. Lindenberg, Siegwart. (1995) 'Short-term prevalence, social approval and the governance of employment relations.' In: J. Gronenwegen (Ed.): Transaction Cost Economics and Beyond. Boston, Dordrecht, London: Kluwer. pp.129-147. Lindenberg, Siegwart. (1996) 'Choice-centered versus subject-centered theories in the social sciences: the influence of simplification on explananda.' European Sociological Review. 12: 147-157. Lindenberg, Siegwart and Bruno Frey. (1993) 'Alternatives, frames, and relative prices: a broader view of rational choice.' Acta Sociologica. 36: 191-205.

References

203

Loewenstein, George. (1987) Anticipation and the Value of Delayed Consumption.' The Economic Journal. 97: 666-684. Loewenstein, George. (1988) ‘Frames of Mind in Intertemporal Choice.’ Management Science. 34 (2): 200-214. Loewenstein, George. (1992) 'The Fall and Rise of Psychological Explanations in the Economics of Intertemporal Choice.' In: George Loewenstein and Jon Elster (Eds.): Choice over Time. New York: Sage. pp. 3-34. Loewenstein, George. (1996) ‘Out of Control: Visceral Influences on Behavior.' Organizational Behavior and Human Decision Processes. 65 (3): 272-292. Loewenstein, George and Jon Elster (Eds.). (1992) 'Choice over Time.' New York: Sage. Loewenstein, George and Richard Thaler. (1989) ‘Anomalies: Intertemporal Choice.’ Journal of Economic Perspectives. 3 (4): 181-193. Loewenstein, George and Drazen Prelec. (1991) 'Negative Time Preference'. American Economic Review. 81: 347-352. Loewenstein, George and Drazen Prelec. (1992) ‘Anomalies in Intertemporal Choice: Evidence and an Interpretation.’ Quarterly Journal of Economics. 107 (2): 573-597. Loewenstein, George and Drazen Prelec. (1993) 'Preferences for Sequences of Outcomes.' Psychological Review. 100: 91-108. Loewenstein, George and Sicherman. (1991) 'Do Workers prefer Increasing Wage Profiles?' Journal of Labor Economics. 9: 67-84. Loewenstein, George, Leigh Thompson, and Max Bazerman. (1989) ‘Social Utility and Decision Making in Interpersonal Contexts.’ Journal of Personality and Social Psychology. 57 (3): 426-441. Luce, Duncan and Howard Raiffa. (1957) 'Games and Decisions.' New York: John Wiley Luce, Duncan and Peter Fishburn. (1991) ‘Rank- and sign-dependent linear utility models for finite first-order gambles.’ Journal of Risk and Uncertainty. 4: 29-59. Mazur, James. (1987) 'An adjustment procedure for studying delayed reinforcement.' In: M.L. Commons, J., J. Mazur, J. Navis, and H. Rachlin: 'Quantitative analysis of behavior: The effect of delay and intervening events on reinforcement value.' Hillsdale, NJ: Erlbaum. pp. 55-73. Machina, Mark. (1982) 'Expected utility analysis without the independence axiom. Econometrica. 50: 277-323. Mano, Haim and Richard Oliver. (1993) ‘Assessing the Dimensionality and Structure of the Consumption Experience: Evaluation, Feeling and Satisfaction.’ Journal of Consumer Research. 20: 451-466. Manser, Marylin and Murray Brown. (1980) 'Marriage and Household DecisionMaking: A Bargaining Analysis.' International Economic Review. 6: 21 (1): 31-44. McElroy, Majorie and Mary Jean Horney. (1981) 'Nash-bargained Household Decisions. Towards a Generalization of the Theory of Demand.' International Economic Review. 22 (2): 333-349. McGill, Ann L. and Punam Anand. (1989) ‘The Effect of Vivid Attributes on the Evaluation of Alternatives: The Role of Differential Attention and Cognitive Elaboration.’ Journal of Consumer Research. 16: 188-196. Medin, Douglas L., Robert L. Goldstone, and Arthur B. Markman. (1995) 'Comparison and choice: relations between similarity processes and decision processes.' Psychonomic Bulletin and Review. 2 (1): 1-19.

204

References

Metcalfe, Janet and Walter Mischel. (1999) 'A Hot/Cool-System Analysis of Delay of Gratification: Dynamics of Willpower.' Psychological Review. 106 (1): 3-19. Mincer, Jacob and Haim Ofek. (1982) 'Interrupted Work Careers: Depreciation and Restoration of Human Capital.' Journal of Human Resources. 17 (1): 3-24. Mischel, Walter. (1961) ‘Father absence and delay of gratification: Cross-cultural comparisons.’ Journal of Abnormal and Social Psychology. 63:116-124. Mischel, Walter. (1968) 'Personality and Assessment.' New York: Wiley. Mischel, Walter. (1974) 'Processes in Delay of Gratification.' In L. Berkowitz (Ed.): Advances in Experimental Social Psychology. Vol. 7. New York: Academic Press. pp. 249-292. Mischel, Walter, Youchi Shoda, and Monica Rodriguez. (1992) ‘Delay of Gratification in Children. in: George Loewenstein and Jon Elster (Eds.) ‘Choice over Time.’ New York: Russell Sage. pp. 147-166. Nisbett, Richard and Lee Ross. (1980) ‘Human Inference: Strategies and Shortcomings of Social Judgement.’ Englewood Cliffs: Prentice Hall. O'Donoghue, Ted and Matthew Rabin. (1999) "Incentives for Procrastinators," Quarterly Journal of Economics. 114 (3): 769-816. O'Donoghue, Ted and Matthew Rabin. (2000) "The Economics of Immediate Gratification," Journal of Behavioral Decision Making. 13 (2): 233-250 O'Donoghue, Ted and Matthew Rabin. (2001) "Choice and Procrastination," Quarterly Journal of Economics. 116 (1): 121-160. Olson, Mancur and Martin Baily. (1981) 'Positive Time Preference.' Journal of Political Economy. 89: 1-25. Orphanides, Athanasios and David Zervos. (1998) 'Myopia and addictive behavior.' The Economic Journal. 108 (January): 75-91. Ott, Notburga. (1992) 'Intrafamily Bargaining and Household Decisions.' Berlin: Springer. Parsons, Talcott. (1937) ‘Structure of Social Action.’ Harvard: Harvard University Press. Parsons, Talcott. (1951) The Social System. New York: Routledge. Phelps, Edmund and Robert Pollak. (1968) ‘On Second-Best National Saving and GameEquilibrium Growth.’ Review of Economic Studies. 35: 185-199. Prelec, Drazen and George Loewenstein. (1991) ‘Decision Making over Time and under Uncertainty: A Common Approach.' Management Science. 37 (7): 770-786. Prelec, Drazen and George Loewenstein. (1998) 'The Red and the Black: Mental Accounting of Savings and Debt.' Marketing Science.17 (1): 4-28. Prelec, Drazen, Birger Wernerfelt, and Zettelmeyer. (1997) 'The Role of Inference in Context Effects: Inferring What you Want from What is Available.' Journal of Consumer Research. 24 (June): 118-126. Quiggin, John. (1982) 'A Theory of Anticipated Utility.' Journal of Economic Behavior and Organization. 3: 323-343. Quiggin, John and John Horowitz. (1995) ‘Time and Risk.’ Journal of Risk and Uncertainty. 10: 37-55.

References

205

Rabin, Matthew. (1998) 'Psychology and Economics.' Journal of Economic Literature. 36(1): 11-46. Rachlin, Howard and Andres Raineri. (1992) ‘Irrationality, Impulsiveness, and Selfishness as Discount Reversal Effects.’ In George Loewenstein and Jon Elster (eds.), Choice over Time. New York: Sage. Rawls. John. (1971) 'A Theory of Justice.' Harvard: Harvard University Press. Read, Daniel and Barbara Van Leeuwen. (1999) ‘Predicting hunger: The effects of appetite and delay on intertemporal choice.’ Organizational Behavior and Human Decision Processes. 76: 189-205. Read, Daniel, George Loewenstein, and Adrian Bivolaru. (1999) ‘Mixing virtue and vice: Combining the immediacy effect and the diversification heuristic.’ Journal of Behavioral Decision Making. 12: 257-273. Read, Daniel and George Loewenstein. (2000) 'Time and Decision: Introduction to the Special Issue.' Journal of Behavioral Decision Making. 13: 141-144. Reinink, Maarten. (2000) 'Afstandsgevoeligheid van voorkeuren.' (Distance-dependence of preferences). Master's Thesis. Groningen Ryan, Bruce and Gerald Adams. (1995) ‘The Family-School Relationship Model.’ in: Bruce Ryan, Gerald Adams, Tomas Gulllotta, Roger Weissberg and Robert Hampton (Eds.): The Family-School Connection. Thousand Oaks: Sage. pp. 3-28. Roelofsma, Peter. (1994) 'Intertemporal Choice.' Ph.D. Dissertation. Amsterdam, Free University. Roelofsma, Peter and Daniel Read. (2000) 'Intransitive Intertemporal Choice.' Journal of Behavioral Decision Making. 13: 161-177. Ross, Catherine and Chia-Ling Wu. (1995) 'The links between education and health.' American Sociological Review. 60: 719-745. Savage, Leonard. (1954) ‘ The Foundations of Statistics.’ New York: Wiley. Samuelson, Paul. (1937) 'A Note on the Measurement of Utility.' Review of Economic Studies. 4: 155-161. Schelling, Thomas C. (1984) ‘The Intimate Contest for Self-Command.’ in: Thomas C. Schelling: ‘Choice and Consequence.’ Harvard: Harvard University Press. pp. 57-82. Schelling, Thomas C. (1992) ‘Self-Command: A New Discipline.’ in: George Loewenstein and Jon Elster (Eds.) Choice over Time. New York: Russell Sage. pp. 167176. Shefrin, Hersh and Richard Thaler. (1992) ‘Mental Accounting, Saving and SelfControl.’ in: George Loewenstein and Jon Elster (Eds.) Choice over Time. New York: Russell Sage. pp. 287-330. Seidl Christian and Stefan Traub (1996) 'Rational Choice and the relevance of irrelevant alternatives.' CentER Discussion Paper No. 9691. Tilburg. Shelley, Marjorie. (1993) ‘Outcome Signs, Question Frames and Discount Rates.’ Management Science. 39 (7): 806-815. Shelley, Majorie. (1994) ‘Gain/Loss Asymmetry in Intertemporal Choice.' Organizational Behavior and Human Decision Processes. 59: 124-159. Shelley, Marjorie and Thomas Omer. (1996) ‘Intertemporal Framing Issues in Management Compensation.’ Organizational Behavior and Human Decision Processes. 66 (1): 42-58. Simonson, Itamar. (1989) 'Choice based on reasons: the case of attraction and compromise effects.' Journal of Consumer Research. 16: 158-174.

206

References

Simonson, Itamar and Amos Tversky. (1992) 'Choice in Context: Tradeoff-Contrast and Extremeness Aversion.' Journal of Marketing Research. 29 (August): 281-296. Slovic, Paul and Sarah Lichtenstein. (1968) ‘Relative importance of probability and payoffs in risk taking.’ Journal of Experimental Psychology Monograph. 78 (2): 1-18. Snijders, Chris. (1996) 'Trust and Commitments.' Ph.D. Dissertation. Amsterdam: Thesis. Snijders, Chris and Jeroen Weesie. (1998) 'Sociale Oriëntaties, Tijdspreferenties, en de Stabiliteit van Relaties.' in: Matthijs Kalmijn, Wim Bernasco and Jeroen Weesie (Eds.): Huwelijks- en samenwoonrelaties in Nederland: De organisatie van afhankelijkheid. Assen: Van Gorcum. Pp. 209-226. Stigler, George and Gary Becker. (1977) ‘De gustibus non est disputandum.‘ American Economic Review. 67: 76-90. Stevenson, Mary. (1986) 'A Discounting Model for Decisions With Delayed Positive or Negative Outcomes.' Journal of Experimental Psychology: General. 115: 131-154. Stevenson, Mary. (1992) 'The Impact of Temporal Context and Risk on the Judged Value of Future outcomes.' Organizational Behavior and Human Decision Processes. 52: 455491 Strathman, Alan, Faith Gleicher, David S. Boninger, and C. Scott Edwards. (1994) ‘The Consideration of Future Consequences: Weighing Immediate and Distant Outcomes of Behavior.’ Journal of Personality and Social Psychology. 66 (4): 742-752. Strotz, Robert. (1956) 'Myopia and Inconsistency in Dynamic Utility Maximization.' Review of Economic Studies. 23: 166-180. Thaler, Richard. (1980) 'Towards a Positive Theory of Consumer Choice.' Journal of Economic Behavior and Organization. 1: 39-60. Thaler, Richard. (1981) ‘Some Empirical Evidence on Dynamic Inconsistency.’ Economic Letters. 8: 201-207. Thaler, Richard. (1990) 'Saving, fungibility, and mental accounts.' Journal of Economic Perspectives. 4: 193-205. Thaler, Richard and Eric Johnson. (1990) 'Gambling with the house money and trying to break even: The effects of prior outcomes on risky choice.' Management Science. 36: 643-660. Thomas William and Florian Zaniecki. (1927) 'The Polish Peasant in Europe and America.' New York: Knopf Tversky, Amos. (1969) 'Intransitivity of Preferences.' Psychological Review. 76 (1): 3148. Tversky, Amos and Daniel Kahneman. (1973) ‘Availability: A Heuristic for Judging Frequency and Probability.’ Cognitive Psychology. 5: 207-232. Tversky, Amos and Daniel Kahneman. (1991) ‘Loss Aversion in Riskless Choice: A Reference Dependent Model.’ Quarterly Journal of Economics. 107: 1039-1061. Tversky, Amos and Daniel Kahneman. (1992) ‘Advances in Prospect Theory: Cumulative Representations of Uncertainty.’ Journal of Risk and Uncertainty. 5: 297323. Tversky, Amos and Craig Fox. (1995) ‘Weighing Risk and Uncertainty.’ Psychological Review. 102: 269-283. Tversky, Amos and Itamar Simonson. (1993) 'Context-dependent preferences.' Management Science. 39: 1179-1189.

References

207

Tversky, Amos, Paul Slovic, and Daniel Kahneman. (1990) ‘The Causes of Preference Reversal.’ American Economic Review. 80 (1): 204-217. Tversy, Amos, Samuel Sattah and Paul Slovic. (1988) 'Contingent Weighting in Judgment and Choice.' Psychological Review. 95 (3): 371-384. Varey, Carol and Daniel Kahneman. (1992) 'Experiences extended across time: Evaluation of moments and episodes.' Journal of Behavioral Decision Making. 5: 169185. Vlek, Charles and Gideon Keren. (1992) ‘Behavioral decision theory and environmental risk management: Assessment and resolution of four ‘survival’ dilemmas.’ Acta Psychologica. 80: 249-278. Von Neumann, John and Oskar Morgenstern. (1982) [1944]. Theory of Games and Economic Behavior. Princeton: Princeton University Press. Wahlund, Richard and Jonas Gunnarsson. (1996) ‘Mental Discounting and Financial Strategies.’ Journal of Economic Psychology. 17: 709-730. Wakker, Peter and Amos Tversky. (1993) 'An Axiomatization of Cumulative Prospect Theory.' Journal of Risk and Uncertainty. 7: 147-176. Wathieu, Luc. (1997) 'Habits and the Anomalies in Intertemporal Choice.' Management Science. 43 (11): 1552-1563. Weber, Elke and Britt Kirsner. (1997) 'Reasons for Rank-Dependent Utility Evaluation.' Journal of Risk and Uncertainty. 14: 41-61. Weber, Max. (1904) 'The Protestant Ethic and the Spirit of Capitalism.' New York Wedell, Douglas. (1991) 'Distinguishing among models of contextually induced preference reversals.' Journal of Experimental Psychology: Learning, Memory, and Cognition. 17-767-778. Wedell, Douglas and Jonathan Pettibone. (1996) ‘Using Judgments to Understand Decoy Effects in Choice.’ Organizational Behavior and Human Decision Processes. 67 (3): 326-344. Wernerfelt, Birger. (1995) 'A Rational Reconstruction of the Compromise Effects: Using Market Data to Infer Utilities.' Journal of Consumer Research. 21: 627-633. White, Halbert. (1980) ' A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity.' Econometrica. 48: 817-838. Wu, George and Richard Gonzalez. (1996) ‘Curvature of the Probability Weighting Function.’ Management Science. 42 (12): 1676-1690. Wunderinck, Sophia. (1995) ‘Is family planning an economic decision?’ Journal of Economic Psychology. 16: 377-392. Yaari, Menahem. (1987) 'The dual theory of choice under risk.' Econometrica. 55: 95115.

Questionnaire used in the first experimental session

209

Appendix A Questionnaire used in the first experimental session Booklet concerning choice behavior This is research concerning behavior in several decision situations. We would like to ask you some questions over some decision problems. Try to imagine that you really have to choose between the offered options. Circle what you think to choose in reality. Take your time to think about your answer but try not to think too long. There are no wrong or right answers, but we are only interested in your own opinion. Do not look back. Thank you very much for your participation. If you have questions concerning the meaning or aims of one of the questions, please contact me.

Drs. Alexander Gattig University of Groningen Department of Sociology/ICS .......................................................................................................................................................... What do you prefer? (circle): A:

f 100 over 7 days

B:

f 200 over 10 days

What do you prefer? (circle): A:

f 100 over 7 days

B: f 120 over 14 days .......................................................................................................................................................... What do you prefer? (circle): A:

f 100 now

B:

f 110 over 4 weeks

Indicate on the following scale how important the two aspects of the situation (receiving the amount quickly and receiving a high amount) were to you. If you are only interested in receiving a high amount then choose the value 1. If you are only interested to receive the amount quickly, choose the value 9. If you find both aspects equally important, choose then the value 5. Circle: Amount of money most important 1

2

3

4

5

6

7

8

Immediacy of receipts most important 9

210

Appendix A

.......................................................................................................................................................... The following question concerns the distribution of money for a good friend. You have to decide yourself between the different options for your friend. Choose the options of which you think that it is the best for your friend. First, write down the name of your friend: _____________ What do you prefer? (circle) C: Your friend gets f 100 over 10 weeks D: Your friend gets f 110 over 14 weeks .......................................................................................................................................................... Imagine you participate in a gamble with three possible results. Your win or loss is determined by randomly drawing a marble from an urn. There are 100 marbles in this urn, among which are red, white and blue marbles. Which gamble do you prefer? (circle) A:

You win f 3000, with a probability of 33 % You win f 2700, with a probability of 66 % You win nothing with a probability of 1 % (there are 33 red, 66 white, and 1 blue marble in the urn)

B:

You get f 2700 for sure (there are only white marbles in the urn) .......................................................................................................................................................... Imagine a gamble where you can win an amount of money, if from an urn with 100 marbles among which red and white marble randomly a red marble is drawn. The percentage of red marbles in the urn can vary between 0 and 100. The following question concerns a gamble for a good friend of yours. First, write down the name of the good friend: ___________ What do you prefer? (circle) C:

Your friend wins f 3000, with a probability of 33 % Your friend wins nothing with a probability of 67 % (there are 33 red and 67 white marbles in the urn)

D:

Your friend wins f 2700 with a probability of 34% Your friend wins nothing with a probability of 66 % (there are 34 red and 66 white marbles in the urn) .......................................................................................................................................................... 'Imagine you have to choose between an amount of ƒ 100 which you receive in 10 weeks and an amount of ƒ 110 which you receive in 14 weeks. If you only look at the amount, how large do you estimate the difference in attractiveness between f 100 and f 110? To answer the question, use the following scale from 1 to 9 (1 means 'identical attractiveness, 9 means 'very large difference in attractiveness'). Circle: ‘f 100’ and ‘f 110’ equally attractive

‘f 110’ much more attractive than ‘f 100’ 1

2

3

4

5

6

7

8

9'

Questionnaire used in the first experimental session

211

If you only consider the difference in temporal delay, how large do you consider the difference in attractiveness between '10 weeks delay' and '14 weeks delay' to be? Answer this question on the following scales from 1 to 9 by circling one number (1 means ‘identical attractiveness’, 9 means ‘very large difference in attractiveness’)? ’10 weeks’ and ‘14 weeks’ equally attractive

’10 weeks’ much more attractive than 14 weeks'

1 2 3 4 5 6 7 8 9 ........................................................................................................................................................ Now there follow some questions concerning consumer goods. Your answer to these questions could depend on whether you own such goods or not. In case you own such goods, imagine that this is not the case. For example, you could imagine the good broke down recently. ‘Imagine: you have participated in a lottery. You could not only win monetary prizes but also consumer goods. You were so fortunate to win a flat-screen television set with Dolby stereo to create cinema atmosphere (official price of f2400). You did not have a television set at home. What is the minimal amount for which you would like to sell this television set: f__________? Because there is a temporary shortage in television sets, not every winner can obtain his or her television set immediately. The organization of the lottery therefore asks you to wait five weeks for your prize. For compensation the organization offers you an amount for sure. The compensation is already fixed but not yet known to you. The organization is completely trustworthy. What amount would you regard as minimally acceptable for waiting: f__________? Try to indicate how well you can imagine the above-mentioned television set (f 2400 with a flat-screen and Dolby Stereo). Use a number between 1 (not at all) en 9 (very well imaginable). Circle: Not at all imaginable

Very well imaginable 1 2 3 4 5 6 7 8 9 .......................................................................................................................................................... Imagine: in the same lottery you have won a high-quality long-durable Miele washing machine with low energy use and five-year guarantee by the producer (official price: f 2400). You did not have a washing machine at home; you did your washing at a laundry. What is the minimal amount for which you would like to sell this washing machine f________ ? Because there is a temporary shortage in washing machines, not every winner can obtain his or her washing machine immediately. The organization of the lottery therefore asks you to wait nine weeks for your prize. For compensation the organization offers you an amount for sure. The compensation is already fixed but not yet known to you. The organization is completely trustworthy. What amount would you regard as minimally acceptable for waiting: f__________? Try to indicate how well you can imagine the above-mentioned washing machine (f 2400 long-durable with a low energy use). Use a number between 1 (not at all) en 9 (very well imaginable). Circle: Not at all imaginable 1

2

3

4

5

6

7

8

Very well imaginable 9

212

Appendix A

.......................................................................................................................................................... ‘Some people distinguish between goods. Some goods are pleasant and fun, give pleasure and appeal to your senses, e.g., perfume (type A). Some goods are useful, practical, functional, something that helps you to achieve a goal, e.g., a vacuum cleaner (type B). Do you consider the television set mentioned above as (circle) A: B: C: D:

Mainly type A Mainly type B Both None

Do you consider the washing machine mentioned above as (circle) A: Mainly type A B: Mainly type B C: Both D: None .......................................................................................................................................................... ‘Imagine: you have participated in a lottery. You could not only win monetary prizes but also consumer goods. You were so fortunate to win a television set and a washing machine. It is a flat-screen television set with Dolby stereo to create cinema atmosphere (official price of f2400) and a high-quality longdurable Miele washing machine with low energy use and five-year guarantee (with an official price of f2400, too). You did not have a television set or a washing machine at home. You watched television at friends or parents and did your washing at a laundry. What is the minimal amount for which you would like to sell this television set: _________? What is the minimal amount for which you would like to sell this washing machine: _______? Unfortunately, there is a temporary shortage in prizes, hence, not every winner can obtain his or her prize immediately. The organization of the lottery therefore asks you to wait for your prize. The organization is completely trustworthy. You have two possibilities (circle which you prefer): A:

You get the television set immediately and the washing machine in four weeks.

B: You get the washing machine immediately and the television set in four weeks.’ .......................................................................................................................................................... Imagine you have participated in a lottery. You could not only win monetary prizes but also consumer goods. You were so fortunate to win an exclusive flat-screen television set with Dolby stereo in order to create cinema atmosphere (official price: f 2400). You did not have a television set at home. Because there is a temporary shortage in television sets, not every winner can obtain his or her prize immediately but only after a four weeks delay. The organization of the lottery now asks you to wait an additional two weeks for the television set in order to provide other people with such a television set sooner. They are willing to pay you something for this additional delay. What is the minimal amount to wait two more weeks (in total six weeks) for the television set? f _______ What do you prefer? (circle) A:

An outstanding CD-player in two days

B:

f 500 in one week

Questionnaire used in the first experimental session

213

.......................................................................................................................................................... What do you prefer? (circle) A:

f 100 in 2 days

B:

f 105 in 3 days

C:

f 120 in 14 days

Indicate on the following scale how important the two aspects of the situation (receiving the amount quickly and receiving a high amount) were to you. If you are only interested in receiving a high amount then choose the value 1. If you are only interested to receive the amount quickly, choose the value 9. If you find both aspects equally important, choose the value 5. Circle: Amount of money most important 1

2

3

4

5

6

7

8

Immediacy of receipts most important 9

What do you prefer? (circle) A:

f 100 in 2 days

B: f 150 in 14 days .......................................................................................................................................................... Imagine you are the owner of a little shop where audio- en video-recorders are being sold. You participated in a lottery and you won (for private use) an exclusive flat-screen television set with Dolby stereo to create cinema atmosphere (official price: f 2600). Because there is a shortage in television sets, the organization of the lottery asks you to wait two months for your television set. They are willing to compensate you some money for the delay. What amount would you regard as minimally acceptable for waiting? f __________ Imagine you bought a new computer for f 2700. You need the textprogram and the statistical program for studying. The producer calls to tell you that there are some internal problems and that they can deliver the computer only after two months delay. They offer you a compensation for waiting. Waiting two months for the computer implies that you have to do statistic assignments during this time in one of the ample public facilities. What amount would you regard as minimally acceptable for waiting? f _______ Could you indicate on the following scale from 1 to 9 how much you are interested in computer games (‘1’ is ‘very much interested’ and ‘9’ is ‘not at all interested’)? Circle: Very much interested 1

Not at all interested 2

3

4

Thank you very much for your participation!

5

6

7

8

9

214

Appendix A

In addition to these decision problems, subjects also answered questions concerning their personal background used for Chapter 5, and answered questions concerning the CFCscale and the money-scale also used for Chapter 5. Because the two scales are already presented in the appendices for Chapter 5 they will not be reported here again and we will turn directly to the questions concerning the personal background. Again, a Dutch and an English version will be presented. Fonts have been adjusted from '12' to '11' and dotted lines indicate page breaks in the original version. Items numbers refer to the respective indices for variables introduced in Chapter 5. Questionnaire Now there follow some questions concerning your personal background. Circle the number of the answer that is most applicable to you or fill in the answer. 1. What is your sex? (circle) Male Female

1 2

2. What is your birth date?

19 ______

3. How many brothers and sisters do you have (including step-brothers and -sisters)?

______

4. What is the birth date of your brothers and sisters?

Brothers

Sisters

19 ___ 19 ___ 19 ___ 19 ___

19 ___ 19 ___ 19 ___ 19 ___

----------------------------------------------------------------------------------------------------------------The following questions concern the housing conditions of the family where you were raised. We always ask separately for the housing conditions when you were 5 and 14 years old. (circle or fill in) 5. How many people lived at your home? (including yourself) When you were 5

When you were 14

When you were 5

When you were 14

6. How many sleeping rooms and living rooms were at home?

Questionnaire used in the first experimental session

215

7. In what type of home did you live? When you were When you were 5 14 Apartment or flat 1 1 In a semi-detached house 2 2 In a house 3 3 In a caravan or boat 4 4 ----------------------------------------------------------------------------------------------------------------8. Did you have an own room? When you were When you were 5 14 yes yes no no 9. How strongly do you agree with the following statements concerning the neighborhood where you were raised? (age 14) disagree completely

disagree

do not agree or disagree

agree

agree completely

Other adults than parents told children what they were allowed to do

1

2

3

4

5

My parents had a good reputation in the neighborhood

1

2

3

4

5

People in the neighborhood knew each other

1

2

3

4

5

I visited the neighbors at home

1

2

3

4

5

1

2

3

4

5

I felt at home in the neighborhood I felt insecure when I was at home alone 1

2

3

4

5

Many of my friends came from the same neighborhood

1

2

3

4

5

It was a neighborhood with much fluctuation

1

2

3

4

5

It was a wealthy neighborhood

1

2

3

4

5

We were wealthier than the average 1 2 3 4 neighborhood ----------------------------------------------------------------------------------------------------------------Your parents 10. Are your parents still alive? If not, when did your father die? When did your mother die?

yes no 19 ________ 19 ________

5

1 2

216

Appendix A

11. Did your parents always stay together or did they separate? (divorce)

always stayed together separated your age then was

1 2 ____

12 What is the highest education of your parents? (circle one answer per column) Lower education Lower training for a job Mavo, ulo, mulo141 Havo, mms Vwo, hbs, atheneum, gymnasium Mean training for a job Higher training for a job University Post-academic (PhD etc.) Don't know

father 1 2 3 4 5 6 7 8 9 10

mother 1 2 3 4 5 6 7 8 9 10

13. What was the direction of your parents' education (circle one answer per column)? father mother General 1 1 Teaching 2 2 Language/History 3 3 Agricultural 4 4 Technical/exact 5 5 Transport, communication or traffic 6 6 (Para)-medical 7 7 Economic, administrative, commercial 8 8 Law 9 9 Social/-cultural 10 10 Personal or social support 11 11 Arts 12 12 Public security 13 13 ----------------------------------------------------------------------------------------------------------------14. Were your parents members of a church or religion? If so, which? (circle) No member of church or religion Catholic Dutch reformed Re-reformed Else ________________________ Don't know

141

father 1 2 3 4 5 6

These as well as the following abbreviations denote specific forms of the diverse Dutch educational system. One line comprises roughly equivalent stages.

mother 1 2 3 4 5 6

Questionnaire used in the first experimental session

217

15. How often visited your parent the church or attended religious celebrations? (circle) Never Once or several time per year Once a month Once a week or more often Don't know

father 1 2 3 4 5

mother 1 2 3 4 5

16. What type of profession did your parents have when you were 14? If someone stopped working or died, answer the question for his or her final job before you were 14. (circle) Self-employed Government/governmental institution In a company Does not apply

father 1 2 3 4

mother 1 2 3 4

17. If self-employed, how many employers in their company or institution? (circle one answer per column) father mother None 1 1 1-2 employees 2 2 3-10 employees 3 3 11-24 employees 4 4 More than 24 employees 5 5 18. To what extent were your parents autonomous at their work? father mother Very little 1 1 Little 2 2 Somewhat 3 3 Much 4 4 Very much 5 5 ----------------------------------------------------------------------------------------------------------------19. To what extent were your parents responsible at their work? Very little Little Somewhat Large Very large

father 1 2 3 4 5

mother 1 2 3 4 5

20. Did your mother work? Before your birth In the year of your birth When you were 5 When you were 14

Not 1 1 1 1

Part-time 2 2 2 2

Full-time 3 3 3 3

218

Appendix A

21. How strongly do you agree with the following statements concerning your parents? Independence

disagree completely

disagree

My parents have a harmonious relation

1

My father has much self-control My parents have a good reputation in the neighborhood My parents find self-reliance very important My parents find farsightedness very important My parents find it important that I am wellbehaved My parents find it important that I am successful in what I am doing My parents find it important that I enjoy the things that I am doing My mother has much self-control

agree

disagree completely

2

do not agree or disagree 3

4

5

1

2

3

4

5

1

2

3

4

5

1

2

3

4

5

1

2

3

4

5

1

2

3

4

5

1

2

3

4

5

1 1

2 2

3 3

4 4

5 5

----------------------------------------------------------------------------------------------------------------22. Could you state how much you agree with the following statements over your raising (when you were 14)? disagree completely

disagree

do not agree or disagree

agree

agree completely

1 1

2 2

3 3

4 4

5 5

1

2

3

4

5

1

2

3

4

5

I do not want to disappoint my parents

1

2

3

4

5

My parents knew with whom I was out My parents knew where I was when I was absent from home I often spoke to my parents about things that were important to me I had a strict raising

1

2

3

4

5

1

2

3

4

5

1 1

2 2

3 3

4 4

5 5

At home problems were rather hidden than solved At home I was allowed more than my friends At home I had more tasks than my friends My parents asked with what I was busy at school

-----------------------------------------------------------------------------------------------------------------

Questionnaire used in the first experimental session 23. Could you state how much you agree with the following statements? disagree disagree do not completely agree or disagree

219

agree

agree completely

I have problems to finish unpleasant tasks I like to try out everything even if it is risky I have problems to get out of bed in the morning I have difficulties to stay away from unhealthy things

1

2

3

4

5

1

2

3

4

5

1

2

3

4

5

1

2

3

4

5

I am very disciplined

1

2

3

4

5

I do not worry much about tomorrow

1

2

3

4

5

I can deal with money very well 1 2 3 4 I should try to control myself better when I have fun 1 2 3 4 I like to do things others would consider strange or risky 1 2 3 4 ----------------------------------------------------------------------------------------------------------------Parental engagement at children's raising and education 24. Did you attend a kindergarten? 25. What type of primary school did you attend? several answers possible)

yes no Public Catholic Dutch-reformed 3 Re-reformed Special school (e.g., Montessori, Free School, Frenet, Dalton etc.)

26. When you were attending primary school, did your parents attend parent's evenings? never sometimes 2 often always

5 5 5

1 2 1 2 4 5

1 3 4

27. When you were attending primary school, did your parents supported activities at school (e.g., parental commission, reading books to children etc.)?

never sometimes often always

1 2 3 4

28. When you were attending primary school how many children's books were there at home?

less than 10 about 10 –20 about 20-50 more than 50

1 2 3 4

220

Appendix A

29. Did your parents help you with your homework or controlled it? (circle one answer per column) primary school 1 2 3 4 5

Daily once or several times a week once or several times a month very rarely Never

high school 1 2 3 4 5

----------------------------------------------------------------------------------------------------------------Now we want to know something concerning the strictness of 'rules' at home with respect to several domains. Answer the questions in comparisons to your friends 30. When you were attending primary school how strict were the rules with respect to eating? much less strict

slightly less strict

comparable to friends

slightly more strict

much more strict

Waiting before you were allowed to eat

1

2

3

4

5

Finishing your plate

1

2

3

4

5

1

2

3

4

5

Eating sufficient vegetables or fruit

1 1

2 2

3 3

4 4

5 5

Table manners (eating with hands, sipping etc.)

1

2

3

4

5

Eating sweets

1

2

3

4

5

Eating with the whole family Washing hands before eating

31. When you were attending primary school how strict were the rules with respect to playing?

Making noise What you were allowed to play Where you were allowed to play With whom you were allowed to play When you had to be home

much less strict 1 1 1 1 1

32. When you were attending elementary school who enforced the rules at home?

slightly less strict 2 2 2 2 2

comparable to friends 3 3 3 3 3

slightly more strict 4 4 4 4 4

(mainly) mother (mainly) father mother and father about the same -----------------------------------------------------------------------------------------------------------------

much more strict 5 5 5 5 5 1 2 3

Questionnaire used in the first experimental session

221

33. When you were attending primary school, which tasks did you have to carry out at home? (almost) never

sometimes

often

(almost) always

does not apply

Helping with the garden (e.g., mow the lawn)

1

2

3

4

5

Do the washing up

1

2

3

4

5

Tidy up your room

1

2

3

4

5

Prepare the table

1

2

3

4

5

Do shopping

1

2

3

4

5

Take care of little children

1

2

3

4

5

34. When you were attending primary school, how often did you and your parents in the following leisure activities? never

once a year

once a month

once a week

To the zoo, swimming-bath, forest or beach

1

2

3

4

Cinema or theater

1

2

3

4

Playing games

1

2

3

4

Having serious discussions

1

2

3

4

Accompanying your leisure activities (e.g., sport or music)

1

2

3

4

35. When you were attending high school, how often did you and your parents in the following leisure activities? never

once a year

once a month

once a week

To the zoo, swimming bath, forest, or beach

1

2

3

4

Cinema or theater

1

2

3

4

Playing games

1

2

3

4

Having serious discussions

1

2

3

4

Accompanying your leisure activities (e.g., sport or music)

1

2

3

4

-----------------------------------------------------------------------------------------------------------------

222

Appendix A

36. How strongly stimulated your parents the following types of knowledge and abilities? General development

very little 1

little 2

somewhat 3

strongly 4

very strongly 5

1

2

3

4

5

1

2

3

4

5

1

2

3

4

5

Practical abilities (e.g., timber work, painting, working in the garden, cooking) Social abilities Cultural knowledge and abilities

37. How strongly stimulated your parents the following norms and values? very little little somewhat strongly very strongly Work ethics 1 2 3 4 5 Relating to other children 1 2 3 4 5 Relating to the family 1 2 3 4 5 Relating with friends 1 2 3 4 5 Good manners 1 2 3 4 5 Self-development 1 2 3 4 5 Body health care 1 2 3 4 5 General moral development 1 2 3 4 5 ----------------------------------------------------------------------------------------------------------------Health and life styles 38. What is your weight?

______ kg

39. What is your height?

______ cm

40. How is your general health state compared to other people of your age (circle)?

much worse slightly worse about the same somewhat better much better

1 2 3 4 5

41 How often during the last year did you personally have had contact with your doctor?

None 1-2 times 3-4 times 5 times or more 4

1 2 3

42. How often per week do you have breakfast?

_______ days per week

43. How often per week do you eat junk food as a warm meal?

_______ days per week

44. How often during the day do you eat snacks, e.g., cheese, chips, nuts, etc.?

_______ times per day

45. How often do you eat during the day sweet snacks, e.g., chocolate, cake, sweets etc.?

________ times per day

Questionnaire used in the first experimental session

223

never less than once per week about once per week about every 2 or 3 days at least once per day

1 2 3 4 5

46. How often do you eat fruit?

47. Do you drink alcoholic beverages (wine, beer, drinks or cocktails)?

never less than once per month about one to three times per month once per week several times per week daily

1 2 3 4 5 6

48. If you drink alcohol during the week, how nothing many glasses of alcohol (wine, beer, 1-2 glasses per time drinks or cocktails) do you drink on 3-5 glasses per time average per time? 6-10 glasses per time ----------------------------------------------------------------------------------------------------------------49. If you drink alcohol during the weekend, nothing how many glasses of alcohol (wine, beer, 1-2 glasses per time drinks, cocktails) do you drink on average 3-5 glasses per time average per time (including Friday)? 6-10 glasses per time

1 2 3 4

50. How often are you drunk?

1 2

51. How much do you smoke?

never once or several times per year once or several times per month 3 once or several times per week 4 nothing 1-10 cigarettes/cigars per day 11-20 cigarettes/cigars per day more than 20 cigarettes/cigars per day

52. How old were you when smoking your first cigarette/cigar? 53. How often do you use drugs?

54. How old were you when using drugs for the first time?

1 2 3 4

1 2 3 4

1: never smoked 2: ______ years

never once or several times per year once or several times per month once or several times per week

1 2 3 4

1: never used drugs 2: _______ years

55. How often do or did you engage in sporting activity? (circle one answer per column) primary school high school never 1 1 once or several times per year 2 2 once or several times per month 3 3 once or several times per week 4 4 daily 5 5

now 1 2 3 4 5

224

Appendix A

56. Did you engage in team sport or individually? primary school high school now individually 1 1 1 in a team 2 2 2 does not apply 3 3 3 ----------------------------------------------------------------------------------------------------------------57. How old were you when you had a relationship for the first time? By 'relationship we understand a period of at least 3 month duration where two people see each other as "partner” but do not live together or a married. _______ years 58. How often did you have a relation that lasted at least 3 months? (including current relation)

________ times

59. Are you currently living together with someone?

yes no

1 2

60. How often do use contraceptives when having sex?

never sometimes usually always does not apply (never had sex)

1 2 3 4 5

61. How often did you ask for professional help (e.g., a nurse or doctor) because of a sexually transmitted disease?

never once 2-3 times 4 times or more 4

1 2 3

62. How often did you ever do the following things?

Stayed away from school without valid reasons

never

once

2-3 times

4-6 times

6 times or more

1

2

3

4

5

Used bus, tram, or metro without paying 1

2

3

4

5

Driving a car or motorcycle without driving license

1

2

3

4

5

Driving a car or motorcycle under influence of alcohol of drugs

1

2

3

4

5

Left a café, restaurant etc. without paying

1 1

2 2

3 3

4 4

5 5

Went to a concert, cinema etc. without paying

1

2

3

4

5

Stole something in a shop

1

2

3

4

5

Deliberately crossed a red light or stop sign

Questionnaire used in the first experimental session

225

----------------------------------------------------------------------------------------------------------------Accidents The following questions concern traffic accidents. They concern accidents where you were involved as pedestrian, automobilist, or when riding a bicycle or motorcycle. 63. How frequently do you participate in traffic (circle one answer per column): Pedestrian Cyclist Car driver or motor cyclist 0-3 times per month 1 1 1 once a week 2 2 2 daily, less than 30 minutes 3 3 3 daily, between 30 minutes and 1 hour 4 4 4 daily, more than 1 hour 5 5 5 64. How many accidents did you ever have? (circle) Pedestrian 1 2 3 4 5

None one 2–3 4-5 6 or more

Cyclist 1 2 3 4 5

Car driver or motor cyclist 1 2 3 4 5

People can also have other accidents, e.g., falling, almost drown, get burned (including fire and fireworks), get injured from machines, (e.g., sawing or drilling), or almost suffocate. 66. How many accidents did you ever have where you were considering to ask for or did ask for medical help . none one 2-3 4-5 6 or more Sport of physical exercising 1 2 3 4 5 Falling (falling from somewhere or simply falling down) 1 2 3 4 5 Burning (gas, fire etc.) 1 2 3 4 5 Drowning, poisoning, or suffocation 1 2 3 4 5 Tools and machines 1 2 3 4 5 Weapons or physical violence 1 2 3 4 5 ----------------------------------------------------------------------------------------------------------------Leisure activities 67. Ho often do you do the following things? never Visit museum or theater

1

several times per year 2

several times per month 3

several times per week 4

daily

Reading literary books

1

2

3

4

5

Reading newspapers

1

2

3

4

5

Reading extra (not required) literature for your study

1

2

3

4

5

5

226

Appendix A

Now we would like to ask something about the making of music. 68. What for an instrument do or did you play (leave empty if this question does not apply)? Instrument when you were 5 when you were 14 Now 69. How often do or did you practice?

Never Several times per year Several times per month Several times per week Daily, less than 30 minutes Daily, more than 30 minutes Does not apply

when you were 5 1 2 3 4 5 6 7

when you were 14 1 2 3 4 5 6 7

now 1 2 3 4 5 6 7

70. When did you seriously start to learn to play an instrument, by taking courses or by seriously studying yourself? ___________ years 71. What is your average gross month income? (including work income or regular and irregular payments from your parents for, e.g., insurances, expenses for studying, books, vacations, or money at Christmas or your birthday) f _________ 72. How often do you exceed your bank account by more than f 500?

never 1 sometimes 2 frequently 3 (almost) always 4 ----------------------------------------------------------------------------------------------------------------Education 73. How carefully did you do your homework at high school?

bad casual average carefully very carefully

1 2 3 4 5

74. If you compare your final results at school with what you think you could have achieved, how do you consider your results? Results could have been much better Results could have been somewhat better Results were about what could have been expected (achieved) You had to work very hard to achieve your results

1 2 3 4

Questionnaire used in the first experimental session 75. How carefully do you prepare courses and exams?

227

bad casual average carefully very carefully

1 2 3 4 5

76. How often do you party before exams?

never sometimes often (almost) always 4

1 2 3

77. How often did you following extra courses for your study (e.g., extra methods courses or courses for which you did not receive credits)?

never sometimes often (almost) always

1 2 3 4

78. Did you ever apply? (for things that were important to you, e.g. , a paid job or traineeship)

yes no

1 2

79. If so, how carefully did you prepare?

bad casual average carefully very carefully

1 2 3 4 5

80. Did you have to redo classes at school? at primary school at high school 81. How much delay in your study is there?

yes 1 1

no 2 2

__________ months

82. What is your field of studying?

_______________________

83. What are your hobbies?

1. _________________________ 2. _________________________ 3. _________________________

84. What is your political preference?

85. What is your nationality? Thank you very much for your participation!

1. Right-wing 2. Slightly right-wing 3. Moderate 4. Slightly left-wing 5. Left-wing 6. No preference _____________________________

The logistic regression model

229

Appendix B The logistic regression model

The logistic regression model is used to study the impact of a set of covariates (independent variables) on dichotomous outcome variable (dependent variable). In standard logistic regression analysis (used in Chapters 2, 3, 4 and 5) observations are assumed to be independent. In this appendix some of the models underlying assumptions will be laid out. The following discussion of the logistic regression model is based on Hosmer and Lemeshow (1989). The logistic regression model differs form the linear regression model with respect to the underlying parametric model and in some assumptions. As to the parametric model, in a logistic regression model the conditional mean E(Yx), i.e., the expected value of Y given x, is limited between 0 and 1 and approaches 0 and 1 gradually. That is, the change in E(Yx) per unit-change in x becomes progressively smaller when approaching 0 or 1. The curve of this conditional mean is S-shaped and resembles a plot of a cumulative distribution function of a random variable. A prominent underlying distribution function for E(Yx) is the logistic distribution function:

π ( x) =

e β 0+ β 1x 1 + e β 0+ β 1 x

This distribution function can be transformed into the so-called logit transformation:  π ( x)  g ( x) = ln   1 − π ( x) 

= β0 + β1x Now the logit g(x) is linear in its parameters, may be continuous, and can take all values from -∞ to +∞. The second difference in comparison to linear regression models is that in a linear regression is assumed that the errors follow a normal distribution. In a logistic regression these errors are assumed to follow a binominal distribution and the analysis will be based on this statistical distribution. As to model fit, linear regression uses least squares to estimate the unknown parameters in the model. Logistic regression maximizes a likelihood function, such that the resulting estimates of β0 and β1 agree most closely with the data. Maximizing the function requires iterative measures. Standard statistical software such as SPSS and STATA displays log likelihood values obtained during the iterative process.

230

Appendix B

To asses significance of a variable observed values of the dependent variable (in the context of this book: preference between choice options) are compared to the predicted values with and without the variable in question. For this comparison the likelihood functions of the model under study and the saturated model are used:  likelihood of the current model  D = −2 ln   likelihood of the saturated model  It is necessary to multiply the expression by minus two is to obtain a quantity with a known distribution. Now, the values of D with and without the respective variable are compared. The change in D after inclusion of the respective variable is obtained by: G = D(for the model without the variable) - D(for the model including the variable) This statistic G follows a chi-square distribution. An alternative measure for the significance of individual variables are Wald statistics. Here the coefficient β1 is divided by its standard error: W =

β1 SE ( β 1)

In STATA, this value is reported as z-scores (see Chapter 7). To obtain the values given as 'Wald' in SPSS (see Chapters 2, 3, 4, and 5) z-scores have to be set to the power 2. The coefficients β0, β1, …, βn represent the change in the log odds for an increase of '1' unit in the respective independent variable x. The log odds for a continuous independent variable for a change in c units in x results from the logit difference: g(x + c) - g(x) = cβ1 From this quantity odds ratios can be obtained. This odds ratio is important since it approximates how likely or unlikely it is for the outcome to be present given the respective values of x. Hence, often the odds ratio approximates the relative risk. The way the odds ratio is obtained from the log odds depends on whether the independent variable is dichotomous, polytomous, or continuous. For example, for continuous variables the odds ratio is obtained by exponentiating the logit difference. Some final general remarks concerning the two statistical programs used in this dissertation might also be in order. Coefficients and standard errors are identical for SPSS and STATA, as is the chi-square value. However, whenever I compared respective values of Pseudo R2 between the two programs STATA generally reports lower levels of Pseudo R2. Note, that SPSS standardly offers two values for the Pseudo R2 ('Cox & Snell' and 'Nagelkerke', the latter being usually slightly higher). Further, STATA generally uses smaller values for the log likelihood.

Samenvatting (Summary in Dutch) 1 Inleiding

Hoe vaak maken we niet keuzes die niet alleen nu, maar ook in de toekomst consequenties hebben. We houden bij dat soort keuzes rekening met zowel de directe kosten en baten, alsook de toekomstige. We verschillen echter wat betreft onze bereidheid met zulke toekomstige consequenties rekening te houden. In sommige situaties zijn we ongeduldiger en letten we meer op de baten die we direct kunnen realiseren. Ook zijn er verschillen tussen mensen. De ene mens is ongeduldiger dan de ander. Dit boek behandelt drie vragen over intertemporele keuzes. Allereerst wordt onderzocht wanneer mensen vooral op de korte termijn zijn gericht en wanneer juist op de lange termijn. Vervolgens behandelt het de vraag welke mensen de langetermijneffecten vooral voor ogen houden, oftewel, welke mensen geduldiger zijn om op hun beloning te wachten. Tenslotte wordt gekeken of de geobserveerde gedragingen met betrekking tot intertemporele keuzes gegeneraliseerd kunnen worden naar risicovolle, interpersoonlijke en ruimtelijke keuzes. Deze samenvatting staat stil bij de hoofdstukken in dit boek. Ze bespreekt de belangrijkste empirische bevindingen en doet enkele suggesties voor toekomstig onderzoek. Hoofdstuk 1 introduceerde het standaard economische model dat aangeeft hoe intertemporele keuzes op een rationele manier gemaakt zouden moeten worden, maar ook hoe rationeele mensen deze keuzes daadwerkelijk maken. De aanname in dit ‘Discounted-Utility’-model (DU-model) is dat mensen directe baten meer op prijs stellen dan toekomstige. De waardering daalt naarmate de baten later in de tijd geconsumeerd kunnen worden. Voor ieder tijdstip van gelijke lengte zou volgens het model deze waardering met een gelijke hoeveelheid af moeten nemen. Volgens dit model is deze afname (discontovoet) gelijk of nagenoeg gelijk aan de op dat moment geldende marktrente. Empirisch onderzoek naar gedrag in intertemporele keuzesituaties toont aan dat het DU-model tal van anomalieën kent. Zo is er van een constante discontovoet geen sprake. Mensen wegen uitstel op korte termijn veel zwaarder dan een identieke vertraging op lange termijn. Verder is gebleken dat mensen veel geduldiger zijn als het gaat om grote bedragen dan om kleine bedragen. Voor kleine bedragen verlangen zij een verhoudingsgewijs grotere compensatie. Eveneens in strijd met het DU-model is dat mensen voor uitkomsten van gelijke waarde verschillende discontovoeten hanteren. Zo is de discontovoet van onverwachte winsten veel groter dan die van een vast inkomen. Tenslotte toont onderzoek aan dat mensen uitstel van bepaalde baten erger vinden dan dat zij het onverwachts met een zelfde tijdseenheid naar voren halen van die baten waarderen. Zo vragen mensen meer compensatie om ontvangst van een consumptiegoed een jaar uit te stellen dan ze bereid zijn te betalen om dit goed een jaar eerder te ontvangen. Dit is niet in overeenstemming met het model omdat dit enkel de waardering van het object en uitstel in de tijd (absoluut) in aanmerking neemt. Een tijdverschil van een jaar zou dus altijd tot een gelijke verschil in waardering leiden onafhankelijk of het om een vertraging of om een eerdere ontvangst gaat.

232

Samenvatting

Een vaak gehoorde reactie van economen is een trivialisering van deze anomalieën. Zij wijzen bijvoorbeeld op het feit dat de anomalieën experimenteel in hypothetische situaties gegeneraliseerd zijn en dat het in de realiteit om echte bedragen gaat. In de realiteit zal men zich wel rationeel gedragen. Verder zouden geobserveerde irrationele gedragingen op toeval berusten en bij aggregatie van de bevindingen verdwijnen. Hoofdstuk 1 besteedde ruime aandacht aan dit soort argumenten. De conclusie was dat de anomalieën op grond van zowel theoretische als empirische argumenten nadere bestudering behoeven. Ook werden enkele alternatieve modellen (e.g. Loewenstein & Prelec 1992) besproken, maar deze blijken net als het standaard model tekorten te vertonen. Zo veronderstellen ze dat voor objecten met een gelijke waarde altijd een zelfde discount rate gebruikt wordt. Dit is in strijd met onderzoek dat aantoont dat de discontovoet varieert per context. Na beschrijving van deze anomalieën werden de drie eerder genoemde onderzoeksvragen geïntroduceerd. Verder werd een alternatief aangedragen: een framingtheorie gekoppeld aan een model voor keuzegedrag (het ‘discriminatiemodel’), beide ontwikkeld door Lindenberg. Het model veronderstelt dat mensen beperkt rationeel zijn als gevolg van cognitieve beperkingen. Gegeven deze cognitieve beperkingen zijn zij er wel op uit om hun nut te maximaliseren. Volgens de framing-theorie letten mensen verschillend op verschillende aspecten van een keuzesituatie, afhankelijk van het doel dat zij willen bereiken. Omdat doelen deze selectieve aandacht (frames) aansturen, kunnen wij vereenvoudigend een frame ook aanduiden met een bepaald doel. Bij intertemporele keuzes kunnen twee frames onderscheiden worden: ‘het zo snel mogelijk ontvangen van de beloning’ (oftewel ‘snelheid’) en ‘het ontvangen van een zo groot mogelijke beloning’ (of kortweg ‘winstmaximalisatie’). 2 Wanneer gedragen zich mensen bijziend?

In hoofdstuk 2 werden enkele assumpties van de framing-theorie getoetst. Bijvoorbeeld de vraag of voorkeuren inderdaad afhankelijk zijn van de saillantie van frames en of deze saillantie afhangt van de mate waarin het frame een bepaalde optie als beste keuze aanwijst. Proefpersonen kregen een eenvoudig intertemporele keuze voorgelegd, bestaande uit een temporeel en een geldelijk attribuut. Zij moesten bijvoorbeeld kiezen of zij liever ƒ100 nu of ƒ110 over 4 weken willen ontvangen. Vervolgens dienden ze aan te geven hoe belangrijk zij elk van de twee attributen vonden. De redenering was dat deze attributen beschouwd kunnen worden als frames. De beoordeling ervan kan daarom als saillantie van frames geïnterpreteerd worden. Twee experimenten toonden dat de saillantie inderdaad een goede voorspeller is van de uiteindelijke voorkeur. Mensen in het frame waarin snelheid het belangrijkste doel is, hadden een voorkeur voor een snellere maar kleinere uitkomst en vice versa. Verder werd gekeken of ook het discriminerend vermogen van de frames bepalend is voor de voorkeur van mensen. Hier waren de uitkomsten gemengd. In een experiment waarin de frames niet discrimineerden, waren de voorkeuren, zoals verwacht, gelijk verdeeld. Echter, ook in het experiment waar het temporele attribuut veel beter tussen de opties discrimineerde dan het attribuut voor uitkomsten, en de theorie dus een voorkeur voor de eerste optie voorspelt, waren de voorkeuren gelijk verdeeld. In een ander experiment werd geprobeerd veranderingen in voorkeur tussen keuzeproblemen aan veranderingen in het onderscheidingsvermogen van frames te relateren. Dat wil zeggen, als in keuzeprobleem A de grotere maar latere optie sterker

Samenvatting

233

geprefereerd wordt dan in keuzeprobleem B, dan moet, in vergelijking met het temporele attribuut, het uitkomst attribuut, bijvoorbeeld ƒ100, in A een betere onderscheiding tussen de opties bieden dan in B. Het experiment vertoonde samenhang tussen voorkeur en onderscheidingsvermogen van frames, maar de correlaties waren niet significant. Omdat er geen veranderingen in de voorkeur tussen de keuzeproblemen A en B waren, blijft het overigens de vraag of deze bevinding wijst op tekortkomingen van het design of van het model. Omdat de andere voorspellingen van het model over het algemeen juist waren, ligt de eerste verklaring meer voor de hand. Om andere voorspellingen van het model te toetsen werden onderscheidingsvermogen en saillantie van de frames ook indirect experimenteel gemanipuleerd. Daartoe werden enkele experimenten uit onderzoek naar consumentengedrag uitgevoerd. Rationaliteit impliceert gewoonlijk dat de relatieve aantrekkelijkheid van twee opties niet van alternatieven afhankelijk is. Echter, volgens het discriminatiemodel kunnen andere opties wel het onderscheidingsvermogen van frames beïnvloeden. Als gevolg daarvan kunnen ook de saillantie van frames en de voorkeuren tussen opties veranderen. Deze voorspelling is in overeenstemming met consumentenonderzoek naar contexteffecten. Een serie experimenten laat zien dat de toevoeging van opties die het onderscheidend vermogen van de frames veranderen een voorkeur tweegbrengt voor het frame dat het beste discrimineert. De optie die volgens dit frame het beste scoort, wordt vaker geprefereerd. In hoofdstuk 3 werd nagegaan welke frames tot kortzichtig gedrag leiden en welke frames lange(re)-termijnoverwegingen saillanter maken. De voorspelling was dat hedonische aspecten van consumptie eerder tot het frame van snelheid leiden dan functionele overwegingen. Omdat hedonische gebruik direct nut oplevert, is dit gebruik ceteris paribus saillanter dan het gebruik van functionele goederen. De voorspelling is daarom dat mensen het erger vinden om hedonische dan functionele goederen uit te stellen. In drie experimenten werd deze hypothese bevestigd. In één experiment was dit effect zo sterk dat mensen zelfs liever direct een hedonisch goed ontvingen dan direct een waardevoller functioneel goed. Deze bevindingen zijn in tegenspraak met zowel het ‘Discounted Utility’-model alsook met zijn meest gebruikte tegenhanger, het hyperbolisch discounting-model. Verder werd onderzocht of de voorkeur voor hedonische goederen blijft bestaan als mensen aan moeten geven hoeveel geldelijke compensatie zij verlangen bij uitstel van hedonische dan wel functionele goederen. In de literatuur wordt het noemen van geldbedragen vaak als equivalent voor ongeduld gehanteerd. Als dit juist is, dan zou de onmiddellijke voorkeur voor hedonische goederen ook in de geldelijke compensatie tot uitdrukking moeten komen. De framing-theorie veronderstelt echter dat met de introductie van geld de voorkeur voor hedonische goederen verzwakt wordt. Proefpersonen zullen bedenken dat geld ook voor andere dan hedonische goederen ingezet kan worden. Dit maakt dat het oorspronkelijke ‘snelheidsframe’ (aangewakkerd door het hedonische goed) verdwijnt en daarmee ook de voorkeur voor dit type goed. De resultaten bevestigden de framing-assumptie. In hoofdstuk 4 werd opnieuw de voorkeur onderzocht voor hedonische en functionele goederen in de tijd. Dit keer werd gekeken hoe gevoelig de voorkeuren zijn voor de manier waarop de goederen worden beschreven. In tegenstelling tot wat verwacht zou kunnen worden, blijkt de levendigheid van de beschrijvingen in het algemeen geen invloed te hebben op de voorkeur. Enkele onderzoekers veronderstellen dat dit bij

234

Samenvatting

intertemporele keuzes wel eens anders zou kunnen zijn. Hun voorspellingen zijn echter tegenstrijdig. Waar sommigen redeneren dat de levendigheid mensen ongeduldiger maakt, veronderstellen anderen het tegendeel. In lijn met framing-theorie veronderstelt dit boek dat levendige beschrijvingen bij hedonische goederen het ‘snelheidsframe’ aanwakkert, terwijl een levendige beschrijving van een functioneel goed weinig effect heeft. Eerst werd een experiment ontworpen om te testen of levendigheid een vergelijkbare invloed heeft op de keuze welk goed men liefst het eerste ontvangt, het hedonische of het functionele goed. Uitkomst was dat in deze keuzesituatie levendigheid een veel grotere invloed had op de voorkeur van hedonische goederen dan op die van functionele goederen. Net als in hoofdstuk drie verdween dit effect zodra de respondenten compensatie voor vertraagde ontvangst van de goederen in geld moesten uitdrukken. De proefpersonen verlangden zelfs een grotere compensatie voor het uitstellen van functionele goederen. 3 Wie gedraagt zich bijziend?

In hoofdstukken de 2 tot en met 4 is onderzocht wanneer mensen zich myopisch gedragen. In hoofdstuk 5 is vervolgens onderzocht wie er vooral geneigd zijn om zich myopisch te gedragen en wie niet. Omdat geduld in veel situaties voordelig is, werd geredeneerd dat ouders hun kinderen geduld proberen bij te brengen. Zij proberen hun kinderen te leren om hun natuurlijke, myopische gedrag te controleren. Ouders verschillen echter in hun vaardigheden, prikkels en restricties om dat te doen. Zo hebben ouders die meer verantwoordelijk gehouden worden voor het gedrag van hun kinderen meer prikkels om hun kinderen 'goed' op te voeden. Als een huishouden veel kinderen telt, is er per kind minder tijd om zelfcontrole aan te leren. In dit boek worden data met gedetailleerde informatie over de opvoeding van de respondenten in verbinding met een psychologische schaal gebracht die meet in hoeverre de respondenten met toekomstige gevolgen van hun acties rekening houden. Het blijkt inderdaad zo te zijn dat respondenten van wie de ouders meer prikkels hadden om hun kinderen 'goed' op te voeden, geduldiger zijn. Ook is het zo dat geduldigere ouders, gemeten door hogere opleidingsniveau of een beroep waarvoor veel zelfcontrole nodig is, geduldigere kindren hebben. De samenstelling van de familie blijkt echter weinig invloed op het geduld van de respondenten te hebben. De manier waarop ouders hun kinderen gedrag leren, heeft daarentegen wel een effect op hun geduld. Respondenten waar ouders een striktere opvoeding toepasten en die door hun ouders meer stimulatie ondervonden, gaven aan sterker met toekomstige gebeurtenissen rekening te houden. Onderzoek naar oorzaken van verschillen in geduld is alleen zinvol als deze verschillen gevolgen voor daadwerkelijk gedrag hebben. Daarom werd geduld, zoals volgens de psychologische schaal gemeten, ook gebruikt om mogelijke gevolgen van myopisch gedrag te voorspellen. Voorbeelden waar zulke verschillen in geduld gevolgen kunnen hebben, zijn gezonde levenswijze, hoe serieus mensen op school of universiteit studeren en hoe groot de kans is dat mensen zich risicovol gedragen. Voor al deze onderwerpen waren er verschillen tussen mensen die als geduldig of ongeduldig ingedeeld zijn. Geduldige mensen consumeerden minder alcohol en drugs, rookten minder vaak, sportten vaker, deden hun huiswerk op school zorgvuldiger, volgden vaker extra lessen op de universiteit en vertoonden ook minder vaak afwijkend gedrag.

Samenvatting

235

Deze verschillen zijn, zoals gezegd, gebaseerd op een psychologische schaal om geduld te meten. In experimenten naar intertemporele voorkeuren is het echter gebruikelijker om geduld te meten door middel van discontovoeten te berekenen. Daarbij wordt de financiële compensatie gemeten die mensen voor het uitstellen van geldbedragen vragen. Al in hoofdstukken 3 en 4 kwam aan de orde dat het vragen aan mensen om hun gevoelens van ongeduld in geldbedragen te vertalen ook hun framing en hun ongeduld beïnvloedt. Het is dan ook maar de vraag of het berekenen van discontovoeten een nuttige meetinstrument voor ongeduld is. Om dit punt verder te onderzoeken, werd getoetst of socialisatie door de ouders ook de discontovoeten beïnvloedt en of men discontovoeten even goed als voorspeller voor daadwerkelijk gedrag kan gebruiken als de psychologische schaal. De resultaten bevestigden de bevindingen van de hoofdstukken 3 en 4. Als geduld door middel van geldbedragen wordt gemeten, is geduld veel slechter voorspelbaar dan geduld zoals gemeten via de psychologische schaal. Bovendien zijn discontovoeten veel slechter in staat dan de psychologische schaal om daadwerkelijke gedrag te voorspellen. Uiteindelijk was er ook geen sprake van een samenhang tussen de twee meetinstrumenten. Ten eerste waren de meetinstrumenten niet met elkaar gecorreleerd, ten tweede was er niet een enkele factor die geduld voor beide meetinstrumenten voorspelde. Zo scoren scoren kinderen van ouders met een hogere status op de psychologische schaal als minder ongeduldig dan andere kinderen, maar de status van hun ouders had geen invloed op discontovoeten. Deze resultaten maken opnieuw duidelijk dat verschillende manieren om ongeduld te meten in feite verschillende aspecten van ongeduld meten. Het berekenen van discontovoeten uit de genoemde financiële compensatie voor uitgestelde geldbedragen laat belangrijke aspecten van bijziend gedrag buiten beschouwing. Framing-effecten zijn dus ook van belang voor meetinstrumenten. 4 Het generaliseren van effecten

In hoofdstuk 6 werden de normatieve concepten en de empirische afwijkingen daarvan voor risicovolle keuzes en intertemporele keuzes besproken. Aangetoond werd dat zowel de normatieve axioma’s alsook de empirische afwijkingen in allebei keuzedomeinen zeer sterk op elkaar lijken. In dit boek wordt geredeneerd dat deze overeenkomst niet toevallig is, maar een algemeen menselijk gedrag weerspiegelt: mensen beoordelen uitkomsten ten aanzien van een multi-dimensioneel referentiepunt. Zij prefereren uitkomsten die zo groot mogelijk zijn, zeker zijn, 'nu' verkrijgbaar zijn, die zij zelf ontvangen en die 'hier' plaatsvinden. Als een of meerdere van deze aspecten niet geldig zijn, worden de uitkomsten minder sterk gewaardeerd. De empirische overeenkomsten in risicovolle en intertemporele keuze zijn daarom niet beperkt tot deze twee gebieden, maar zullen ook in ruimtelijke en interpersoonlijke keuzes optreden. Bovendien zal discounting in alle vier domeinen op soortgelijke wijze gebeuren en tot identieke discount-functies moeten leiden. In hoofdstuk 7 werden enkele anomalieën voor risicovolle en intertemporele keuze gerepliceerd en er werd getoetst of deze ook bij ruimtelijke en interpersoonlijke keuzes optreden. Daarvoor was het nodig adequate stimuli voor deze twee keuzedomeinen te creëren. In ruimtelijke keuze moesten mensen kiezen tussen winkels en supermarkten met verschillende hoeveelheden artikelen. Deze supermarkten konden of

236

Samenvatting

nieuw geopend of voorgoed gesloten worden. Voor sociale afstand moest een eigen meetinstrument ontwikkeld worden om deze dimensie vergelijkbaar te maken met de eenvoudig kwantificeerbare dimensies als risico, tijd en ruimte. Voor dit meetinstrument zijn fictieve profielen van personen gemaakt. Deze profielen zijn gedefinieerd door sociaal-economische status, leeftijd, hobby's, politieke voorkeur en nationaliteit. De gewichten van de afzonderlijke dimensies voor de uiteindelijke sociale afstand zijn bepaald door middel van een pretest. Daarin moesten mensen aangeven hoe belangrijk ze de afzonderlijke dimensies vonden. Zo beschouwden mensen in de pretest in verschillende situaties hobby's als drie keer zo belangrijk als nationaliteit. Een profiel met dezelfde hobby als de 'gemiddelde respondent', maar met een verschillende nationaliteit, kende dus een minder grote sociale afstand dan een profiel met een identieke nationaliteit maar met verschillende hobby's. Om te controleren of de door ons gebruikte maat ook daadwerkelijk in overeenstemming met de perceptie van de respondenten was, werd de respondenten gevraagd ook hun inschatting van de sociale afstanden van de profielen aan te geven. Zoals voorspeld, traden veel van de effecten in risicovolle en intertemporele keuzes ook in ruimtelijke keuzes op. Voorbeelden zijn het common ratio effect en het sign effect. Dus is er inderdaad evidentie voor een algemeen menselijke neiging objecten minder sterk te waarderen als zij op grotere afstand van het referentiepunt verkrijgbaar zijn. Sterker nog, het feit dat er voor de ruimtelijke dimensie een common ratio effect wordt gevonden, maar geen common difference effect, vormt een aanwijzing dat de discount-functie voor ruimtelijke afstand exponentieel is in plaats van hyperbolisch, zoals bij intertemporele keuzes, of S-shaped, zoals bij risicovolle keuzes. Voor interpersoonlijke keuzes waren de resultaten minder duidelijk. Ten eerste leek de waarneming van de stimuli door de mannen anders te zijn dan verwacht. De stimulus set (dat wil zeggen de opties waartussen mensen moesten kiezen) moest daarom aangepast worden. Ten tweede leek de waarneming van de stimuli door de vrouwen wel in overeenstemming met de verwachtingen te zijn, maar geen van de verwachte effecten trad op. Ondanks het feit dat voor hen de stimulus set minder goed werkte, konden daarentegen voor mannen wel enkele effecten aangetoond worden. Zo traden de common ratio effect en de common difference effect op voor de sociale dimensie, wat suggereert dat de discount-functie voor sociale afstand hyperbolisch in plaats van exponentieel is. Toekomstig onderzoek zou moeten ophelderen of dit verschil tussen mannen en vrouwen een artefact is, waar verschillen in de waarneming van de keuzeopties aan ten grondslag liggen, of dat er daadwerkelijk een verschil tussen discounting van sociale afstand tussen mannen en vrouwen bestaat. 5 Het meten van ongeduld

In dit proefschrift gaat het heel vaak daarom hoe ongeduld op de beste manier kan worden gemeten en of verschillende meetinstrumenten voor ongeduld tot soortgelijke resultaten zullen leiden. De meest gebruikelijke meetinstrument voor ongeduld, uitstellen van hypothetische geldbedragen en het vragen van adequate financiële compensatie voor dit ongemak, leken veel minder betrouwbaar en leidden tot andere resultaten dan het eenvoudige vragen aan mensen welke van twee uitkomsten ze eerder willen ontvangen. De eerder gebruikte framing-theorie voorspelt dat het vragen van geldbedragen ertoe leidt dat mensen eerder op functionele dan op hedonische aspecten letten en zich dus geduldiger gedragen. Vragen over het meten van ongeduld zullen dus als theoretische

Samenvatting

237

vragen moeten worden beschouwd. De theorie voor het meten en voor het voorspellen van ongeduld moet dezelfde zijn. Naast deze theoretische overwegingen, is het niet echt betrouwbaar om ongeduld alleen via discount rates te meten. Het is heel moeilijk voor mensen adequate bedragen te bepalen voor situaties waarmee zij helemaal geen ervaring hebben. In een eerdere experiment over de delay/speed-up asymmetrie vermeldde Loewenstein (1988: 208) het verbazingwekkende resultaat dat respondenten een uitgestelde diner gemiddeld iets sterker waarderen dan een onmiddellijk te consumeren diner: 'This result again testifies to the unimportant role played by time delay when subjects are simply asked how much they would pay to consume at different points in time'. 6 Suggesties voor verder onderzoek

Ten slotte wordt ingegaan op enkele vragen en problemen die in dit boek niet voldoende aan de orde kwamen. Er wordt bijvoorbeeld onderzocht wanneer mensen zich bijziend gedragen en wie zich bijziend gedraagt. Het zou een natuurlijke vervolgstudie kunnen zijn om deze resultaten met elkaar te verbinden. Men kan de persoonlijke meetinstrumenten uit hoofdstuk 5 gebruiken om voorspellingen te doen over de de keuzes tussen de consumptiegoederen die in de hoofdstukken 3 en 4 zijn behandeld. In hoofdstuk 7 bleef het onduidelijk of de verschillende resultaten tussen mannen en vrouwen door de stimulus set tot stand kwamen of dat mannen en vrouwen verschillende discount rates voor sociale afstand toepassen. Toekomstig onderzoek zou hier nuttig zijn. In hoofdstuk 6 zijn verder enkele overeenstemmingen tussen risicovolle en intertemporele keuze genoemd die in hoofdstuk 7 niet worden getoetst, bijv. de delay/speed-up asymmetrie. Het zou uitermate nuttig zijn om deze effecten ook voor de andere keuzedomeinen te onderzoeken. In hoofdstuk 5 werd beredeneerd dat bepaalde persoonskenmerken, bijvoorbeeld prikkels om vaardigheden te leren, het geduld van de respondenten beïnvloeden. Deze argumenten zullen in principe ook voor andere soorten van voorkeuren, bijvoorbeeld tussen risicovolle of ruimtelijke keuzes, gelden. Toekomstig onderzoek zou kunnen worden verricht naar overeenkomsten tussen bepaalde persoonskenmerken in alle drie keuzedomeinen. Een verwante vraag is of dezelfde mensen ongeduldig en risico-avers zijn en bovendien uitkomsten – ruimtelijk gezien – graag hier willen ontvangen. Men zou dus niet alleen discount functies met elkaar kunnen vergelijken, maar ook of het temporele attribuut, het kansattribuut en het ruimtelijke attribuut met elkaar uitgewisseld kunnen worden. Stel iemand prefereert (ƒ100; nu) boven (ƒ110, in vier weken). Verhoogt deze voorkeur zijn of haar voorkeur voor het spel (75; 1) tegenover het spel (100; 0.8)? Een laatste vraag met betrekking tot overeenstemming tussen de drie keuzedomeinen is die naar de effecten in geval er keuzes met meerdere attributen aan de respondenten voorgelegd worden. Eerder onderzoek naar intertemporele keuzes toont aan dat er relatief minder discounting is als er een kansattribuut aan een intertemporele optie toegevoegd wordt. Soortgelijke effecten kunnen verwacht worden, als het kansattribuut wordt vervangen door een ruimtelijke attribuut. Als het om toepassingen van de in dit boek gevondenen resultaten gaat, dan zou het nuttig zijn om te kijken of mensen ook een grotere voorkeur hebben hedonische goederen onmiddellijk te ontvangen dan functionele goederen, wanneer mensen daadwerkelijke keuzes tussen echte goederen moeten maken. Dit is naar verwachting met

238

Samenvatting

name het geval als de goederen levendig gepresenteerd worden en mensen geen expliciete financiële trade-offs moeten maken. Bijvoorbeeld valt te verwachten dat vaker (excessieve) inkopen met een creditkaart vaker worden gedaan als het om hedonische goederen gaat. Verder valt te verwachten dat mensen eerder acties met negatieve langetermijngevolgen ondernemen als daar onmiddellijke hedonische consumptie tegenover staat. Een andere voorbeeld is dat mensen makkelijker hun lange-termijnoverwegingen volgen als het om functionele overwegingen gaat. Hedonische consumptie zou daarentegen vaker op krediet gefinancierd worden. Het lijkt ook mogelijk om de persoonlijke kenmerken die in hoofdstuk 5 geduld blijken te beïnvloeden in analyses omtrent andere toepassingsgebieden te hanteren. In veel economische analyses wordt tijdvoorkeur gebruikt. Als mensen als geduldig ingeschat worden, zullen zij bijvoorbeeld meer sparen en ander gedrag met betrekking tot familie en gezin vertonen.

Curriculum Vitae Alexander Gattig was born in Bremen, Germany, on October, 23rd, 1969. He completed his secondary education in 1989. From 1989 to 1995 he studied Sociology, Economics, and Philosophy at the University of Bremen. In 1996 he obtained his Master's degree in Sociology. From September 1995 he worked as a Ph.D. student at the Interuniversity Center for Social Science Theory and Methodology (ICS). His dissertation project resulted in this book. In 1997 he spent a four-weeks traineeship at the 'Center for Advanced Study in the Behavioral Sciences' and the Graduate School of Business at Stanford University, Palo Alto, USA.

ICS dissertation series

The ICS-series presents dissertations of the Interuniversity Center for Social Science Theory and Methodology. Each of these studies aims at integrating explicit theory formation with state-of-the-art empirical research or at the development of advanced methods for empirical research. The ICS was founded in 1986 as a cooperative effort of the universities of Groningen and Utrecht. Since 1992, the ICS expanded to the University of Nijmegen. Most of the projects are financed by the participating universities or by the Netherlands Organization for Scientific Research (NWO). The international composition of the ICS graduate students is mirrored in the increasing international orientation of the projects and thus of the ICS-series itself. 1) 2) 3) 4) 5) 6)

7)

8) 9)

10) 11) 12) 13)

14)

C. van Liere, (1990), Lastige Leerlingen. Een empirisch onderzoek naar sociale oorzaken van probleemgedrag op basisscholen. Amsterdam: Thesis Publishers. Marco H.D. van Leeuwen, (1990), Bijstand in Amsterdam, ca. 1800 - 1850. Armenzorg als beheersings- en overlevingsstrategie. ICS dissertation, Utrecht I. Maas, (1990), Deelname aan podiumkunsten via de podia, de media en actieve beoefening. Substitutie of leereffecten? Amsterdam: Thesis Publishers. M.I. Broese van Groenou, (1991), Gescheiden Netwerken. De relaties met vrienden en verwanten na echtscheiding. Amsterdam: Thesis Publishers. Jan M.M. van den Bos, (1991), Dutch EC Policy Making. A Model-Guided Approach to Coordination and Negotiation. Amsterdam: Thesis Publishers. Karin Sanders, (1991), Vrouwelijke Pioniers. Vrouwen en mannen met een 'mannelijke' hogere beroepsopleiding aan het begin van hun loopbaan. Amsterdam: Thesis Publishers. Sjerp de Vries, (1991), Egoism, Altruism, and Social Justice. Theory and Experiments on Cooperation in Social Dilemmas. Amsterdam: Thesis Publishers. Ronald S. Batenburg, (1991), Automatisering in bedrijf. Amsterdam: Thesis Publishers. Rudi Wielers, (1991), Selectie en allocatie op de arbeidsmarkt. Een uitwerking voor de informele en geïnstitutionaliseerde kinderopvang. Amsterdam: Thesis Publishers. Gert P. Westert, (1991), Verschillen in ziekenhuisgebruik. ICS dissertation, Groningen Hanneke Hermsen, (1992), Votes and Policy Preferences. Equilibria in Party Systems. Amsterdam: Thesis Publishers. Cora J.M. Maas, (1992), Probleemleerlingen in het basisonderwijs. Amsterdam: Thesis Publishers. Ed A.W. Boxman, (1992), Contacten en carrière. Een empirisch-theoretisch onderzoek naar de relatie tussen sociale netwerken en arbeidsmarktposities. Amsterdam: Thesis Publishers. Conny G.J. Taes, (1992), Kijken naar banen. Een onderzoek naar de inschatting van arbeidsmarktkansen bij schoolverlaters uit het middelbaar beroepsonderwijs. Amsterdam: Thesis Publishers.

15)

16) 17) 18)

19) 20) 21)

22)

23) 24)

25) 26) 27)

28)

29)

30)

31)

Peter van Roozendaal, (1992), Cabinets in Multi-Party Democracies. The Effect of Dominant and Central Parties on Cabinet Composition and Durability. Amsterdam: Thesis Publishers. Marcel van Dam, (1992), Regio zonder regie. Verschillen in en effectiviteit van gemeentelijk arbeidsmarktbeleid. Amsterdam: Thesis Publishers. Tanja van der Lippe, (1993), Arbeidsverdeling tussen mannen en vrouwen. Amsterdam: Thesis Publishers. Marc A. Jacobs, (1993), Software: Kopen of Kopiëren? Een sociaal-wetenschappelijk onderzoek onder PC-gebruikers. Amsterdam: Thesis Publishers. Peter van der Meer, (1993), Verdringing op de Nederlandse arbeidsmarkt. sector- en sekseverschillen. Amsterdam: Thesis Publishers. Gerbert Kraaykamp, (1993), Over lezen gesproken. Een studie naar sociale differentiatie in leesgedrag. Amsterdam: Thesis Publishers. Evelien Zeggelink, (1993), Strangers into Friends. The Evolution of Friendship Networks Using an Individual Oriented Modeling Approach. Amsterdam: Thesis Publishers. Jaco Berveling, (1994), Het stempel op de besluitvorming. Macht, invloed en besluitvorming op twee Amsterdamse beleidsterreinen. Amsterdam: Thesis Publishers. Wim Bernasco, (1994), Coupled Careers. The Effects of Spouse's Resources on Success at Work. Amsterdam: Thesis Publishers. Liset van Dijk, (1994), Choices in Child Care. The Distribution of Child Care Among Mothers, Fathers and Non-Parental Care Providers. Amsterdam: Thesis Publishers. Jos de Haan, (1994), Research Groups in Dutch Sociology. Amsterdam: Thesis Publishers. K. Boahene, (1995), Innovation Adoption as a Socio-Economic Process. The Case of the Ghanaian Cocoa Industry. Amsterdam: Thesis Publishers. Paul E.M. Ligthart, (1995), Solidarity in Economic Transactions. An Experimental Study of Framing Effects in Bargaining and Contracting. Amsterdam: Thesis Publishers. Roger Th. A.J. Leenders, (1995), Structure and Influence. Statistical Models for the Dynamics of Actor Attributes, Network Structure, and their Interdependence. Amsterdam: Thesis Publishers. Beate Völker, (1995), Should Auld Acquaintance Be Forgot...? Institutions of Communism, the Transition to Capitalism and Personal Networks: the Case of East Germany. Amsterdam: Thesis Publishers. A. Cancrinus-Matthijsse, (1995), Tussen hulpverlening en ondernemerschap. Beroepsuitoefening en taakopvattingen van openbare apothekers in een aantal West-Europese landen. Amsterdam: Thesis Publishers. Nardi Steverink, (1996), Zo lang mogelijk zelfstandig. Naar een verklaring van verschillen in oriëntatie ten aanzien van opname in een verzorgingstehuis onder fysiek kwetsbare ouderen. Amsterdam: Thesis Publishers.

ICS Dissertation Series 32) 33) 34)

35) 36)

37)

38)

39) 40) 41) 42)

43) 44)

45)

46)

47) 48)

Ellen Lindeman, (1996), Participatie in vrijwilligerswerk. Amsterdam: Thesis Publishers. Chris Snijders, (1996), Trust and Commitments. Amsterdam: Thesis Publishers. Koos Postma, (1996), Changing Prejudice in Hungary. A Study on the Collapse of State Socialism and Its Impact on Prejudice Against Gypsies and Jews. Amsterdam: Thesis Publishers. Jooske T. van Busschbach, (1996), Uit het oog, uit het hart? Stabiliteit en verandering in persoonlijke relaties. Amsterdam: Thesis Publishers. René Torenvlied, (1996), Besluiten in uitvoering. Theorieën over beleidsuitvoering modelmatig getoetst op sociale vernieuwing in drie gemeenten. Amsterdam: Thesis Publishers. Andreas Flache, (1996), The Double Edge of Networks. An Analysis of the Effect of Informal Networks on Cooperation in Social Dilemmas. Amsterdam: Thesis Publishers. Kees van Veen, (1997), Inside an Internal Labor Market: Formal Rules, Flexibility and Career Lines in a Dutch Manufacturing Company. Amsterdam: Thesis Publishers. Lucienne van Eijk, (1997), Activity and Well-being in the Elderly. Amsterdam: Thesis Publishers. Róbert Gál, (1997), Unreliability. Contract Discipline and Contract Governance under Economic Transition. Amsterdam: Thesis Publishers Anne-Geerte van de Goor, (1997), Effects of Regulation on Disability Duration. ICS dissertation, Utrecht Boris Blumberg, (1997), Das Management von Technologiekooperationen. Partnersuche und Verhandlungen mit dem Partner aus Empirisch-Theoretischer Perspektive. ICS dissertation, Utrecht. Marijke von Bergh, (1997), Loopbanen van oudere werknemers. Amsterdam: Thesis Publishers. Anna Petra Nieboer, (1997), Life-Events and Well-Being: A Prospective Study on Changes in Well-Being of Elderly People Due to a Serious Illness Event or Death of the Spouse. Amsterdam: Thesis Publishers. Jacques Niehof, (1997), Resources and Social Reproduction: The Effects of Cultural and Material Resources on Educational and Occupational Careers in Industrial Nations at the End of the Twentieth Century. ICS dissertation, Nijmegen. Ariana Need, (1997), The Kindred Vote. Individual and Family Effects of Social Class and Religion on Electoral Change in the Netherlands, 1956-1994. ICS dissertation, Nijmegen. Jim Allen, (1997), Sector Composition and the Effect of Education on Wages: an International Comparison. Amsterdam: Thesis Publishers. Jack B.F. Hutten, (1998), Workload and Provision of Care in General Practice. An Empirical Study of the Relation Between Workload of Dutch General Practitioners and the Content and Quality of their Care. ICS dissertation, Utrecht.

49)

50)

51)

52) 53) 54) 55) 56)

57) 58) 59) 60)

61)

62) 63) 64) 65) 66) 67) 68) 69)

Per B. Kropp, (1998), Berufserfolg im Transformationsprozeß. Eine theoretisch-empirische Studie über die Gewinner und Verlierer der Wende in Ostdeutschland. ICS dissertation, Utrecht. Maarten H.J. Wolbers, (1998), Diploma-inflatie en verdringing op de arbeidsmarkt. Een studie naar ontwikkelingen in de opbrengsten van diploma's in Nederland. ICS dissertation, Nijmegen. Wilma Smeenk, (1998), Opportunity and Marriage. The Impact of Individual Resources and Marriage Market Structure on First Marriage Timing and Partner Choice in the Netherlands. ICS dissertation, Nijmegen. Marinus Spreen, (1999), Sampling Personal Network Structures: Statistical Inference in Ego-Graphs. ICS dissertation, Groningen. Vincent Buskens, (1999), Social Networks and Trust. ICS dissertation, Utrecht. Susanne Rijken, (1999), Educational Expansion and Status Attainment. A CrossNational and Over-Time Comparison. ICS dissertation, Utrecht. Mérove Gijsberts, (1999), The Legitimation of Inequality in State-Socialist and Market Societies, 1987-1996. ICS dissertation, Utrecht. Gerhard G. Van de Bunt, (1999), Friends by Choice. An Actor-Oriented Statistical Network Model for Friendship Networks Through Time. ICS dissertation, Groningen. Robert Thomson, (1999), The Party Mandate: Election Pledges and Government Actions in the Netherlands, 1986-1998. Amsterdam: Thela Thesis. Corine Baarda, (1999), Politieke besluiten en boeren beslissingen. Het draagvlak van het mestbeleid tot 2000. ICS dissertation, Groningen. Rafael Wittek, (1999), Interdependence and Informal Control in Organizations. ICS dissertation, Groningen. Diane Payne, (1999), Policy Making in the European Union: an Analysis of the Impact of the Reform of the Structural Funds in Ireland. ICS dissertation, Groningen. René Veenstra, (1999), Leerlingen - Klassen - Scholen. Prestaties en vorderingen van leerlingen in het voortgezet onderwijs. Amsterdam, Thela Thesis. Marjolein Achterkamp, (1999), Influence Strategies in Collective Decision Making. A Comparison of Two Models. ICS dissertation, Groningen. Peter Mühlau, (2000), The Governance of the Employment Relation. A Relational Signaling Perspective. ICS dissertation, Groningen. Agnes Akkerman, (2000), Verdeelde vakbeweging en stakingen. Concurrentie om leden. ICS dissertation, Groningen. Sandra van Thiel, (2000), Quangocratization: Trends, Causes and Consequences. ICS dissertation, Utrecht. Rudi Turksema, (2000), Supply of Day Care. ICS dissertation, Utrecht. Sylvia E. Korupp (2000), Mothers and the Process of Social Stratification. ICS dissertation, Utrecht Bernard A. Nijstad (2000), How the Group Affects the Mind: Effects of Communication in Idea Generating Groups. ICS dissertation, Utrecht Inge F. de Wolf (2000), Opleidingsspecialisatie en arbeidsmarktsucces van

ICS Dissertation Series

70) 71) 72)

73) 74)

75) 76) 77) 78)

79) 80)

81)

82) 83) 84)

sociale wetenschappers. ICS dissertation, Utrecht Jan Kratzer (2001), Communication and Performance: An Empirical Study in Innovation Teams. ICS-dissertation, Groningen Madelon Kroneman (2001), Healthcare Systems and Hospital Bed Use. ICS/NIVEL-dissertation, Utrecht Herman van de Werfhorst (2001), Field of Study and Social Inequality. Four Types of Educational Resources in the Process of Stratification in the Netherlands. ICS-dissertation, Nijmegen Tamás Bartus (2001), Social Capital and Earnings Inequalities. The Role of Informal Job Search in Hungary. ICS-dissertation Groningen Hester Moerbeek (2001), Friends and Foes in the Occupational Career. The Influence of Sweet and Sour Social Capital on the Labour Market. ICSdissertation, Nijmegen Marcel van Assen (2001), Essays on Actor Perspectives in Exchange Networks and Social Dilemmas. ICS-dissertation, Groningen Inge Sieben (2001), Sibling Similarities and Social Stratification. The Impact of Family Background across Countries and Cohorts. ICS-dissertation, Nijmegen Alinda van Bruggen (2001), Individual Production of Social Well-Being. An Exploratory Study. ICS-dissertation, Groningen Marcel Coenders (2001), Nationalistic Attitudes and Ethnic Exclusionism in a Comparative Perspective: An Empirical Study of Attitudes Toward the Country and Ethnic Immigrants in 22 Countries. ICS-dissertation, Nijmegen Marcel Lubbers (2001), Exclusionistic Electorates. Extreme Right-Wing Voting in Western Europe, ICS-dissertation, Nijmegen Uwe Matzat (2001), Social Networks and Cooperation in Electronic Communities. A theoretical-empirical Analysis of Academic Communication and Internet Discussion Groups, ICS-dissertation, Groningen Jacques Janssen (2002), Do Opposites Attract Divorce? Dimensions of Mixed Marriage and the Risk of Divorce in the Netherlands, ICS-dissertation, Nijmegen Miranda Jansen (2002), Waardenoriënties en partnerrelaties. Een panelstudie naar wederzijdse invloeden, ICS-dissertation, Utrecht Anne Rigt Poortman (2002), Socioeconomic Causes and Consequences of Divorce. ICS-dissertation, Utrecht Alexander Gattig (2002), Intertemporal Decision Making, ICS-dissertation, Groningen