1
Inducing cognitive development and learning A Review of short-term training experiments I. The organismic development...
83 downloads
614 Views
10MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
1
Inducing cognitive development and learning A Review of short-term training experiments I. The organismic developmental approach’
:
SIDNEY STRAUSS Tel-Aviv University
A bstruct The author reviewed and interpreted the literature of experimental studies whose purposes were to determine rules of generation that transform a child’s cognitive organization at one stage of development into that of the next, more advanced stage. The categories of the research methodology were consistent with the organismic-developmental approach, and the findings of the studies tended to support most of the hypotheses generated from this approach. Some unresolved theoretical and methodological issues were presented, and research strategies to shed light on these issues were proposed.
The mechanisms or processes governing conceptual progress have long been a central area of concern in cognitive, developmental and educational psychology. Among other problem areas, this concern has recently become focalized (e.g. Tanner and Inhelder, 1956; Elkind and Flavell, 1969; Wallach, 1963; Kessen and Kuhlman, 1962; Sigel and Hooper, 1968; Flavell, 1963; Ripple and Rockcastle, 1964; Bruner, 1964, 1966; Flavell and Hill, 1969; Halford, 1970; Brainerd and Allen, 1971) on the question of how a child progresses through the stages of concept formation delineated by the research of Piaget and his co-workers (as summarized, for example, in Piaget, 1966; Inhelder and Piaget, 1958, 1964). The problem is to determine rules of generation that transform a child’s cognitive organization at one stage of development into that of the next, more advanced stage. A major empirical method of studying generative rules of transformation is ex1. The author would like to express his appreciation to Matal - the Tel Aviv University Elementary School Science Project - for its generous support during the preparation of
this manuszript. Special thanks are due to Jonas Langer and Paul Mussen for their helpful comments on an earlier version of this paper.
Cognitiorl I (4), pp. 329-317
330
Sidney
Strauss
perimental. The basic paradigm is to construct procedures that simulate what is thought to ordinarily occur in a child’s cognitive development and to examine their relative effects upon progress to a more advanced stage of conceptual activity. Numerous intervention studies in cognitive development have been performed in the nine years since Flavell’s (1963) analysis of the few limited studies available at the time of his review. In this paper we will first examine the research reported in the past nine years, and then we will formulate some conclusions that may be drawn from this experimental literature as to the theoretical understanding of the process of cognitive stage transformation that has been gained.
The organismic
stage hypothesis
Since the present review is relevant to the organismic stage hypothesis, we will begin by a brief summary of those features of the hypothesis that set the context for the research to be considered in this paper. The basic organismic assumption about development is that it is a two-fold process of organization and adaptation (Piaget, 1965. 1966). All organisms, however primitive. are born with organized structures, however rudimentarily they are formed. The organism’s organization is the basis for his initial interactions with his environment. Moreover, his innate organization determines the way in which he structures that initial interaction. All organisms arc also born with adaptive functions. Adaptation is the functional basis for the organism’s interaction with his environment. The source of the development of more advanced competence is the organism’s interactions with his environment. His actions upon the environment construct the forms, not the content, of his experience. These experiential products of his constructive activity feed back upon the organization of his structures which originally set the whole functional pattern into operation. When this information is discrepant with the existing functional structures, the resultant conditions of disequilibrium are propitious for progressive organizational and adaptive change in order to accommodate the anomalous forms (Piaget, 1967b. 1968; Langer, 1969a,b). Progressive change means structural and functional advance. Concomitantly, progressive structural transformation means advance in the organism’s forms of interacting with his environment and advance in the forms of conception that he constructs. This, in the most sketchy form, is the self-generative or self-constructive cycle of development and interaction that organismic theory posits to begin at birth and progresses until the individual develops to his most mature form. What is characteristic of development is the qualitative change of forms of reasoning; i.e., development or stage transformation is neither incremental nor additive. Consequently, we
Inducing cognitive development and learning
33 1
can define development as the construction of qualitatively different forms of reasoning which did not exist as part of the structural configuration prior to development. Within organismic-developmental theory, a sharp distinction is made between development and learning. Functionally, learning is the application of an intellectual structure to a wide variety of objects and events. That is, one learns rules about how to apply the structure, and, consequently, the structure becomes increasingly elaborated. The appearance of a learning order (i.e., an order of concept acquisition) suggests that objects vary in their structure and that, therefore, they vary in the resistance they offer to logical conceptualization. Structural decalage between objects explains experimental evidence of concept attainment decalage. The resistance which objects offer to rules of application is thought to be related to the complexity of the object’s parameters. It should be noted that the level of a child’s structural development determines the concepts he will learn. That is, the intellectual structure sets the limits for what can be learned. In this sense learning is subordinate to development. However, this is not to say that learning does not support development. As we shall see, the developmental status of more elaborated and less elaborated structures is quite different (e.g. Youniss, 1971). This paper is intended to review those studies which have attempted to induce structural transformations (development) and structural elaboration (learning) and to assess the consequences of their findings for a theory of cognitive development. The review will be limited to those studies which have attempted to induce these changes from the intuitive to the concrete operations stages. A review of this sort also has the possibility to confirm or disconfirm one of the major tenets of organismic developmental theory, namely, that stage development is sequential. Only a few studies (Turiel, 1966; Kuhn, 1969) have tested this hypothesis directly, and their findings confirmed the hypothesis. We shall consider this hypothesis indirectly and shall return to it in the discussion section. Finally, we should add that a number of methodological problems arose in this review. Each presented its unique issues which will be briefly discussed here. These problems fall into three categories: Stage assessment procedures and criteria, the assessment of transitional Ss, and the fit between some studies’ theoretical aims and research strategies. Let us begin with stage assessment procedures and criteria. The problem here is in the comparison of experiments which have used different assessment criteria. One solution would be to treat them as being equivalent. Gruen (1966) has shown that different criteria produce different interpretations of the results and therefore, different theoretical conclusions. For this reason, this solution is inadequate. We
have adopted
a second
solution.
The most stringent
criterion
has been accepted
as
the most powerful assessment tool. This criterion has been established by Piaget and his co-workers (cf. Inhelder and Sinclair, 3969). Here S is assessed to be at the concrete operations stage for conservation concepts if he: (1) Makes a correct judgment of equivalence, (2) logically justifies that judgment, (3) successfully resists a verbal countersuggestion, and (4) produces a successful performance on a related behavioral task. The second category of methodological issues concerns the assessment of transitional Ss. A transitional S is one who displays at least two functional structures. This becomes an issue because many recent studies have found that transitional Ss are more likely to change levels of reasoning than those who are not. The problem Fvithin this issue arises in an analysis and comparison of studies, some of which have not assessed transitional Ss. It is of greater theoretical importance to make statements about intuitive and transitional Ss. The third category of methodological issues is the problem of research strategy not fitting the intended purpose of the study. Unfortunately. there are a number of studies which fit this description. As a result, many of the to-be-reviewed studies’ procedures will be analyzed in a manner discrepant with the original analyses made by the studies’ authors. With this brief introduction, we shall begin the review of the literature. Three classes of training have been investigated experimentally which are either derived directly from or are theoretically consistent with the organismic stage hypothesis of cognitive development. The three kinds of training focus upon (1) disequi!ibrium. (2) mental operations. and (3) regression.
1. Disequilibrium Disequilibrium is the key concept which Piaget employs to describe and explain structural transformation and elaboration. Tt is a concept of interaction in which a psychological act is defined as the interaction between a structure and the environment and within the structure itself. In other words, each psychological act has two aspects: The external and internal. The external aspect faces outward to interact and function with the environment. This is what Piaget (1960, 1967b) calls adaptation. The internal aspect faces inward to structure itself. This is the organizational nature of structures. When a mental structure is equilibrated, it possesses a stability or balance. The equilibrated structure does not suggest that it is static but rather that its stability is relative to (1) external information acting upon it. adaptational equilibrium, and (2) lack of actual internal contradictions, organizational equilibrium.
Inducing cognitive development
and learning
333
In the adaptational interaction between environmental information and the intellectual structure, experiences are structurally assimilated. If the information is acted upon and the operative system can integrate it, the structure remains in equilibrium. Within the equilibrium model, the function of operations is to anticipate information discrepant with the existing structure in order to compensate for anomalous information. The state of organizational equilibrium can exist under two conditions. The first occurs if an individual is operating entirely at the highest level of formal operational reasoning. The second condition occurs when an individual is operating at any other level or combination of levels and is not ‘aware’ that he is producing contradictory judgments. This dual analysis of equilibrium implies a similar analysis of disequilibrium sources and mechanisms of progressive structural change. In both analyses of disequilibrium, we find the notion that a perturbing event(s) may be followed by a state of structural and functional disequilibrium which energizes structural reorganization (Langer, 1969b). Given these theoretical considerations, we can find two major categories of training experiments: Those which attempt to induce (a) external, adaptational disequilibrium by means of prediction-outcome conflict and (b) internal, organizational disequilibrium through structural mixture conflict. 1.1 Adaptational disequilibrium There have been two theoretical approaches within this model: The Genevan and Harvard approaches of Piaget and Bruner and their co-workers. We shall describe both of these approaches, review the literature relevant to them, and suggest some of the problems of their approaches. 1.1.1 Genevan model. The adaptational model suggests that a child’s actions should result in feedback which does not fit the intellectual structure which gave rise to these actions. This lack of fit could perturb the mental structure, thus inducing a state of disequilibrium. A methodological technique for inducing such a state could first have S predict the outcome of a transformation, such as predicting the level of water when it is poured from one shaped container to another shaped container. S could then be asked to observe the outcome of the transformation; i.e., the final level after the pouring of the water. The underlying hypothesis is that the outcome could conflict with the child’s predictions. For example, a child who is predominantly at the intuitive stage of reasoning might predict that the levels of equivalent amounts of water in two different shaped containers will be equal. The negative feedback to the structure could provide the conditions for restructuring and progress to a more advanced level of conceptualization (cf. Inhelder et al., 1966).
334
Sidney Strauss
Inhelder and Sinclair (1969) described a procedure whose purpose was to induce prediction-outcome (PO) conflict. The results of this experiment were that for Ss who were initially assessed at the intuitive level, only 12.5 % progressed to the transitional phase and none progressed to the concrete operations stage. Thus, progressive development was minimal for intuitive Ss. In marked contrast, 75 % of those Ss who were initially assessed at the transitional phase progressed. Within this group, 38.5 % extended their previous structure while the remaining 36.5 % progressed to concrete operational reasoning. In a related study, Smedslund (1963b) used a PO technique in an effort to induce cognitive disequilibrium and subsequent progressive cognitive reorganization. The procedure involved the anticipation and perception of displacement outcome using a Muller-Lyer illusion apparatus. The results were that 7 of 16 Ss trained in this condition conserved length on a posttest as compared to 1 of 16 control Ss. An objection to the study is that it is difficult to draw conclusion about the findings since operativity was assessed with a task from the figurative domain (the MullerLyer illusion). This is discussed in detail elsewhere (Langer and Strauss, 1972; Strauss, 1972b). 1.1.2 Harvard model. Bruner (1964, 1966, 1967) suggests that some children who appear not to conserve have actually attained the conservation concept in their verbal symbolic representation of nonperceived events. When these same children perceive a transformation, they may be misled by irrelevant perceptual cues, e.g., the height of the liquid in a container. Thus, a child can be a nonconserver in his ikonic representation and a conserver in his symbolic representation. Bruner further suggests that conceptual progress, in the form of conservation judgments in the face of perceived transformations, can occur if there is a mismatch between the ikonic and symbolic modes of representation. That is, if they are in conflict, the child might reorganize his perceptions. A condition which could provide the mismatch is one which would have three parts. First, the misleading perceptual cues of a transformation would be eliminated by techniques such as screening. Second, the child would give a verbal symbolic rendering of the nonperceived transformation. This would presumably be a conservation judgment. Third, the shielding device would be removed in order that the child could perceive the perceptually distorted transformation. Bruner suggests that the discrepancy between a symbolic conservation judgment and an ikonic nonconservation judgment should arouse sufficient cognitive conflict that perceptual reorganization would ensue. In experiments described in detail elsewhere (Strauss, 1972b), both Sonstroem (1966) and Frank (1966) from the Harvard group and Strauss and Langer (1970) employed screening techniques as PO training conditions. Basically, Strauss and
Inducing cognitive development
and learning
335
Langer (1970) replicated Frank’s PO procedure, included a second PO procedure, and compared these procedures to (a) two other training conditions which involved prediction without outcome (PE) and (b) control groups. The major finding for our interests here was that the PO conditions were not more effective in inducing cognitive reorganization than the PG groups and the controls. Although both the Genevan and Harvard groups use a PO model for cognitive conflict inducement, a deeper analysis dissolves the similarity. Schematically, the two theories can be characterized as in Figure 1. Figure 1. Prediction-outcome
conflict models of the Genevan and Harvard groups
Genevan group operational
structure
PREDICTION_
perception
imagery
of transformation. OUTCOME
Harvard group “structure”
______t
PREDICTION
t CONFLICT symbolic -t------_-t representation
imagery
ikonic t perception
representation
of transformation
OUTCOME
The Genevan group suggests that a child’s prediction will be determined by his existing operational structure. Anticipatory imagery is a by-product of this model. The prediction comes into conflict with the observation of the transformation if two conditions are fulfilled. First, the structural prediction and the outcome of the transformation must contradict each other. This is not assured since both Piaget (1966) and Bruner (1966) have shown that at the intuitive stage Ss can correctly predict the empirical outcome of a transformation such as predicting water levels in the example given previously. Second, the outcome product must inform the operatory structure in such a way that the information is anomalous with the structure (cf., Piaget’s 1970 notion of retroaction). This, too, is not assured due to the mechanism of coherence. That is, incoming retroactive information can be assimilated to (distorted by) existing
336
Sidney Strauss
structures in such a way that it tends to fit the structure (Kuhn and Langer. 1968; Turiel, 1969; Rest, Turiel and Kohlberg, 1969; Langer, 1969b). For further clarification and precision, this formulation must take two other factors into consideration. First, it must account for internal, self-generated disequilibrium (Langer, 1969b). That is, it is not enough that external events chosen as conflict-producing are presented to children. Although they are viewed as logically conflicting from the experimenter’s viewpoint, it is not altogether clear that children will find the same event perturbing. Second, it must make a structural description which can explain how the coherence mechanism allows partial fit to occur. This will be the subject of discussion in section 1.2. The Harvard group’s PO model presumes the existence of a mental structure despite the lack of a clear definition of its nature. Structures have the tools of physical action, imagery and language as media to represent reality. Thus, structural products could be verbal predictions of an event or images of nonperceived events. The conflict which Bruner describes is not between the structure and retroactive feedback as the Genevans would have it. Rather, the conflict is between products of the structure, e.g., the ikonic and symbolic modes. In effect, we have an interaction between modes of representation which are external to the structure. Nevertheless, the structure is affected by the conflict (indicated by the bi-directional arrow) since conflict resolution is theoretically followed by relatively consistent dominance of the symbolic mode. This dominance must be structurally recorded. If not recorded, we would not expect the symbolic mode (conservation) to dominate a child’s reasoning. Within the Harvard approach, there are two basic problems. The first is identical to the first problem raised by the Genevan approach. Namely, it was found that there are intuitive Ss who are able to anticipate correct levels due to previous empirical observations. One can then question how such children can be brought into conflict since there would be no discrepancy or mismatch between their predictions (symbolic rendering) and observations (ikonic rendering). The second basic problem in the Harvard PO model is a general one concerning the relations between the figurative and operative domains. As indicated earlier, the Harvard model suggests that interactions and subsequent conflict between aspects of the figurative domain (language, perception, imagery) will induce change in the operative domain (structure). Piaget (1966) and Piaget and Inhelder (1971, 1969) have provided empirical evidence for the issue of the role of the figurative domain in the inducement of structural reorganization. With respect to imagery, Inhelder (1965, p. 16) summarized their findings by stating that ‘the insufficiently co-ordinated centralizations and irreversibility of preoperative thought go together with an essentially static imag-
Inducing cognitive development
and learning
337
ery . . . since its role is to imitate and not to construct, operation
by its very constructiveness rapidly increases mobility and, in turn, acts on symbolic imagery.’ The problem this presents for the Harvard PO model is that training where two aspects of the figurative domain interact would seem even less likely to induce a reorganization of function-structures. 1.2 Organizational disequilibrium This category is often referred to as transitional reasoning. Unlike the adaptational disequilibrium model, it does not refer to a set of intervention conditions. Rather, it refers to a child’s structural organization. Piaget (1970) and Turiel (1959) have described structural organizations where a child possesses at least two functional structures of reasoning. Assessments of structural profiles indicate that a child applies one structure predominantly and applies others less (see Figure 2).
Figure 2. Profile of moral stage usage on Kohlberg
moral judgment
interview
(from Rest, Turiel and Kohlberg, 1969)
Percent
stage usage
Stage level relatwe
to dominant
stage
Two general types of structural mixture have been studied. The first type is within a concept. Empirically this means that when given a series of tasks on, for example, length conservation, a child will conserve on at least one but not all of the tasks. This suggests that this child has constructed and operates with both the concrete
338
Sidney Strauss
operations
and intuitive
structures.
The second type of structural
mixture
is between
concepts. An example of this type could be a child who applies the concrete operational structure for discontinuous quantity conservation and applies the intuitive structure to the solution of length conservation problems. In both types of structural mixture, children have been more likely to construct the concrete operational structure for length conservation than those children who were strictly intuitive operational, i.e., displayed no measured structural mixture (Strauss and Langer, 1970; Langer and Strauss, 1972). In the latter study, it was found that 19 of 23 Ss who were at the intuitive stage for length conservation on the pretest changed to the concrete operations stage on the posttest if they were also transitional within or concrete operational for discontinuous quantity on the pretest. In contrast, 15 of 17 Ss who were assessed to be at the intuitive stage for both length and discontinuous quantity on the pretest did not change their conceptual level for length conservation on the posttest. Other studies (Coxford, 1964; Beilin, 196.5; Turiel, 1969; Inhelder and Sinclair, 1969) also indicated that children who displayed structural mixture were more likely than those who displayed no measured structural mixture to progressively transform their cognitive structures. One can interpret this to mean that structural mixture may be an inherently unstable state that is particularly susceptible to change. The instability might arise from situations in which a child with structural mixture may be producing contradictory judgments. That is, in one situation a child might apply the concrete operations structure, while in a related situation he might apply the intuitive structure. The situation in which a child successively produces contradictory judgments has been hypothesized to be one which could produce internal, organizational perturbations and subsequent structural and functional disequilibrium (Langer, 1969b). The findings reported above were made on a post-hoc basis. There have also been studies whose purpose was to exploit the natural state of structural mixture in order to induce organizational disequilibrium. One such study was conducted by Inhelder and Sinclair (1969). They attempted to determine whether the acquisition of length conservation (for which S was intuitive) could be facilitated by applying numerical operations (for which S was operational) to the evaluation of length. The posttest results were that 35 % of the Ss made no progress at all. Of the remaining 65 % slightly more than one-fourth were assessed to have applied the concrete operations structure to both number and length conservation problems. A second type of research strategy to induce organizational disequilibrium by exploiting structural mixture involves exposing a child to a model who solves problems with the use of the structure which is one above (+ 1) the child’s predominant structure (0). Theoretically, if a child perceives the discrepancy between
Inducing cognitive development
and learning
339
his own predominant structure and the modelled structure, one might expect a state of disequilibrium and subsequent structural elaboration. It should be noted that the model’s + 1 structure is already present in the child’s structural configuration and is the next structure which will become predominant. Thus, we have the potential of a more advanced but less used structure informing a structure which is less advanced but more used. An attendent hypothesis is that if a model presents Ss with reasoning which is two structures above (+ 2) his predominant level (0), he will be less likely to change to the +2 structure because the +2 structure either does not exist or hardly exists in the child’s structural configuration. Here one structure cannot inform another. Also, this type of change would contradict the organismic hypothesis that it is not possible to skip stages of reasoning. Kuhn (1969; 1972), basing her work on Turiel’s (1966) study which pioneered this research strategy, had models present Ss with + 1 or + 2 classification reasoning. Of 81 experimental Ss, 19 % showed clear progressive structural change. Of those who changed to the + 1 structure, 67 % received + 1 structural reasoning while the remaining 33 % received +2. In using a measure of partial change, it was found that almost 69 % of the + 1 and +2 Ss changed partially. It was also found that all Ss who changed after receiving the + 1 structure of reasoning changed to that structure on the posttest. In contrast, among those Ss who received the +2 condition and changed, only 36 % changed to +2 reasoning structure while the remaining Ss changed to + 1 reasoning. Other research conducted in the intellectual and moral domains has provided us with similar findings (Murray, 1972; Turiel, 1966, 1969; Turiel and Rothman, 1972; Blatt and Kohlberg, 1972). In sum, it becomes clear that despite the centrality of equilibration in Piaget’s theory of structural development, few studies have been conducted with the purpose of or which meet the criteria for inducing organizational disequilibrium. The major evidence we have that training studies of this type might be successful has been accumulated in a post-hoc manner. This area seems to have considerable potential for generating research with other than post-hoc analyses which can produce data which are relevant to organismic stage hypotheses. For adaptational and organizational disequilibrium, a common developmental problem is found in the coherence mechanism. It was pointed out earlier that coherence is the phenomenon in which environmental information is distorted by the structure in order to allow that information to become structurally integrated. The function of this mechanism is to maintain structural integrity (Werner and Kaplan, 1963). It was suggested that the question raised in section 1.1, which addresses itself to the determination of the types of structures which allow partial fit, must have a
340
Sidney Strauss
structural
solution.
The
data
from
organizational
disequilibrium
studies
suggest
that in both the intellectual and moral domains the percentage of Ss who changed to + 1 reasoning is quite low. One interpretation of these findings is that -!- 1 information was presented to Ss who possessed varying amounts of structural mixture. It would be consistent with the organismic stage hypothesis that the presentation of + 1 information to a child who can apply the + 1 structure to only a small number of phenomena in the environment is not very likely to result in the assimilation of that information to the + 1 structure. Conversely, the presentation of + 1 information to a child whose + 1 structure is more elaborated is more likely to result in assimilation to the + 1 structure. One of the tasks of future research in this area should be to measure and accurately determine the above. Finally, a philosophical problem common to the adaptational and organizational models of equilibration is the tautological nature of our explanations of disequilibrium and change. If we find conceptual change, we conclude that there was structural disequilibrium. On the other hand, we define disequilibrium as a source of change. In defining elements of the system by each other, we cannot come very close to satisfactory explanations of cognitive advance. This suggests a need for a measure of disequilibrium which is independent of structural change. This motivation should serve as a guide for future investigations.
2.
Mental operations
There are two basic strategies in this type of training. The first (1) is to induce mental operations that are characteristic of one stage more advanced than the child’s current level of competence. To date, all the research on this idea has been limited to training children at the intuitive stage to progress to the concrete operations stage. Consequently, the procedure has attempted to train intuitive Ss for one of the logical operations that normally develops at the concrete operational stage (either the operations of addition, subtraction, or reversibility). A basis for the second research strategy (2) is the hypothesis that underlying the intuitive stage is a structure whose operational components are segregated or even syncretically fused. The research strategy here is to induce operational integration since an integrated structure is thought to underlie concrete operational reasoning. 2.1.1 Addition/subtraction (AS). The role of the AS operations in concrete operational conception, such as conservation, has been a subject of considerable theorizing and empirical research (Piaget, 1965). Empirical and anectodal information (Smedslund, 1961e, 1961f; 1962; 1963b) suggests that the AS operationis necessary, but ap-
Inducing cognitive development and learning
341
parently not sufficient, for the concrete operational conception of conservation. That is, a child may produce judgments that indicate the use of AS operations without conserving; yet conservation is always accompanied by the AS operations. Accordingly, organismic stage theory would not predict that training of this nature would induce conceptual advance. This point and the potential value of this type of research strategy will be discussed at the end of this section. In one of the first AS experiments, Smedslund (1961f) attempted to measure the effects that this type of training for continuous quantity conservation would have upon conservation of discontinuous quantity and vice versa. The results of significance for this review were that 30 of the 44 Ss partially progressed in their ability to conserve; however, the number of Ss who definitely conserved was small. Other experiments conducted by Smedslund (1961b, c), Gruen (1965) Smith (1968) Wallach and Sprott (1964) Wallach et al. (1967) Feigenbaum and Sulkin (1964) and Wohlwill and Lowe (1962) which attempted to induce the use of AS mental operation have found similar results. Relatively few Ss changed to concrete operational reasoning after having received training of this nature. These results are not totally compelling because (a) the AS training was always confounded with other conditions within the training; (b) none of the experimenters, with the exception of Smith (1968) and Smedslund (1961f). accounted for partial advances in the level of the child’s conceptual activity; or (c) the procedures were generally not described in enough detail. In sum, the effect of AS training upon conceptual progress to concrete conception appears to be partial or negligible. A more precise operational determination awaits (a) careful assessment of the child’s starting position; (b) AS training conditions that are not confounded, i.e., where other operations are not present; (c) insuring that the AS training actually leads the child to operations of addition and/or subtraction when attempting to deal with the conceptual test items; and (d) careful assessment of the child’s final position. 2.1.2 Reversibility. Reversibility enjoys a central position in Piagetian theory yet it is fraught with confusion and a certain amount of vagueness. In contemporary studies, there has been an obfuscation of the differences between renversibilitk and rkversibilitk (Piaget, 1967b). The former suggests the empirical return of a deformed object to its original configuration. For example, if we displace an object from position X to position Y, we can empirically cancel the effect of this displacement by returning the object from Y back to X. Rhersibilitt, on the other hand, has been interpreted in two ways. The first interpretation is that it is a mental operation which allows one to mentally cancel the effect of a transformation. In the case of the grouping structure which underlies
342
Sidney
Strauss
concrete operational reasoning, Piaget (1960, p, 29) differentiates between two types of reversibility operations: ‘From the standpoint of mental development, inversion (negation or elimination) and reciprocity (symmetry) form two kinds of reversibility, whose beginnings are already to be seen at the lower developmental levels.’ The second interpretation of reversibility is that of a reversible structure. This refers to the internal coordination of the system of operations which allows one to mentally transform the environment. For example, Piaget (1966, p. 11) ‘. . . defines intelligence in terms of the progressive reversibility of the mobile structures’. With these definitions, it is possible to have a reversible structure which contains two reversible operations. In general, research in this area has aimed towards conservation concept acquisition. The line of investigation has generally been to present an intuitive stage S renversibilite’ (R) operations; i.e., evidence is presented that shows S that one can return to the original configuration without any relevant changes having occurred to the transformed object. However, it has been shown that intuitive Ss can produce renversibilite judgments (Bruner et al. 1966; Wallach and Sprott, 1964; Piaget, 1968; Murray and Johnson, 1969). As in the case of AS operations, the above suggests that R is necessary but not sufficient for concrete operational reasoning. Therefore, one should not expect R training to induce either learning or development. In an experiment that is strikingly similar to the present interpretation of Smedslurid’s (1961e) work (see section 2.2) Wallach and Sprott (1964) attempted to induce number conservation through training on R transformations. On the posttest, 14 out of 15 training Ss and none of 15 controls produced conservation judgments. It appears that the training involved renversibilit&; however, it should be noted that one-half of the training procedures also contained either addition or subtraction. It is possible that since the training involved two operations necessary for conservation, cognitive change resulted. This interpretation will be discussed in section 2.2. In a follow-up experiment, Wallach et al. (1967) sought to determine whether: (a) Conservation could be induced by training either for reversibility or AS, and (b) training for and attainment of number conservation would transfer to conservation of continuous quantity. In order to avoid combining reversibility and additionsubtraction in the AS training, a screen was used to prevent Ss from seeing the transformed objects being restored to their original state. It was found that R training on number facilitated number conservation acquisition while AS did not. It was also found that neither R nor AS training on continuous quantity (liquid) was effective in inducing posttest conservation for the same concept. In their interpretation, the authors suggest that because R training was successful in conservation of number but not in continuous quantity, R by itself was not sufficient
Inducing cognitive development
and learning
343
to produce conservation. Nevertheless, the authors maintained that R is necessary for conservation, but it is not as important as the combination of the R operation and the removal (by use of the screen) of the distorting effect due to inaccurate perceptual cues. Unfortunately, these findings are open to question because of methodological problems. The most salient among them is that it is doubtful that reduction of misleading cues by screening leads to judgments of conservation (Sonstroem, 1966; Strauss and Langer, 1970; Wohlwill and Lowe, 1962; Smedslund, 1961d). A study which also reported the success of R training was conducted by Goldschmid (1968). Significantly more Ss successfully acquired conservation after R training than Ss who were administered training for the compensation operation and a combination of R and compensation training. Because the diagnostic techniques and training procedures were not described in sufficient detail, we are unable to interpret these results. Briefly, we can see that research related to this mental operation has not been productive in inducing progressive structural reorganization. We can explain this, in part, by the confusion surrounding the concept of reversibility. When considered a mental operation, we find training which attempts to indicate to intuitive Ss that an equal but opposite action returns a deformed object back to its original configuration (renversibilitk). However, as noted earlier, intuitive Ss can produce this judgment. On the other hand, we can consider the reversibility of a structure which is operationally integrated. Operational integration, which implies that each operation relies on the others, is not a likely product of training which attempts to induce or strengthen R. We shall now review the literature which can be interpreted to have facilitated operational integration. 2.2 Operational coordination As we have just shown, underlying intuitive reasoning are articulated, yet segregated, mental operations which constitute logical semi-groups; intuitive children can perform AS, R, or compensation operations and not conserve (for compensation operations, see Larsen and Flavell, 1970; Gelman and Weinberg, 1972). One of the basic premises of the organismic-developmental approach is that underlying operational stages are mental structures whose organization comprises a ‘structured whole’ (structure d’ensemble). These structures are composed of mental operations which, Piaget has argued, are internalized actions. The structure is coordinated in such a way that the operational system can now compensate for changes in the empirical world or in the operational system itself. Given the differences between the intuitive and concrete operations structures
344
Sidney Strauss
sketched above, a theoretically plausible hypothesis is that experimental procedures which attempt to integrate segregated mental operations would have a great likelihood of inducing cognitive advance. Few studies were designed to test this hypothesis, yet many can be interpreted to have fulfilled the requirements for such a test. Let us now consider evidence from this type of training strategy. In studying conservation of weight acquisition, Smedslund (1961e) attempted to induce cognitive conflict in children by performing AS transformations (e.g. the addition and removal of material from one ball of plasticene) and deformations (e.g. flattening out the other ball of plasticene). He found that during training, 5 of 13 Ss consistently used the AS operations while the remaining 8 Ss were continually dominated by the action of the deformation. Four of five AS Ss gave several conservation judgments on the posttest. Smedslund (1961e) reported that there was little, if any, conflict. A closer inspection of the training procedure suggests that the deformation technique could well be described as a renversible transformation. In some instances the child would not only be presented with the flattening out of the ball into a pancake but also with the pancake then being rolled back into a ball. Here we have evidence that when initially intuitive Ss receive training which includes more than one operation (in this case, renversibility and AS), the likelihood of conceptual progress is relatively high. An attempt to replicate Smedslund’s (1961e) study was performed by Winer (1968). The first result of significance confirmed Smedslund’s report of no conflict. Also, 43 % of the AS subjects conserved while none of the deformation and control Ss became concrete operational. Winer’s training procedures lend themselves to the interpretation that they involved an AS operation on all transformations and a renversible operation on onethird of the trials. On the other hand, the deformation set training procedure involved no AS operations but had a renversible operation on one-third of the trials. Thus, it appears that our interpretation of Smedslund’s findings have been supported. We must add a note of caution here since a weak criterion was used in Winer’s study as opposed to a stronger criterion used by Smedslund. In another attempt to replicate Smedslund’s findings, Smith (1968) employed the same procedure with intuitive and transitional Ss. For both groups of Ss there were no significant differences between this training procedure and a control. Thus, we have two studies which are related to our interpretation of Smedslund’s (1961e) study - whose findings contradict each other and whose assessment procedure are questionable. In a related experiment, Beilin (1965) used a verbal rule instruction training condition which contained a demonstration of a renverse operation and a verbal explanation of AS. It was found that this condition was significantly superior to others
Inducing cognitive development
and learning
345
in inducing change and operational thought. As in the Smedslund study, there were attempts to replicate (Mermelstein et al., 1967; Mermelstein and Meyer, 1967; Smith, 1968) and an inability to draw conclusions about Beilin’s study since the replication studies did not truly replicate Beilin’s conditions, and the findings contradicted each other. The same phenomenon was found in an original study by Sigel et al. (1966) and failure to replicate their multiple operations training by Mermelstein et al. (1967) and Mermelstein and Meyer (1967). Other studies can also be interpreted to have trained for the use of more than one mental operation (Gruen, 1965; Wallach and Sprott, 1964; Goldschmid, 1968; Inhelder and Sinclair, 1969; Bearison, 1970; Rothenberg and Orost, 1969). They, too, have reported success in inducing operational thought. It appears, then, that this type of training procedure has been relatively successful. This is consistent with the organismic-developmental hypothesis which suggests that structural and functional integration of operations is a prerequisite to concrete operational reasoning. In sum, there were two types of training strategies employed in research pertaining to mental operations. The first type attempted to train for the acquisition of isolated mental operations (addition, subtraction, or reversibility). The second type attempted to train for operational integration. The results indicated that the former strategy was quite unsuccessful while the latter was markedly successful in inducing structural transformation and elaboration. In fact, training for operational integration is among the most successful of all the methods reviewed in this paper. These findings are consistent with organismic-developmental theory. A research strategy which attempted to train for an isolated mental operation would not be thought to successfully induce either structural transformation or elaboration since the effect of such training might be either to segregate that operation or to exercise it (Piaget, 1970). Nevertheless, this type of training has potential for obtaining data which would be of considerable developmental interest. For example, it would be of interest to determine if this type of training would have the effect of differentiating mental operations from structures in which these operations were fused syncretically. Jt would also be of interest to determine the effects of this type of training on already differentiated operations. For example, would this type of training alter the developmental status of these operations in such a way that it would increase the probability of their integration? To date, no research of this sort has been conducted, yet it would yield important information about development and learning. This brings us to the second type of training strategy. As indicated previously, operational integration of functional structures is one of the criteria of the concrete operations stage. A strategy whose purpose is to integrate functionally segregated operations would, therefore, be thought to induce structural transformation and
346
Sidney Strauss
elaboration.
The training
strategy’s
success confirms
this developmental
hypothesis.
It should be noted that the strategy does not explain how S internalizes that information and coordinates his own mental operations. Structural transformation to the concrete operations stage can be facilitated through this general training method since the intuitive structure is defined by its operational segregation and the concrete operations structure by its operational integration. Structural ezaboration is somewhat more difficult to explain because presumably one has already constructed an integrated structure. It is theoretically plausible that when a structure interacts with objects that are more complex structurally and for which there are no rules of application, the tendency is to interact in an operationally segregated manner. That is, objects which are more complex structurally might have the tendency to re-segregate the operations at higher levels. One can conceive of this process occurring at all conceptual levels (decalage). This does not negate the organismic hypothesis that structural application is recursively isomorphic. It merely suggests that an analogous resegregation-integration process occurs cyclically at increasingly higher levels.
3.
Regression
This phenomenon has, more often than not, been viewed as a by-product of training. Few studies (Kingsley and Hall, 1967; Hall and Simpson, 1968) were designed with the explicit purpose of inducing regression through training. Nevertheless, in order to be more complete, a theory of cognitive development must account for regressive as well as progressive development. As Turiel(1969) has suggested, part of the problem of analysing regression centers upon the relationship between content and the structural form underlying it. That is, similar judgments might be the products of different structures. In part, the problem is definitional since there are two types of regressions: (a) Reversion to a subordinated level, and (b) regressive cognitive reorganization. For the first type, one should recall the law of hierarchic integration (Werner, 1948) which suggests that less developed systems become subordinated to and regulated by more developed systems. In addition, structural mixture comes into play here. In view of both of these phenomena, a stage can be characterized as a dominant or preferred mode of action. Conceptual change, whether progressive or regressive, can be viewed as a shift in preferred modes. A condition which might evoke reversion to a subordinated level might be the inability to apply the regulating system, e.g., due to the stress of a situation or the novelty or difficulty of a task (Werner, 1948). This regression is relatively shortterm. That is, the more developed system will predominate again once the stress
Inducing cognitive development
and learning
347
has been removed, the task is no longer novel, or the parameters of the more difficult task are understood. This type of regression is also local since all behavior does not become regulated by the less developed system. In other words, cognitive reorganization does not ensue from the above mentioned conditions. The second type of regression is that which involves cognitive reorganization. Werner (1948) and Langer (1969a, 1970) suggest that regressive development is characterized by changes in the mental structure which proceed from a differentiated organization to one which is more syncretic. The classic exemplars of regressive cognitive reorganization are found in cases of brain damage and psychopathic disruption (Goldstein, 1939). This type of regression concerns almost all behavior; i.e., regression is not local. It is also relatively long-term. Before we begin the review of the literature, it should be noted that an excellent review of extinction training was recently written by Miller (1971). One of the points which he made was that extinction of conservation is ubiquitous, rather easy to obtain, and a majority of Ss seem susceptible to such training. The reader is referred to Miller’s article for a detailed account of these findings since they will be treated in less detail here. In contrast to Miller’s interpretation, we shall begin by presenting evidence that indicates that neither type of regression occurs often under normal or stress conditions. It is important to note that those who have conducted studies which suggest this have been among the most careful in their assessments of structural levels. The differences between the two types of findings are not simply a matter of amount of regression. Rather, the differences lie in the models used to explain the phenomena of extinction and regression. Later, we shall attempt to explain the findings of one of the models by the other. The first model has a clear theoretical viewpoint which is based upon Werner’s (1948) concept of hierarchic integration. We stated earlier that, among other things, this means that less developed systems become regulated by more developed systems. In effect this means that the lower systems are not lost. Instead, they are thought to be transformed into the higher, regulating system. If we apply this to stage theory. we find that structural transformation involves a. reorganization to higher, more adaptive structures. If each structure is transformed into a qualitatively different form, we would expect that a child would have difficulty retrieving previous forms. In other words, this model would not predict the occurrence of regression unless conditions would be obtained in which the higher system would lose its regulatory function. In addition, these theoretical considerations relate to the issue of sequentiality alluded to in the introduction. Despite the prediction that one would not expect regression, one must account for its occurrence in a theoretically consistent manner. In section 1.2, we indicated
348
Sidney Strauss
that a child who has more structural mixture is more likely to oscillate between use of different structures than a child who has less structural mixture. It was argued that a state of structural mixture implied a state of natural potential disequilibrium and that one would expect to find more shifting of preferred modes of action at this time. Although cognitive change is generally progressive in the sense that the new organization is more adaptive, regression can also ensue if the child’s actions or the environmental information increases the likelihood of a more syncretic, less differentiated mental organization. Empirical evidence supporting this contention is found in the results of a study conducted by Strauss and Langer (1970) who trained for operational thought inducement. It was found that of 105 Ss, 6 % regressed. Five of the seven who regressed were transitional. Others reporting more regression among transitional than strictly intuitive Ss were Beilin (1965) Schwartz and Langer (1968) Turiel (1966) Kohlberg and Kramer (1969) and Turiel and Rothman (1972). In all of the cited studies, the percentage of Ss regressing was exceedingly small. Let us now look at some of these studies in more detail. Turiel (1969) presented data which suggest that regression in moral reasoning occurs rarely. For example, Turiel (1966) working with Kohlberg’s (1963) moral development model, conducted an experiment where children were put into a hypothetical moral dilemma and were provided ‘advice’ which was characteristic of stages one above (+ 1) and one below (-1) S’s predominant stage. Of particular interest for our discussion was the finding that the -1 condition had a small effect. Turiel (1969, 1971, 1972) interpreted these and other findings as an indication that although the stage below could be understood, the children could not accept it because they had previously rejected that mode of thought. Rest (1969) and Rest et al. (1969) also presented data which support this interpretation. It was found that children exhibited a high comprehension of -1 reasoning yet they rejected it as inadequate. In addition, it was found that children judged statements above their dominant stage as being ‘better’ than statements below their stage. In a similar study, Kuhn (1969, 1972) presented Ss -1 structures of classification. Of 19 Ss who were administered this condition, only one S produced the -1 structure of reasoning consistently on the posttests. Although the differences were not significant, Ss who saw models present reasoning at Ss’ dominant (0) level were more likely to change progressively than Ss in the -1 condition were likely to change regressively. These findings support the organismic-developmental model’s hypothesis that regression to a formerly predominant structure is unlikely. However, as indicated earlier, there are many studies whose findings do not support this hypothesis. As Miller (1971) pointed out so capably, these studies are quite divergent and do not present us with a clearly formulated alternative model. Indeed, we can find pro-
Inducing cognitive development
and learning
349
of both the organismic (Smedslund) and neo-behaviorist (Hall and his associates) positions whose findings were similar. Thus, the defining criterion of the second ‘model’ is not a theoretical position. Instead, it is defined by similar results from empirical studies whose purposes were to study the feasibility of extinguishing conservation responses. The prototype experiment was conducted by Smedslund (1961~). In this study, he tested Ss who had acquired conservation of weight by means of empirical demonstration (E) versus Ss who had ‘naturally’ acquired conservation of weight (N) on their ability to resist extinction, i.e., resist regressing to intuitive reasoning. In the extinction procedure, E surreptitiously removed clay from one of two objects which had been previously judged equivalent in weight. The results indicated that none of the E condition Ss resisted extinction while 6 of 13 Ss in the N condition successfully resisted extinction. Smedslund concluded that the organismic-developmental viewpoint had been substantiated since 46 % of the natural conservers had resisted the extinction experience. One must certainly question why the remaining 54 % did not resist extinction. We shall interpret these findings in the general discussion of regression. Subsequent related research has also produced similar findings. For example, Hall and Kingsley (1968) conducted an experiment which attempted to replicate Smedslund’s findings. Although not a true replication, they found that 0 of 17 Ss resisted extinction. Kingsley and Hall (1967) were also not successful in an attempt to replicate Smedslund’s findings. They found that none of 15 natural conservers resisted extinction while 3 of 17 Ss, who had apparently become operational through training (different from Smedslund’s), resisted extinction. However, the assessment of conservation acquisition was not very stringent and, therefore, questionable. Similar findings were also reported by Smith (1968) Brison (1966) Hall and Simpson (1968), and Halpern (1965). Smedslund (1968) indicated that the failure of Kingsley and Hall (1967) and Hall and Simpson (1968) to replicate Smedshmd’s (1961~) findings was due to their inadequate diagnosis of the presence or absence of conservation. Because Smedslurid’s procedure was not standardized and enabled the experimenter to probe more into the children’s reasoning, we believe Smedslund’s (1961~) findings are more reliable. In an attempt to account for all of the regression research results in a single theoretical framework, we shall reinterpret the studies of the second model within organismic-developmental stage theory. That is, in the second model’s studies, natural conservers were presumably applying the concrete operations structure of reasoning to the solution of the tasks presented to them. In accordance with Turiel’s (1966) convention, let us call this the 0 level. These Ss were shown an empirical ponents
350
Sidney Strauss
event which violated conservation. Here a child observed a phenomenon for which he had previously argued when his former (intuitive) structure was predominant. That is, a child who is applying the intuitive structure produces nonconservation judgments. Thus, the presentation of a nonconservation empirical event would be equivalent to Turiel’s -1 condition. As indicated earlier Turiel(1966), Turiel and Rothman (1972), Rest et al. (1969) and Kuhn (1969) found that the vast majority of Ss rejected -1 reasoning. Yet it was also indicated that the results of the second model’s studies showed considerable acceptance of -1 reasoning. There are several ways of interpreting this apparent discrepancy. First, in the reinterpreted studies, one can question whether the natural conserver applied the concrete operations structure predominantly. If not, this could be a source of discrepancy since Turiel’s definition of predominant stage usage (0) was use of this stage at least 50 % more than use of other stages (Turiel and Rothman, 1972). Thus, the extent of structural elaboration is an important variable to be considered here. For example, it would be of developmental interest to determine whether a structure which is less elaborated (e.g. is applied only to number conservation) will be more likely to reject -1 information regarding number than a structure which is more elaborated (e.g. is applied to number, length and continuous quantity problems). The converse has already been tested, but attempts at replication should be made because of the questions raised about assessment techniques. Here we are speaking of determining whether a more elaborated structure is more, or less, likely to reject -1 information from structurally more complex concepts such as weight. If the findings of Smedslund (1961b), Hall and Kingsley (1968), Smith (1968) and Kingsley and Hall (1967) can be replicated, we would then have data which suggest that a more elaborated structure is not very likely to reject this type of -1 information. This conclusion must be qualified not only by the lack of careful assessment techniques but also by the additional consideration of the amount of consolidation of a particular structure. For example, one would imagine that a child who recently learned rules of application to weight problems is less likely to reject -1 information than a child who has learned the same rules of application and who has consolidated them in using them for a considerable period of time. A second explanation for the discrepancy between findings might be due to the difference in the medium of presentation of -1 information. In the case of moral reasoning. Turiel (1966) Turiel and Rothman (1972) and Rest et al. (1969) presented the -1 structure in argument form. Kuhn (1969, 1972) presented arguments and classifications of materials which embodied a structure of reasoning which was -1 to Ss’ predominant structure. These forms of presentation are quite different from each other and from those of a -1 empirical outcome found in conservation
Inducing cognitive development
and learning
3.51
extinction studies. One would imagine that the violation of an expected empirical law is of a different order of complexity than an argument or a classification of material which one had previously used predominantly. One might, therefore, expect somewhat different results, depending upon the form and content which would be presented to the child. In sum, it appears that, contrary to Miller’s (1971) interpretation, ‘extinction’ or regression is not ubiquitous. There are two categories of data which have produced apparently contradictory results. The first category suggests that the rejection of a -1 structure of reasoning or a -1 empirical event is characteristic of Ss at all levels of reasoning. The second category allows for the assimilation of -1 reasoning to the existing intellectual structure. It was argued that some of the apparent discrepancy between these categories might be accounted for by methodological weaknesses in the latter category. Future research in this area should concentrate on obtaining a comprehensive profile of Ss’ pretest stage usage so that one can assess which structural configurations are likely to reject -1 information and which are not. In addition, it would be of theoretical interest to determine the influence of the (1) medium of -1 reasoning presentation (verbal, imaginal, etc.), and (2) content of the -1 information (moral reasoning concepts, physical concepts, logical concepts, etc.) on the acceptance or rejection of -1 reasoning.
Discussion Three classes of training studies, which are consistent with organismic-developmental stage hypotheses, have been reviewed. The first two, disequilibrium and mental operations, involved attempts to assess the effects of these types of training upon cognitive advance. The third, regression, has generally been viewed as a by-product of training studies. We shall discuss these three classes in terms of: (a) Their potential for inducing structural transformation and elaboration, and (b) the implications of their findings for sequentiality. The disequilibrium models present different answers to the problem of inducing structural transformation and elaboration. The adaptational (prediction-outcome) model suggests that both could ensue if proper training techniques would be employed. Structural transformation could be induced through retroaction if the child’s structure would be sufficiently elaborated and consolidated. Here mental actions would feed back to a structure which is susceptible to transforming itself. Mental actions on empirical feedback in the form of verification procedures could be one source of disequilibrium. The role of the figurative domain for structural transformations is discussed elsewhere (Strauss, 1972a, 1972b).
352
Sidney Strauss
A somewhat different argument can be made for structural elaboration. It is one of the organismic hypotheses that structural elaboration results when one gains an understanding of an object’s parameters. If one achieves this and then applies the highest level structure to the object, conceptual products such as conservation are produced. Understanding an object’s parameters does not come only from information about what they are but also about what they are not. The latter is no less important than the former. Thus, retroaction can serve structural elaboration. Empirical investigations which have studied the above hypotheses have begun only recently, and it is much too early to draw conclusions about their efficacy. Nevertheless, this seems to be a fertile area for research. The organizational disequilibrium model of structural mixture can predict only structural elaboration. The very definition of structural transformation as the generation of a structure never before constructed negates its inclusion as an outcome of this model. The little data which we have about this model suggests that it is among the most promising for inducing cognitive elaboration. Mental operations training also suggests different structural outcomes depending upon the training model. For training which exercises only one mental operation, we would predict neither structural transformation nor elaboration. Empirical outcomes match this prediction. Operational development and learning might ensue from this type of training and, in effect, might change the developmental status of the structure. Thus, this type of change is hypothesized to effect structural transformation or elaboration only indirectly. This hypothesis requires empirical verification since it has not yet been tested. In contrast to the training model in which only one operation is exercised, the operational integration model, in which several operations are exercised, would predict that both structural transformation and elaboration would ensue. Structural transformation could be induced by this training since the use of several operations might facilitate operational integration. Integrated structures are more advanced than structures which are syncretically fused or whose operations are differentiated but segregated. There is no experimental evidence for this since no study has been addressed to this problem. On the other hand, there is an abundance of quite successful operational integration training studies in which structural elaboration has ensued. This evidence suggests that once a structure has been operationally integrated for lower level concepts, the same structure can, through operational integration training, become reintegrated for structurally more complex concepts. Because most studies in this area were methodologically questionable, more precise research is required. The regression studies can also be interpreted in the light of structural transformation and elaboration. Regression can be thought of as a structural transformation
Inducing cognitive development
and learning
353
since it involves a reorganization to a structure which has been transformed and which the child no longer uses. Reversion to a subordinated level is a form of structural elaboration since it involves the application of a formerly predominant structure to a particular content. Although the data from research in this area vary, our interpretation of them suggested that regression is unlikely and reversion to a subordinated level was also unlikely but could result if the structure were not consolidated at a particular level. Finally, it was suggested in the introduction that evidence confirming or disconfirming the hypothesis of structural sequentiality was scanty. Most of the evidence that we have comes from Turiel’s (1966, 1972) and Kuhn’s (1969, 1972) studies. Here we saw that most cognitive advance was to the structure one above (+ 1) the child’s predominant structure (0). Few Ss moved to the + 2 structure even when they were presented a +2 model. It is worthy of note again that of the +2 condition Ss who progressed, the majority moved to the + 1 structure of reasoning. Most studies reviewed here have not made as fine distinctions in structural advance as Turiel and Kuhn. The result was that Ss were assessed as either intuitive or concrete operational. Therefore, the only structure which Ss could have constructed while skipping the concrete operations structure would be that of the formal operations stage. No study investigated this since Ss were usually below eight years of age. Additional evidence related to sequentiality comes from regression studies. Despite the findings that regression and reversion to a subordinated level do not occur often, it should be noted that in cases where reversion occurred, there was no reversion to the -2 level in Turiel’s and Kuhn’s studies. Similarly, there was no reported evidence in other experiments where Ss reverted from the concrete operations to the preconceptual stage. Thus, evidence from studies which reported success in inducing cognitive advance or regression supports the organismic-developmental hypothesis of sequential@. In sum, it appears that research consistent with the organismic-developmental approach has considerable potential to resolve or better define outstanding theoretical issues. A close inspection of the history of the past decade of research in the area of experimentally inducing cognitive development indicates that the problems became more clearly defined, relevant data was accumulated, some apparently unsuccessful research strategies were abandoned, and assessment techniques became increasingly more refined. In short, the quality of research has improved considerably in the ten years since Flavell’s original review. If short-term training maintains its richness for generating data relevant to developmental hypotheses, it can be argued fairly that the past’s improvements augur well for the future.
354
Sidney Strauss
REFERENCES Bearison, D. J. (1969) The role of measurement operations in the acquisition of conservation. Devel. Psychol., 1, 653660. Beilin, H. (1965) Learning and operational convergence in logical thought. 1. Exper. child Psychol.,
2, 3 17-339.
Blatt, M. and Kohlberg, L. (In press) The effects of classroom discussion upon children’s level of moral judgment. In L. Kohlberg and E. Turiel (Eds.) Recent research
in moral
development.
J. H. (1963) The development psychology of Jean Piaget. New Jersey, Van
Flavell,
Nostrand. Flavell, J. H. and Hill, J. P. (1969) Developmental psychology. In P. H. Mussen and M. R. Rosenzweig (Eds.) Annual review of psychology. Palo Alto, Annual Reviews Inc. Gelman, R. and Weinberg, D. H. (1972) The relationship between liquid conservation ;;y_3;mpensation. Child Devel., 43,
New
York, Holt, Rinehart and Winston. Brainerd, C. J. and Allen, T. W. (1971) Experimental inductions of the conservation of ‘first-order’ quantitative invariants. Psychol. Bull., 75, 128-144. Brison, D. W. (1966) Acceleration of conservation of substance. J. gen. Psychol., 109, 3 1 l-322. Brown, R., Donaldson, M., Goodnow, J., Levin, H., Restle, F., and Suppes, P. (1966) A behavioral approach to instruction: General statement of research problems and strategies. In J. S. Bruner (Ed.) Learning about learning: A conference report. Washington, U.S. Government Printing Office. Bruner, I. S. (1964) The course of cognitive growth. Am. Psychol., 19, l-5. Bruner, I. S., Olver, R. R., and Greenfield, P. M. (1966) Studies in cognitive growth. New York, John Wiley and Sons. Bruner, J. S. (1967) 7th Annual Harvard University Review. Center for cognitive studies. Coxford, A. F. (1964) The effects of instruction on the stage placement of children in Piaget’s seriation experiments. The Arithmetic Teacher, 10, 4-9. Elkind, D. and Flavell, J. H., Eds., (1969) Studies in cog;nitive development: Essays in honor of Jean Piaget. Oxford,
Oxford University Press. Feigenbaum, K. D. and Sulkin, H. (1964) Piaget’s problem of conservation of discontinuous quantities: A teaching experience. J. g-en. Psycho/., 105, 91-97.
Goldschmid, M. L. (1968) Role of experience in the acquisition of conservation. Paper presented American Psychological Association. Goldstein, K. (1939) The organism. New York, American Book Company. Gruen, G. E. (1965) Experience affecting the development of number conservation in children. Child Develop., 36, 963-979. -Gruen, G. E. (1966) Note on conservation: Methodological and definitional considerations. Child Devel., 37, 979-983. Halford, G. S. (1970) A theosry of the acquisition of conservation. Psychol. Rev., 77, 302-316.
Halford, G. S. and Fullerton, T. J. (1970) A discrimination task which induces conservation of number. Child Devel., 41, 205-213. Hall, V. C. and Kingsley, R. (1968) Conservation and equilibration theory. J. gen. Psychol., 113, 195-213. Hall, V. C. and Simpson, G. J. (1968) Factors influencing extinction of weight conservation. Merrill-Palmer Q., 14, 197-210. Halpern, E. (1965) The effects of incompatibility between perception and logic in Piaget’s stages of concrete operations. Child Devel.,
36, 491-497.
Inhelder, B. (1965) Operational thought and symbolic imagery. Mono. Sot. Res. Child Devel.,
Inhelder,
30, 14-18.
B. and Piaget, J. (1958) The growth
of logical thinking from childhood to adolescence. New York, Basic Books. Inhelder, B. and Piaget, J. (1964) The early growth of logic in the child: Classifica-
Inducing cognitive development and learning
tion and seriation. New York, Harper and Row. Inhelder, B., Bovet, M., Sinclair, H., and Smock, C. D. (1966) On cognitive development. Am. Psychol., 21, 160-164. Inhelder, B. and Sinclair, H. (1969) Learning cognitive structures. In P. Mussen, J. Langer, and M. Covington (Eds.), Trends and issues in developmental psychology. New York, Holt, Rinehart,
and Winston. Kessen, W. and Kuhlman, C., Eds. (1962) Thought in the young child. Mono. Sot. Res. Child Devel.,
27, 65-82.
Kingsley, R. C. and Hall, V. C. (1967) Training conservation through the use of learning sets. Child Devel., 38, llll1126. Kuhn, D. (1969) Patterns of imitative behavior in children from 3 to 8: A study of imitation from a cognitive-developmental perspective. Unpublished doctoral dissertation, University of California, Berkeley. Kuhn, D. (1972) Mechanisms of change in the development of cognitive structures. Child Devel.,
43, 833-844.
Kuhn, D. and Langer, J. (1968) Cognitive developmental determinants of imitation. Unpublished manuscript. Kuhn, T. S. (1962) The structure of scientific revolutions. Chicago, The University of Chicago Press. Langer, J. (1969a) Theories of development. New York, Holt, Rinehart, and Winston. Langer, J. (1969b) Disequilibrium as a source of development. In P. Mussen, J. Langer and M. Covington (Eds.), Trends and issues in developmental New York, Holt, Rinehart,,
psychology.
chael’s manual of New York, Wiley.
psychology.
and Winston. Langer, J. (1970) Werner’s theory of development. In P. H. Mussen (Ed.), Carmichild
Langer, J. and Strauss, S. (1972) Appearance, reaIity and identity. Cog., I, 105 128. Larsen, G. Y. and Flavell, J. H. (1970) Verbal factors in compensation performance and the relation between conservation and compensation. Child Devel., 41,
355
965-977.
Mermelstein, E., Carr, E., Mills, D., and Schwartz, D. (1967) The effects of various training techniques on the acquisition of the concept of conservation of substance. Unpublished manuscript. Mermelstein, E. and Meyer, E. (1967) Number training techniques and their effects on different populations. Unpublished manuscript. Miller, S. A. (1971) Extinction of conservation: A methodological and theoretical analysis. Merrill-Palmer Q., 17, 3 19-334. Murray, F. B. (1968) Cognitive conflict and reversibility training in the acquisition of length conservation. J. ed. Psychol., 59, 82-87.
Murray, F. B. and Johnson, P. E. (1969) Reversibility in nonconservation of weight. Psychon. SC., 16, 285-286. Murray, F. B. (1972) The acquisition of conservation through social interaction. Devel.
Psychol.,
6, 1-6.
Piaget, J. (1960) Logic and psychology. New York, Basic Books. Piaget, J. (1965) The origins of intelligence in children. New York, Norton and Company. Piaget, J. (1965) The child’s conception of number. New York, Norton and Company. Piaget, J. (1966) Psychology of intelligence. New Jersey, Littlefield, Adams and Company. Piaget, J. (1967a) Cognition and conservations: Two views. Contemp. Psychol., 12, 530-533.
Piaget, J. (1967b) The role of the concept of equilibrium in psychological explication. In D. Elkind (Ed.), Six psychological studies. New York, Random House. Piaget, I. (1968) Quantification, conservation and nativism. Science, 162, 976-979. Piaget, J. (1970) Piaget’s theory. In P. H. Mussen (Ed.), Carmichael’s manual of child psychology. New York, John Wiley. Piaget, J. and Inhelder, B. (1971) Mental imagery in the child. London, Routledge and Kegan Paul. Piaget, J. and Inhelder, B. (1969) The psy-
356
Sidney Strauss
chology of the child. New York, Basic Books. Rest, J. (1969) Hierarchies of comprehension and preference in a developmental stage model of moral thinking. Unpublished doctoral dissertation, University of Chicago. Rest, J., Turiel, E., and Kohlberg, L. (1969) Level of moral development as a determinant of preference and comprehension of moral judgments made by others. 1. Pets.. 37, 225-252. Rothenberg, B. B. (1969) Conservation of number among four- and five-year old children. Child Devel., 40, 383-406. Rothenberg, B. B. and Orost, J. H. (1969) The training of conservation of number in young children. Child Devel., 40, 707-726.
Ripple, R. E. and Rockcastle, (1964) Piaget rediscovered:
V. N., Eds.,
A report of the conference on cognitive studies and curriculum development. Cornell Uni-
versity, School of Education. Schwartz, C. and Langer, J. (1968) Aspects of classificatory conceptualization. Unpublished manuscript. Sigel, I. E., Roeper, A. and Hooper, F. H. (1966) A training procedure for acquisition of Piaget’s conservation of quantity: A pilot study and its replication. Brit. J. ed. Psychol.,
36, 301-311.
Sigel, I. E. and Hooper,
F. H., Eds. (1968) Logical thinking in children. New York, Holt, Rinehart and Winston. Smedslund, J. (1961a) The acquisition of conservation of substance and weight in children. I. Introduction. Scan. J. Psychol., 2, 11-20. Smedslund, J. (1961b) The acquisition of conservation of substance and weight in children. II. External reinforcement of conservation of weight and of the operation and subtraction. Scan. J. Psychol., 2, 71-84.
Smedslund, J. (1961~) The acquisition of conservation of substance and weight in children. III. Extinction of conservation of weight ‘normally’ and by means of empirical controls on a balance scale. Scan. J. Psychol.,
2, 85-87.
Smedslund, J. (1961d) The acquisition of conservation of substance and weight in children. IV. An attempt at extinction of the visual components of the weight concept. Scan. J. Psychol., 2, 153-155. Smedslund, J. (1961e) The acquisition of substance and weight in children. V. Practice in conflict situations without external reinforcement. Scan. J. Psychol., 2, 156-160. Smedslund, J. (1961f) The acquisition of conservation of substance and weight in children. VI. Practice on continuous versus discontinuous material in conflict situations without external reinforcement. Scan. J. Psychol., 2, 203-210. Smedslund, J. (1962) The acquisition of conservation of substance and weight in children. VII. Conservation of discontinuous quantity and the operations of adding and taking away. Scan. J. Psychol., 3, 69-77.
Smedslund, J. (1963a) Patterns of experience and the acquisition of the conservation of length. Scan. J. Psycho/., 4, 257-264. Smedslund, J. (1963b) Patterns of experience and the acquisition of concrete transitivity of weight in eight-year-old children. Scan. 1. Psychol., 4, 251-256. Smedslund, I. (1968) Conservation and resistance to extinction: A comment on Hall and Simpson’s article. Merrill-Palmer Q., 14, 211-214.
Smith, I. D. (1968) The effects of training procedures upon the acquisition of conservation of weight. Child Devel., 39, 515-526.
Sonstroem, A. Mc.K. (1966) On the conservation of solids. In J. S. Bruner, R. R. Olver, and P. M. Greenfield (Eds.), Studies in cognitive growth. New York, John Wiley and Sons. Strauss, S. (1969) Psychological foundations of science education. In Junior science Conference. Rehovot, Weizmann Institute Press. Strauss, S. (1972) Learning and developmental theories of Gagne and Piaget: Implications for theories of curriculum development. Teachers Coil. Rec., 74, 81-102. Strauss, S. (In preparation, a) Inducing cogni-
Inducing cognitive development
tive development: A review of shortterm training experiments. II. The functionalist approach. Strauss, S. (In preparation, h) Inducing cognitive development: A review of shortterm training experiments. III. The hehaviorist-approach. Strauss, S. and Langer, J. (1970) Operational thought inducement. Child Devel., 41, 163-175. Tanner, J. M. and Inhelder, B., Eds. (1956) Discussions on child development. Vol. 4. New York, International Universities Press. Turiel, E. (1966) An experimental test of the sequentiality of developmental stages in the child’s moral development. J. Pers. sot. Psychol., 3, 611-618. Turiel, E. (1969) Developmental processes in the child’s moral thinking. In P. Mussen, J. Langer and M. Covington (Eds.), Trends and issues in developmental psychology. New York, Holt, Rinehart
and Winston. Turiel, E. (1972) Stage and transition in moral development. In R. M. Travers (Ed.), Second handbook of research on teaching. Chicago, Rand, McNally and Com-
pany. Turiel, E. and Rothman, G. R. (1972) The influence of reasoning on behavioral choices at different stages of moral
arrd learning
357
development. Child Devel., 43, 741-756. Wallach, L., Wall, A. J. and Anderson, L. (1967) Number conservation: The roles of reversibility, addition-subtraction, and misleading perceptual cues. Child Devel.,
38, 425-441.
Wallach, L. and Sprott, R. L. (1964) Inducing number conservation. Child Devel., 35, 1057-1071.
Wallach, M. A. (1963) Research on children’s thinking. In H. W. Stevenson (Ed.), Child psychology: 62nd yearbook of the National Society for the Study of Education. Chicago, University of Chicago
Press. Werner, H. (1948) Comparative of mental
development.
psychology New York, In-
ternational Universities Press. Werner, H. and Kaplan, B. (1963) Symbol formation. New York, John Wiley and Sons. Winer, G. A. (1968) Induced set and acquisition of number conservation. Child Devel., 39, 195-205. Wohlwill, J. and Lowe, R. C. (1962) An experimental analysis of the development of the conservation of number. Child Devel.,
33, 153-157.
Youniss, J. (1971) Classificatory schemes in relation to class inclusion before and after training. Hum. Devel., 15, 171183.
R&urn&
L’auteur analyse des etudes experimentales portant sur les rbgles d’engendrement qui font passer l’organisation cognitive de l’enfant dune ttape de developpement 1 une &tape ult&ieure. Les m6thodes de recherche et les resultats tendent tous B confirmer la
plupart des hypotheses fond&es sur une approche ‘organisationelle’ du dtveloppement. L’auteur examine les probltmes methodologiques et theoriques non resolus et prtsente des strategies de recherche qui pourraient aider ?I Cclaircir ces problemes.
Effects
of biasing
context
Resolving ambiguity: in the unattended earl
J. R. LACKNER2 M. F. GARRETT Massachusetts
Institute of Technology
A b&act Ambiguous sentences and a disambiguating context sentence were dichotically presented to subjects who were instructed to attend to the channel over which the ambiguous sentences .were presented. Subjects were required to paraphrase the sentence in the attended channel immediately upon its presentation. The disambiguating material (in the unattended channel) was presented at a level 5 to 10 db less intense than the attended channel; in post-test reports subjects were unable to produce any information about the content of the unattended channel. Nonetheless, for four types of ambiguity tested, the bias contexts significantly influenced the interpretation of the ambiguous sentences. This result is taken to indicate both that there is structural analysis of the material in the unattended channel and that, during their input, multiple readings are computed for ambiguous sentences.
Ambiguous sentences have come to have the same significance for psychologists concerned with understanding language comprehension as ambiguous figures have for psychologists studying visual perception. When, for example, Gestalt psychologists constructed dot patterns which could be perceived as rows or columns depending on the set of the observer, they were attempting to demonstrate the necessity of appealing to the internal organization of the observer as well as to the physical properties of the stimulus in an adequate theory of visual perception. The psycholinguist’s concern with ambiguous sentences has a similar motivation: Sentence comprehension requires a level of description which refers to the functional relations among the constituents of a sentence. Such a functional description cannot 1. This research was supported in part by NIMH Grant No. S-POl-MH13390 and NIH
Grant No. HDOS 168-02. 2. Also at Brandeis University. Cognition
1 (4).
pp. 359-372
360
.I. R. Lackner
and M. F. Garrett
be adequately expressed on the basis simply of the words of the sentence and their succession. Consequently, just as an adequate theory of visual perception must account for the multiple grouping phenomena described by the Gestalt psychologists, so must an adequate theory of language perception account for the multiple possible analyses of most sentences. Although the pervasiveness of linguistic ambiguity is well established, in most language situations it is unnoticed. Therefore, one aspect of normal language comprehension must involve selecting (on the basis of linguistic and extralinguistic context) an ‘appropriate’ reading for an ambiguous sentence while rejecting the others. The nature of this selection process is little understood. It is necessary to consider how the presence of ambiguity might affect computation of a sentence’s interpretation. Several recent experiments have addressed the question whether linguistic ambiguity has perceptual consequences. The results of these experiments do not permit a simple statement. Foss, Bever, and Silver (1968) compared subjects’ responses to ambiguous and unambiguous stimulus sentences using a ‘verification task’ (S’s task is to report the truth value of a sentence in relation to an immediately following picture). They found a significant increase in verification times only for cases in which the picture presented required the less preferred of the two readings associated with the ambiguous stimulus item. Mehler, Carey and Bever (1970) obtained a similar result using a slightly different paradigm. They also used a verification task but expressly manipulated Ss’ expectations about the interpretation of the ambiguities. Again, verification times were greater only when the presented picture demanded the ‘unexpected’ interpretation. One might interpret these results as indicating that listeners normally select only one reading for a sentence, even when there are multiple possible readings. It is only when posterior context reveals a wrong selection that ‘perceptual’ consequences of the ambiguity are found. On the other hand, MacKay (1966) found significant differences between ambiguous sentences and their unambiguous control counterparts in a sentence completion task. In this task, s’s when presented with the initial fragment of a sentence were required to provide a grammatical and meaningful completion for it. Apparently, the subjects in this experiment were not aware that some of the fragments were ambiguous, but, nonetheless, showed slower completion times for the ambiguous fragments. Presumably this is the result of interference from the ‘unused’ meaning. More recently, Foss (1970) has found an effect of lexical ambiguity in a phoneme monitor task. Subjects were required to indicate by pressing a key the point of occurrence of a particular speech sound during the aural presentation of a sentence. Reaction time to a target phoneme in a word immediately following an ambiguous lexical item was slower than reaction time following an
Resolving ambiguity
361
unambiguous lexical item. Both of these experiments suggest that the processing of ambiguous sentences differs from that of unambiguous sentences, although the nature of this difference remains to be determined. However, it is clear from the pattern of results described that measures of the perceptual consequences of ambiguity should be made during the ongoing processing of the ambiguous sentence. This is a reasonable assumption both on the basis of the experimental findings that have been discussed as well as on informal grounds. It seems unlikely that listeners regularly postpone the interpretation of a constituent much beyond the sentence containing it. For example, a recent experiment on sentence completion by Bever, Garrett and Hurtig (1971) showed an effect of ambiguity like that found by MacKay but only for incomplete constituents. It seems safe to assume that the structural analysis of a sentence is begun by a listener during its reception and that an interpretation of the sentence is assigned either before or at most a few hundred milliseconds after its termination. If this assumption is correct, then at least three major computational procedures are available to a listener when an ambiguity is encountered: (1) He can pursue both (all) of the structural options for the sentence; (2) he can postpone the assignment of structure for a time (so that some posterior context can be used to determine the appropriate reading); (3) h e can choose a single structural option on the basis of prior context or a priori likelihood and accept the possibility of error. It is presumed that in all three cases only one reading is finally selected. Experiments using the verification procedure seem to support (3) whereas experiments using the phoneme monitor task and the sentence completion task are consistent with either (1) or (2). The experiment to be described was intended to facilitate a choice among these three computational possibilities and to further our knowledge of the types of linguistic analyses that are performed on the ‘unattended channel’ in a selective listening situation.
Experimental
procedure
The procedure involved the simultaneous presentation of an ambiguous sentence to one of a listener’s ears and a sentence furnishing a disambiguating context to his other ear. For instance, a listener might be presented simultaneously with an ambiguous sentence such as (a) in one ear and a biasing context sentence (b) in his other ear. (a) The spy put out the torch as our signal to attack. (b) The spy extinguished the torch in the window.
362
J. R. Lackner and M. F. Garrett
Listeners were instructed which ear to attend to and were told to paraphrase the sentence that they heard in the attended ear. They were told to start the paraphrase before the sentence ended. The attended ear was always the ear receiving the ambiguous sentences. Subjects were told that one of the goals of the experiment was to determine how well people can paraphrase sentences of different kinds in the presence of distracting materials. They were not told until afterwards that the material presented to the two ears was sometimes related and that some of the sentences presented to the attended ear were ambiguous. The onset of the sentence presented in the unattended ear always lagged by between 200 to 750 msec the onset of the sentence in the attended ear. The unattended sentence always ended at the same time or slightly before the attended sentence. During presentation the volume of the two sentences was such that the intensity of the unattended sentence overall was 5-10 db less than that of the attended sentence. The intensity of the attended sentence was maintained as close to 70 db (re .0002 dyne/cm2) as possible. Either sentence presented monaurally was clearly understandable. There are several assumptions underlying the use of this experimental technicme. One is that in a dichotic listening situation the subject performs some form of analysis on the material in his ‘unattended’ ear. Moray (1959) has demonstrated, for instance, that the threshold for recognition of one’s own name is lower in the unattended ear than for other words (i.e., it is easier to detect one’s own name). It is also possible to obtain a conditioned galvanic skin response to a word in the unattended channel that previously had been presented in isolation paired with an electric shock (Moray, 1970). In this latter situation, subjects, although showing a conditioned GSR, do not report hearing the conditioned word. Triesman (1964) has presented strong evidence that some lexical processing of unattended messages must take place. In one of her experiments, bilinguals shadowed a message in one ear and ‘ignored’ a message in the other; the rejected message was a translation of the attended message. When one of the messages led the other by 3.5 set or less, many of her subjects reported that the meaning of the messages was identical. Effectiveness of the biasing context in the present experimental situation depends on some form of linguistic analysis being performed on the unattended message, and the demonstration of biasing would be prima facie evidence of the occurrence of such analysis. Another assumption underlying the experimental procedure concerns the possibility of influencing a subject’s reading of an ambiguous sentence through simultaneously presented biasing contexts. If the biasing context is to have any influence the subject must determine more than one reading of an ambiguous sentence while he is hearing it, or he must delay the assignment of structure long enough for the
Resolving ambiguity
363
material in the unattended channel to be received and processed. If option (3) of the computational procedures described above is true, then there should be no influence of the biasing context on his choice of analysis of the ambiguous sentences. However, if the biasing context does exert an influence, then either computational procedure (1) or (2) might be correct.
Methods
45 ambiguous sentences of different linguistic types were constructed. These consisted of 15 lexical ambiguities 3 (e.g., ‘The sailors enjoyed the port in the evening’); 10 particle-preposition ambiguities (e.g., ‘The boy looked over the old stone wall’); 10 surface structure or bracketing ambiguities (e.g., ‘Jack left with a dog he found last Saturday’); 10 deep structure ambiguities (e.g., ‘The corrupt police can’t stop drinking’). Three copies of each of the ambiguous sentences were prepared from a master recording. Each of these copies was paired with a different context sentence (‘paired’ means that for an ambiguous sentence on channel 3 a context sentence was simultaneously recorded on channel 2). Two of the context sentences were ‘biasing’ sentences that encouraged a single reading for the sentence, either reading (Rl) or reading (R2). Classification of an interpretation as Rl or R2 was arbitrary. The third context sentence was irrelevant or neutral (N) with respect to the two readings of the ambiguous sentence. Fifteen rtnambiguous ‘padding’ sentences were also recorded and paired with (unambiguous) context sentences that were unrelated to their meaning; three copies of these 15 pairs were made. These unambiguous sentences were included to reduce the likelihood that subjects would notice a relation between the material in the attended and unattended channels. The ambiguous sentences were assembled into three counterbalanced experimental groups; each experimental group contained all the ambiguous sentences; onethird of the ambiguities for a given group were paired with their Rl biasing contexts, one-third with their R2 biasing contexts, and one-third with their N contexts. The 15 unambiguous sentences (with their simultaneous context sentences) were included in each group, yielding 60 stimuli (dichotic pairs) for each group. Thus, half the sentences a subject heard in the attended ear had a biasing context (30 of the ambiguous sentences) and half had an irrelevant context (the 15 unambiguous
3. Five of the lexical ambiguities were homophones but not homographs. The results for these five were not different from those of
the other lexical ambiguities and hence they are not presented separately.
364
J. R. Lackner and M. F. Garrett
stimuli and 15 ambiguous sentences paired with neutral contexts). The order of presentation was separately randomized for each of the three experimental groups. Forty-five subjects without known hearing defects were selected for participation. Following warm-up and practice trials, each subject listened to one of the three experimental groups of sentences. The subjects were arbitrarily assigned into groups of 15, all the members of one group hearing the same experimental tape. Thus there were 15 separate paraphrases of each ambiguous sentence for each of its three experimental contexts (Rl, R2, N). Seven of the subjects in each group heard the material to be paraphrased in their right ear, and eight heard the material in their left ear.
RMlltS ‘Interpretive bias’ was assessed for each ambiguity by determining the frequencies of its two readings when it was paired with its neutral-context sentence. For example, if twelve paraphrases were of the Rl interpretation and three were R2, then the ambiguity was strongly weighted toward its Rl meaning (i.e., upon hearing this sentence in a neutral context most subjects interpret it as Rl). Since the two meanings of an ambiguity that is presented in a neutral context are rarely reported equally often, there is an asymmetry in the possible effects of the biasing contexts. If an ambiguous sentence, generally interpreted as Rl, is presented with an Rl bias context, then only a slight increase in Rl readings can be expected (e.g., if in neutral context, twelve of fifteen reports are Rl, then only three of the fifteen reports could possibly be changed by the Rl context). The effect of the biasing context sentence on the interpretation of each ambiguous sentence was assessed by subtracting the proportion of Rl interpretations that occurred in the presence of the neutral context from the proportion of Rl interpretations that occurred in the presence of the Rl and R2 biasing contexts respectively.4 This procedure yielded two proportions whose signs and magnitudes indicated both the direction and strength of the biasing effects for each sentence. The significance of the changes from neutral to bias contexts for each type of ambiguity (lexical, particle-preposition, bracketing, deep) was evaluated by Wilcoxon matched4. Proportions were always used in the analysis because it was never the case for any ambiguity that all responses to it could be scored. Sometimes the paraphrase of an ambiguous sentence was itself ambiguous and could not be included in the analysis. As-
signing readings to the paraphrase sentences was performed independently by two different scorers. Paraphrases over which there was dispute were omitted from consideration (less than 1 %).
Resolving ambiguity
365
pairs, signed-ranks tests. The bias contexts exerted a strong influence on the interpretation of all ambiguity types. In every condition, but one, the bias effects were significant (p < .05, one tail). There is some indication that the bracketing ambiguities were less influenced than the others; the R2 bias was significant, but the Rl bias was not (although it approached significance; p < .lO). No differential biasing effect was found as a function of which ear received the biasing context. A summary of the experimental results is presented in Tables l-4.
Table 1. Lexical ambiguities. Entries at left are numbers of sentences showing a change in the incidence of RI readings. The mean proportion of RI for each condition is given on the right _____~~ Number of sentences showing: Nature of biasing context
An increase in Rl readings
Rl bias
Mean proportion of Rl for 15 sentences in:
A decrease in Rl readings ~-.
12
1
No change 2
....______
.___ R2 bias
2
12
Neutral context _~ -_-...__
Bias context .69
41 1
.26 ~_____
Table 2. Particle/preposition
ambiguities. Entries at left are numbers of sentences showing a change in the incidence of RI readings. The mean proportion of RI for each condition is given on the right ~___ ._ _~__ --___ Number of sentences showing:
Nature of biasing context
An increase in Rl readings
Mean proportion of Rl for 10 sentences in:
A decrease in RI readings
No change
Neutral context
Bias context -
Rl bias
8
2
0
1
8
1
_ R2 bias
.57
.68 -__-__ .25
366
J. R. Lackner
and M. F. Garrett
Table 3. Surface structure ambiguities. Entries on the left are numbers of sentences showing a change in incidence of RI readings. The mean proportion of RI readings for each condition is given on the right Mean proportion of RI for 10 sentences in:
Number of sentences showing: Nature of biasing context Rl bias
An increase in Rl readings
A decrease in R 1 readings
6
4
No change
Neutral context
0
.63 ____..__
.56 1
R2 bias
Table
9
0
.30
4. Deep structure ambiguities. Entries on the left are numbers of sentences showing a change in the incidence of RI readings. The mean proportion of RI for each condition is given on the right Number of sentences showing:
Nature biasing context Rl bias
Bias context
Mean proportion of Rl for 10 sentences in:
of An increase in Rl readings 8
A decrease in Rl readings 2
No change
Neutral context
0
Bias context .84
-60 R2 bias
2
8
0
.35
It should be noted that a pilot run of this experiment was attempted using Harvard summer school students but failed because these subjects were unable to paraphrase the one stimulus sentence while ignoring the other. The experiment was resumed in the fall and regular M.I.T. undergraduates were able to perform the task. Usually a subject would sit with eyes closed and head cocked to the side, one hand pushing against the headphone carrying the message to be paraphrased, and immediately blurt out his paraphrase.
Resolving ambiguity
367
The experimental results make two things clear: First, some linguistic analysis of the unattended sentence took place; some indication of the nature of that analysis will be given below. Second, while a subject is listening to an ambiguous sentence and determining its meaning, both its readings are in some sense available to him. If this were not the case, it would have been impossible to bias the interpretation of the ambiguous sentences. Consider the case in which only a single reading of the ambiguity is immediately computed during the sentence’s presentation: When the computed reading and the context matched, there would be no influence of the bias, and when the computed reading and the context did not match, the ‘bias’ context would be equivalent to the neutral context; i.e., there would be no alternative analysis of the ambiguity available to establish the relevance of the bias context. However, it was possible to bias subjects’ interpretations of the ambiguous sentences, both in favor of the preferred and less preferred of its interpretations. Thus, it seems clear that both readings of the ambiguities were computed prior to the formulation of a subject’s paraphrase response, or that neither reading is determined before the biasing information is computed and can exert an influence on the analysis of the ambiguity. The question of whether multiple readings are computed for ambiguities and if so at what point requires further discussion. Before turning to these questions, however, we should consider a possible objection to the interpretation we have offered of our results; namely, that subjects took account of the biasing context after the stimulus sentence had been presented. Two sorts of observations are inconsistent with this view: The immediacy of the subject’s paraphrase response, and the answers they provided in post-test interviews. Subjects were instructed to begin their paraphrase as soon as possible (before the end of the stimulus sentence if they were able), and they complied with this instruction; there was no delay following the presentation of the stimulus sentence during which the subject could alter his interpretation to comply with the bias context. At the end of each experimental session subjects were asked whether they had noticed anything unusual about the material they were paraphrasing, and they were requested to describe as much as they could about the material in their unattended ear. None of the subjects had noticed that the material being paraphrased was often ambiguous. None of the subjects could report anything systematic about the material in the unattended ear. Several were unable to say more than that they thought it was speech; some said that there were words (in English), but most were unable to say whether there were sentences. All subjects were surprised when the nature of the experiment was explained to them and said they had no inkling of its intent. Under monaural presentation most subjects do notice when material they are para-
368
J. R. Lackner and M. F. Garrett
phrasing is ambiguous. If the biasing context were exerting its influence after the subjects had already decided on one interpretation for the ambiguity, they should surely have detected the ambiguity at least in sentences for which the biasing context represented the less likely of the two readings. This did not happen. Even if one accepts the preceding argument as ruling out any sort of gross poststimulus integration of ambiguity and bias, there remains another possible serious objection to our interpretation. One might argue that our focus on sentences as units of analysis in the experiment is misleading. For, while it is true that the sentences of any dichotic stimulus pair were simultaneous, it does not follow that all relevant portions of the two were simultaneous. One might suppose, where the disambiguating information within the bias sentence was temporally prior to the ambiguous portion of the attended sentence, that their interaction might be the same as that for any two successive portions of a normally presented sentence. If we had presented our bias sentence and then the ambiguous sentence, no one would be surprised at the correspondence of bias and ambiguity. (In fact, it is non-trivial to say what the mechanism of disambiguating is in even this apparently simple case.)5 What makes our demonstration compelling is the simultaneity of the ambiguity and its bias context. This, coupled with the lack of an opportunity for post-stimulus decision-making, renders the parallel processing or the delay model preferable to the view that a single reading is immediately computed for an ambiguity. Thus, it is important to examine the internal structure of the two sentences in each dichotic pair to assess the likelihood that only a partial analysis of the bias sentence would suffice, and that such a partial analysis might precede the input of the ambiguous portion of the attended sentence. If the parallel or delay strategies are rejected, one would predict that our experimental effects must arise from cases where a short stretch of the bias sentence could disambiguate and where that stretch preceded the occurrence of the ambiguous portion of the attended sentence. In fact, this possibility can be ruled out on the basis of a post hoc analysis of our present results that considers both the level of analysis required to obtain relevant information from the bias sentence and the temporal relation of the effective portion of the bias to the ambiguity. 5. Consider a variant of Lashley’s (1951) example: The contents of the capsized canoe were saved because of rapidirajtin / by my quickwitted fishing guide. It is not self-evident how to use the interpretation of the first clause to ‘automatically’ select the appropriate rendering of the ambiguous stretch; neither the individual lexical
items nor the entire sense of the first clause is syntactically or semantically incompatible with the ‘writing’ interpretation of the second clause. This kind of example seems to demand a judgment of the plausibility of the alternative situations. (See Forster and Ryder, 1971, for an interpretation of the effects of such variables on perception of sentences.)
Resolving ambiguity
369
The ninety bias cases (two readings for each of forty-five ambiguities) were classified in terms of whether the biasing portion of the context sentence preceded (‘pre-disambiguation’), was simultaneous with (‘simultaneous disambiguation’), or followed (‘post-disambiguation’) the ambiguous portion of the ambiguous sentence. In order to do this, however, it was first necessary to decide which portion of the bias sentence should be counted as providing the effective bias. The results for both classifications are reported below. In forty of the ninety cases the word or phrase assumed to be biasing was assigned to the pre-disambiguation category, and in the remaining fifty cases to the simultaneous (thirty-six) or the posterior (fourteen cases) categories. For the pre-disambiguation cases thirty-two showed the expected biasing effect, six the opposite and two no difference (p < .Ol, sign test); the mean proportion of changed readings was .29. For the simultaneous cases the count was twenty-six in the bias direction, eight in the opposite and two no difference 0, < .Ol); the mean proportion of changed readings was .12. For the posterior bias, the results were thirteen in the bias direction and one in the opposite (p < .Ol); mean change in readings was .30. Evidently the bias contexts were effective in all three temporal relations. Note that in our classification we required that at least one full syllable intervene between the termination of the presumed biasing word or phrase and the onset of the ambiguous portion of the attended sentence to qualify a case as ‘preceding’ context, and, similarly, that one full syllable intervene between the termination of the ambiguous portion of the attended sentence and the onset of the disambiguation to qualify a case as ‘posterior’ context. In fact, the details of our sorting system are not very crucial since the few negative cases that occur in our data were not exclusively borderline instances of either preceding or posterior context.6 If, for example, one were to eliminate the ‘simultaneous’ category by assigning all cases of temporal 6. Our classification was conservative. The classification was done by the two authors independently and results pooled. In cases where there was disagreement as to whether a word or phrase level of the bias sentence was required to yield the relevant information, we assigned the case to the word-level category. Thus, the word-level category contains some cases that are arguably phrase level, but the phrase-level category contains no plausible word-level cases. This practice inflates the word-level category and affects the classification of the temporal relations between bias context and ambiguity as well. The shorter the stretch assumed to be biasing the more likely it will have been fully input
prior to the occurrence of the ambiguous stretch of the attended sentence. Moreover, when we determined membership in the ‘pre’, ‘simultaneous’, and ‘post’ disambiguation categories, we assigned disputed cases to the temporally prior category (i.e. borderline ‘simultaneous’ assigned to ‘pre’ and borderline ‘post’ assigned to ‘simultaneous’). These practices operate to increase the number of cases counted as pre-disambiguating and to preserve the ‘purity’ of the other categories. Given the very strong effects we found, this seemed the most reasonable way to give the alternative hypothesis a chance to account for the results.
370
overlap
J. R. Lackner and M. F. Garrett
in bias context
and ambiguity
either
to the pre-disambiguation
category
(if bias was fully input first) or to the post-disambiguation category (if ambiguity was fully input first), the results would be: Forty-two cases in the bias direction, six in the opposite and two no difference for pre-disambiguation; and twenty-nine in the bias direction, nine in the opposite and two no difference for post-disambiguation (p < .Ol). The th ree category system has the virtue of showing that even the bias contexts that were well past the occurrence point of the ambiguity exerted a significant influence. The classification of the bias context sentences can also be looked at for an indication of the level of analysis of the ‘unattended’ channel. For twenty-eight of the bias sentences it seemed plausible that interpretation of the ambiguity might be determined by a single word, and for the remaining sixty-two a phrasal analysis seemed necessary. Examples of what was counted as ‘word-level’ disambiguation are (the portion of the bias context presumed to be effective is italicized): ‘The detective was cruel to the suspects’ when paired with the ambiguous sentence ‘the detective in charge of the case looked hard’; ‘it took three lawyers to handle the case’ paired with ‘he was not able to handle the case by himself’. Examples of bias sentences classified as requiring a phrase-level analysis are: ‘The sailors liked to be in port at night’ paired with ‘the sailors liked the port in the evening’; ‘I hate relatives who visit often’ paired with the ambiguous sentence ‘visiting relatives can be a bore’. Twenty-six of the word-level bias cases showed changes in the bias direction, one in the opposite direction, and one showed no change (’ < .Ol); the mean proportion of changed readings across these cases was .33. For the phrase-level bias cases, forty-six were in the expected direction, thirteen in the opposite direction, and three showed no change (p < .Ol); the mean proportion of changed readings was .18. Though it might appear so from these figures, it cannot be concluded that wordlevel bias was more effective than phrase-level bias because the types of disambiguating contexts (word or phrase level) were not balanced across ambiguity types or sentences. For example, the ambiguity type (bracketing, see Table 3) which showed the least pronounced effects of the bias was also the one for which all the bias contexts were phrase-level. Thus, one cannot say whether the apparent reduction in the strength of the bias effect is due to the nature of the ambiguity or to the level of analysis required for the bias context sentence to be effective. Regardless of the possible differences in strength of word- and phrase-level bias, however, it is clear that even where a phrase-level analysis was required the bias sentences were effective. The post hoc analysis thus does not compromise our original interpretation of the bias effects. The results indicate that the two channels were being processed independently at least up to the level of a phrasal analysis and that there was either
Resolving ambiguity
37 1
simultaneous computation of both readings of the ambiguities or a delay in their interpretation. The present experiment does not provide a conclusive basis for deciding between delay of processing and computation of multiple readings. Nonetheless, we are inclined to prefer the latter as an explanation for two reasons. First it provides a specific answer to the question of how the relevance of the bias channel to the ambiguity is determined; the delay hypothesis leaves that issue as unclear as the case of overt pre-disambiguation. Second, a parallel processing strategy explains why there should be a difference between ambiguous and unambiguous sentences in processing difficulty. A delay strategy cannot be very convincingly tied just to ambiguity (and hence account for effects like those found by Foss, MacKay or Bever et al.); one can imagine plausible ways of postponing processing by marking certain lexical items as ambiguous, but it is hard to see a reasonable way of ‘flagging’ ambiguities that are not associated with a particular lexical item (i.e., most bracketing and deep structure ambiguities) without, in fact, computing the alternative structures. However, regardless of whether one opts for a parallel processing or delay account of our results, they do seem to rule out the view that only a single analysis is assigned to an ambiguity at the point of its occurrence.
REFERENCES
Bever, T., Garrett, M. and Hurtig, R. (1971)
upon sentence
The interaction of speech perception and ambiguous sentences. Quarterly Progress Report of the Research Laboratory of Electronics, No. 101, MIT. Mehler, J., Carey, P., and Bever, T. (1970) Judging the veracity of ambiguous sentences. 1. verb. Learn. verb. Beh., 9,
Learn.
243-254.
Forster, K. I. and Ryder, L. A. (1971) Perceiving the structure and meaning of sentences. J. verb. Learn. verb. Beh., 10, 28.5-296. Foss, D., Bever, T. and Silver, M. (1968) The comprehension and verification of ambiguous sentences. Percep. Psychophy., 4, 304-306.
Foss, D. (1970) Some effects
of ambiguity
comprehension.
verb. Beh.,
J. verb.
9, 699-706.
Lashley, K. S. (1951) The problem of serial order in behavior. In L. A. Jeffress (Ed.), Cerebral
mechanisms
in behavior.
New
York, Wiley. MacKay, D. (1966) To end ambiguous sentences. Percep. Psychophy., 1, 426-436. Moray, N. (1959) Attention in dichotic listening: Affective cues and the effect of instructions. Q. .I. exper. Psychol., 9, 5’6-60.
N. (1970) Attetztion: Selective processes in vision and hearing. New York,
Moray,
Academic Press. Triesman, A. (1964) Verbal cues, language and meaning in attention. Amer. J. Psychol., 77, 206-214.
372
J. R. Lackner and M. F. Garrett
Des phrases ambigues et une phrase donnant un contexte desambiguant sont propokes separement par deux canaux auditifs a des sujets qui ont recu comme consigne de pr&ter plus attention au canal par lequel arrivent les phrases ambigues. La tlche des sujets consiste a paraphraser la phrase entendue dans le canal privilegit immediatement aprts sa presentation. Le materiel desambiguant est present6 dans l’autre canal avec une intensite moindre de 5 B 10 decibels. Les sujets,
au tours du post-test, sont incapables de dire ce qu’ils ont entendu par le canal non privilegie. Cependant, pour les quatre types de phrases ambigues presentes, le contexte influe de facon significative sur I’interpretation don&e a la phrase ambigue. Le rCsultat est interpretit comme indiquant a la fois qu’il y a une analyse structurelle du materiel prtsent6 par le canal non privilegit et que, pendant leur presentation, il y a calcul des diverses lectures des phrases ambigues.
On the problems of interpreting reasoning data: Logical and psychological approaches*
J. ST. B. T. EVANS Sir John Cass School of Science and Technology.
City of London Polytechnic
A bstract It is argued that current approaches to the study of propositional inference have been unduly influenced by the formaI structure of the problems and the logical consequences of subjects’ responses. In order to understand the psychological basis of subjects’ behaviour it is suggested that at least two types of influence must be distinguished: Those relating to the subjects’ interpretation of the sentences constituting the logical premises of the problems, and those arising from the nature of the mental operations required on a given task. Examples are cited where misinterpretation of results has arisen from an overemphasis on the logical implications of subjects’ behaviour and from a failure to distinguish interpretative from operational influences. It is proposed that a genuine understanding of the psychological determinants of reasoning behaviour can only be achieved by consideration of data from a number of different experimental tasks.
This paper constitutes an attempt to pinpoint some of the difficulties and dangers involved in interpreting the results of experiments on propositional reasoning. In reasoning experiments subjects are normally required to infer conclusions from given premises or to evaluate the truth of logical statements in relation to given evidence. A common feature of these studies is that an arbitrary criterion of the correctness, or otherwise, of subjects’ responses is provided by the rules of the propositional calculus of logic. It will be argued that experimenters have tended to be over-influenced by the logical structures of the problems when interpreting
* Requests for reprints should be sent to Dr. J. St. B. T. Evans, Psychology Section, City of London Polytechnic, Central House An-
nexe, Whitechapel 7PF, England.
High Street, London
Cognition
El
1 (4), pp. 373-384
374
J. St. B. T. Evans
the observed behaviour of their subjects and have tended to overlook psychological explanations in terms of factors quite unconnected with logic. I should like to follow Kelly (1955) and Orne (1962) in emphasising the dangers in assuming that the subjects respond to the independent variable as construed by the experimenter: We must examine all aspects of the task and experimental situation in order to isolate the true determinants of their behaviour. Kelly’s contention that it is at best meaningless and at worst highly misleading to work in the absence of a coherent theoretical framework is also relevant: The paucity of theory has, in the area of propositional reasoning at least, led to the influence of unstated and perhaps unconscious assumptions in the design and interpretation of experiments. The aim here is to seek out and examine critically these assumptions and to suggest criteria which will aid the interpretation of experimental findings. The relation between formal logic and human reasoning has been discussed with particular reference to syllogistic inference by Henle (1962) who reviews the changing fashions of philosophical and psychological opinion on the subject. Syllogistic reasoning has a longer history of study by experimental psychologists than propositional reasoning, and it is interesting to observe the attitudes to formal logic which have emerged in this research. Theories of syllogistic reasoning can be classified into three types: (i) Logical, (ii) illogical and (iii) non-logical. Henle (1962) herself suggests that subjects reason according to the rules of formal logic and explains deviations from the logically correct conclusions by showing that subjects may drop, add or alter the premises given. In other words, the problem the subject actually solves is not the problem defined by the experimenter. A theory of ‘illogical’ reasoning is one that proposes that subjects perform incorrect or inappropriate operations upon the logical structure of the problem, whereas a theory of ‘nonlogical’ reasoning explains subjects’ behaviour in relation to features of the task quite unconnected with logic. The old ‘atmosphere effect’ of syllogistic inference (Woodworth and Sells, 1935; Sells, 1936) is essentially a theory of non-logical reasoning: Subjects are assumed to ignore the logical structure of the problems and evaluate the conclusions of the argument according to whether they match the atmosphere of the premises. For example, a premise containing a negative creates a bias towards choosing a conclusion which contains a negative. The principal critics of this theory, Chapman and Chapman (1959) propose an alternative theory which postulates illogical reasoning processes. The subject is supposed to reason according to logical rules which do not conform to those of formal logic; for example he will accept inferences considered by logicians to be fallacious. (For recent discussions of these theories, see Begg and Denny, 1969; Roberge, 1970.) Before considering the application of these distinctions to current theorising about propositional reasoning, I should like to illustrate the dangers of analysing any
On the problems
of interpreting reasoning data
375
reasoning study in terms of whether responses are logically ‘correct’ or ‘erroneous’. The problem is particularly acute where comparisons between logically valid and logically fallacious inferences are concerned. Evans (1972b), for example, examined the tendencies of subjects to make or withhold inferences about conditional rules, in which the presence and absence of negative components was systematically varied. The rules concerned an arbitrary population of letter-number pairs, for example: ‘If the letter is A then the number is 3’. With an affirmative rule of this sort most subjects make the valid inference (known as the modus tollens) that the letter cannot be A if the number is not 3, but withhold the fallacious inference (known as the affirmation of the consequent) that the letter must be A if the number is 3. When a negative is introduced into the antecedent of the rule, however, as in ‘if the letter is not G then the number is 6’, most subjects withhold the valid inference that the letter must be G if the number is not 6, and make the fallacious inference that the letter cannot be G if the number is 6. Viewed in terms of logical competence, these results are consistent. The introduction of a negative, with its known psychological difficulty, has produced logical errors: The valid inference is withheld while the fallacious inference is affirmed. The view here, however, is that the logical validity of the inference is psychologically irrelevant; the problem to be explained is why any inference tends or does not tend to be made. Hence the negative produces znconsistent effects, one inference being facilitated and the other inhibited, and a more complex psychological explanation is required. The view that human propositional reasoning conforms to the rules of formal logic will not be considered here, despite the influence of Piaget’s theory of formal operations (Inhelder and Piaget, 1958) since virtually all the experiments to be discussed find that subjects’ reasoning deviates to some extent from the laws of logic. The distinction between illogical and non-logical theories of thinking is, however, of considerable importance in this context. The thorough investigations reported by Wason and Johnson-Laird (1972) have been concerned to a large extent with investigating the extent of subjects’ logical competence, and their explanations of the many ‘erroneous’ reasoning patterns observed have been principally based on the assumption of illogical or pseudological rules of inference. Like Chapman and Chapman (1959) they propose that subjects may make ‘illicit’ conversions of premises, or adopt inappropriate logical strategies, e.g. by attempting to find verifying rather than falsifying cases of rules which they are required to test. A close examination of the Wason and Johnson-Laird studies is evidently pertinent to the argument of this paper, but first I should like to make a case for the influence of non-logical variables in a series of experiments conducted by myself, in which subjects were set problems of a highly abstract logical nature. One important factor which is proposed to influence subjects’ behaviour in these
376
J. St. B. T. Evans
experiments
is their
interpretation
of the sentences
which
constitute
the logical
propositions. The usefulness of this kind of approach is suggested by Clark’s (1969a, 1969b) linguistic theory of relational reasoning which, despite the controversy with the imagery theorists (Huttenlocher and Higgins, 1971; Clark, 1971), has generally found good empirical support (Johnson-Laird, 1972). However, Clark has been rightly criticised by Huttenlocher for focussing purely on comprehension and failing to provide any explanation of how the reasoning operations are carried out, for example in combining the information given in two premises. In the case of propositional reasoning it has also been argued (Evans, 1972a, 1972~) that reasoning cannot be explained simply by reference to the subjects’ interpretation of the premises, but that certain ‘task variables’ must be distinguished. The use of this term may be somewhat misleading since it does not refer to variables that are specific to one task but rather to the influence of certain operational requirements of the task which act independently of the subjects’ interpretation of the sentences. Hence these influences will be referred to here as ‘operational variables’. An example of such an operational variable may be illustrated by consideration of one of the findings of Evans (1972b) discussed above, viz. the inability or unwillingness to make the modus tollens inference about a rule with a negated antecedent. It is hard to conceive of any interpretation of Zf not p then cywhich would not necessitate the making of this inference. It would follow from any kind of contingency (implication, equivalence, causality, etc.) that might be conveyed by the ‘If . . . then’ structure and is further inconsistent with the ability of most subjects to find the falsifying case of such a rule (Evans, 1972~). On the other hand there are reasons to believe that an operational difficulty may arise in the process of inferring the falsity of a negative. Firstly, other inferences involving the same operation have also been made with relative infrequency, e.g. the ‘valid’ reductio ad absurdum and the ‘fallacious’ denial of the antecedent (Evans, 1972a). Furthermore this operational difficulty is consistent with a large number of studies of negation per se (for detailed discussion see Evans, 1972a). The most striking operational variable to emerge in the course of my own research was that of ‘matching bias’ (Evans, 1972~). In the original experiment subjects were asked to construct verifying and falsifying cases of conditional rules, in which the presence and absence of negative components were systematically varied l in an attempt to elicit ‘psychological truth tables’. A totally unexpected factor dominated the results: Subjects tended to choose values which matched rather than altered values named in the rules, irrespective of the presence of negatives. For example, 1. Four rules were used of the general form,
If p
then
q,
Zf
p
then not
q.
Zf not
p
then
q
and Zf not p then not
q.
On the problems
of interpreting
reasoning data
377
when asked to falsify a rule such as ‘If there is not a red triangle on the left then there is a blue circle on the right’, most subjects placed a red triangle to the left of a blue circle. Logically this constitutes a combination of a false antecedent and a true consequent: A logical combination which was rarely given to falsify rules with a different combination of negatives. The logically correct combination, true antecedent and false consequent, was least often given on this rule, which is the only one where alteration of both named values would be necessitated, e.g. by placing a green square to the left of a blue triangle. Considering only the logical classification of subjects’ responses would have led to the conclusion that they reasoned differently on rules which had different combinations of negatives; the discovery of the non-logical ‘matching bias’, however, indicates a considerable degree of consistency. Matching did, however, interact with the subjects’ understanding of the sentences since they showed varying degrees of consistency about their treatment of the same logical case on different rules. All four rules, for example, were verified almost universally by the construction of a true antecedent with a true consequent, this case being quite unaffected by matching. The construction of cases containing a falsified antecedent, argued by Wason (1966) to be psychologically irrelevant to the truth of a conditional rule, was almost entirely determined by matching. Constructions of the logically falsifying case, true antecedent plus false consequent, while significantly influenced by matching, were generally observed at some point during falsification of each rule. It would seem that the subjects’ understanding of an ‘If . . . then’ sentence entails some appreciation of when the rule may be true or false, but for cases where such influence is weak the operational effects of matching take over. The relative effects of matching suggest that while the false antecedent cases are psychologically ‘irrelevant’ (Wason, 1966) subjects are more competent at the task of verification than falsification. What exactly is matching bias? In the context of the experiment described one might think it to be a tendency to pick out or select matching values. Bearing in mind the discrepancy between subjects’ behaviour on construction and evaluation tasks in studies of both negation (Wason, 1961) and deductive reasoning (Wason and Shapiro, 1971) I recently attempted to replicate the Evans (3972~) study using an evaluation task. The results were almost identical. Whereas in the earlier experiment subjects had failed to select a case with mismatching values, they now evaluated such a case as neither conforming to nor contradicting the rule. It would seem that subjects not only fail to pick out a mismatching case but fail to see the relevance of such a case when it is presented to them. What of linguistic variables other than those connected with negation? In an experiment reported by Evans (1972a) subjects’ reasoning was compared with two forms of rule which differed in their syntactic but not in their logical structure.
378
J. St. B. T. Evans
Following Greene (1970b) the hypothesis was formulated that the syntactic form of the rule might have a semantic function in orienting the listener to a particular property of the logical structure. The hypothesis was that a rule of the general form ‘If . . . then’ is used in normal language to emphasise one logical property of implication, i.e. the antecedent is a sufficient condition for the consequent, while a rule of the general form ‘. . . only if . . .’ is used to emphasise the different logical property of the implication, i.e. the consequent is a necessary condition for the antecedent. The hypothesis was based on an intuitive appraisal of the different ways in which these two logically equivalent rules are used in normal language and is best illustrated by a concrete example. The sentence ‘If it is a dog then it has four legs’ seems to emphasise the fact that knowing something to be a dog is sufficient reason for supposing that it has four legs, whereas ‘It is a dog only if it has four legs’ emphasises that it is necessary for it to have four legs in order for it to be a dog, i.e. if it did not have four legs then it could not be a dog. The two ‘valid’ inferences associated with implication reflect these two different properties: (i) modus ponens, by which one may infer given that it is a dog the conclusion is that it has four legs, reflects the sufficiency of the antecedent; (ii) modus tollens, by which one may infer from the fact that it does not have four legs that it cannot be a dog, reflects the necessity of the consequent. As predicted there were significantly more modus ponens inferences made on ‘If. . . then’ rules and significantly more modus tollens inferences made data, in the form of concrete on ‘. . . only if. . .’ rules. An analysis of qualitative examples of the two types of sentence invented by the subjects after the experiment, gave added support to the linguistic usage hypothesis. The sentences were rarely used in interchangeable ways and where temporal or causal relations were involved their direction of operation appeared to reverse with the ‘. . . only if . . .’ rules, that is the contingencies were from the logical consequent to the logical antecedent. This is most clearly illustrated for the case of temporal relations where the constraints are grammatical. For example, ‘If I work hard then I shall pass my exams’ cannot be expressed as ‘I work hard only if I shall pass my exams’. Similarly, ‘I will go out only if I finish my essay’ cannot be expressed as ‘If I will go out then I finish my essay’. This study illustrates how the intuitive formation of linguistic hypotheses (recommended by Chomsky) can be verified by psychological experiment. Having discussed some of the evidence for non-logical variables suggested by my own research, I shall now examine the work of Wason and Johnson-Laird who have tended to interpret their own findings in terms of illogical or pseudological reasoning strategies. Much of their recent work has been concerned with a problem known as the ‘selection task’ (Wason, 1966, 1968, 1969). The general nature of this task is best illustrated by an example of a problem used by Wason (1969). Subjects were shown four cards which displayed pictures of a red triangle, a blue triangle, a red
On the problems of interpreting reasoning data
379
circle and a blue circle. From previous experience they knew that each card had a triangle on one side and a circle on the other, and that each figure was either red or blue. They were given the rule, every card which has a red triangle on one side has a blue circle on the other side and asked which cards they needed to turn over in order to find out whether the rule was true or false. Now, only a red triangle paired with a red circle could make the rule false, so logically these are the values, and the only values, which need be turned over. In practice most subjects either turn only the red triangle or both the red triangle and the blue circle. In more abstract terms the rule is of the logical form if p then q, and subjects tend to select p alone or p and q. They thus commit two logical errors: (i) Failure to select not q and (ii) redundant selection of q. Wason has consistently interpreted this as indicating that subjects are biased towards verifying rather than falsifying the rule (for an information processing model based on this see Johnson-Laird and Wason, 1970). Wason’s interpretation arises from consideration of the logical consequences of subjects’ selections; they are more likely to discover the verifying combination of p and q than the falsifying combination of p and not q. However, since the task has only been carried out on the affirmative rule if p then q the findings are also open to the interpretation that subjects are merely selecting the matching values and failing to select the mismatching values. In a recent study (Evans and Lynch, in press) subjects were given the selection task on conditional rules in which the presence and absence of negative components were systematically varied, so that each logical choice would equally often be subjected to matching and mismatching influences. If subjects were verifying as Wason proposes then they should continue to select values which would make the antecedent and consequent true and continue to omit values which would make the consequent false (for reasons discussed by Wason, 1966, he would also not expect the false antecedent to be selected). Evans and Lynch in fact showed that the selection of all four logical alternatives was largely determined by matching. If, as it would seem, matching bias is the main cause of the selections observed in Wason’s task, then it appears that its action is dependent upon the abstract nature of the materials. Recent studies have shown that subjects tend to select the logically correct alternatives when the problem is presented with thematic materials (Wason and Shapiro, 1971; Johnson-Laird, Legrenzi and Legrenzi, 1972; Lunzer, Harrison and Davey, 1972). This is quite consistent with the supposition that operational and interpretative factors interact in determining subjects’ behaviour, since strengthening the meaningfulness of the conditional premise (by using it in a realistic context) reduces the effectiveness of operational factors such as matching bias. Let US now turn to the technical difficulties involved in distinguishing the effects of interpretative and operational variables. My contention is that this can only be
380
J. St. B. T. Evans
satisfactorily achieved by consideration of data from a number of different experimental situations. In order to say that subjects’ responses reflect their interpretation of a given rule, it is necessary to show consistency in their reasoning on different tasks when due allowance has been made for the influence of the different operational variables that have been identified. For example, an analysis based on four of my own experiments and on the work of Peel (1967) led to the conclusion that whereas a rule of the form if p then not q tends to be interpreted as though it expressed the logical relation ‘p implies not q’; if not p then q tends to be interpreted as though it expressed the relation ‘not p is equivalent to q’. This is, of course, a purely formal description of subjects’ behaviour, which can only be understood by looking at the kinds of situation in which such propositions would hold. Thus, ‘p implies not q’ is true of a situation where p and q are ‘incompatible’ (Peel, 1967), so that the subject might actually reformulate if p then not q as ‘p and q cannot both be true’. Similarly, ‘not p is equivalent to q’ would hold for a situation of exclusive disjunction between p and q, so the subject may reformulate if not p then q as ‘either p or q but not both’. It would be interesting to test for such reformulations by the collection of ‘thinking aloud’ protocols although this has not yet been done. The point here is that a formal description of a subject’s interpretation of a rule, e.g. in terms of equivalence or implication, has no explanatory value psychologically although it might help to define the kind of situation in which the sentence would normally be used. Such meaning as a subject is able to ascribe to a sentence in an abstract task surely relates to the normal linguistic usage of its syntactic formulation. The failure to recognise an explicit distinction between interpretative and operational variables has led some authors to an implicit assumption that their work reflects one or the other type of influence. Legrenzi (1970), for example, assumes that eliciting a subject’s ‘truth table’ for a conditional rule yields a direct measure of its structure in ‘natural logic’ (corresponding to my conception of ‘interpretation’) against which the personal validity of subjects’ reasoning with such a rule might be assessed. It was shown in our previous discussion of the Evans (1972~) study that measures of psychological truth tables are just as prone to the influence of operational variables, such as matching bias, as are direct measures of inference. In another recent study (Johnson-Laird and Tridgell, 1972) the authors assumed that their results reflected the influence of an operational variable without any control for interpretive factors .* Johnson-Laird and Tridgell predicted on the basis of studies of negation (e.g. Wason, 196.5; Greene. 1970a, 1970b) that it would be harder to reason that a negative component of a rule was falsified by an affirmative 2. I have no argument with their highly interesting finding that contradiction is more
naturally indicated implicit negative.
by an explicit
than
an
On the problems
of interpreting reasoning data
381
premise than that an affirmative component was falsified by a negative. Previous research has shown that whereas a negative is normally used to deny a prior affirmative, affirmatives are rarely intended to make any implication about prior negatives. Johnson-Laird and Tridgell specifically predicted that subjects would more readily infer the conclusion ‘John is intelligent’ from the premises (1) ‘either John is intelligent or he is rich’ and (2) ‘John is not rich’, than the premises (1) ‘Either John is intelligent or he is not rich’ and (2) ‘John is rich’ (the conclusion follows logically in either case). Subjects took less time to reach their conclusions, and more frequently gave the correct conclusion on the first problem, results which Johnson-Laird and Tridgell take as clear confirmation of their hypothesis. Another possibility is that the first premise of the second problem, a disjunctive in which one alternative is negated, is relatively difficult to understand or interpret. In order to distinguish these interpretations of their results one should first of all test for comparable difficulty on tasks involving different sorts of sentences but the same kind of operation: If the difficulty is operational it should show up here as well. Secondly one should look at experiments involving the same types of sentence but with different reasoning measures to see if there are any consistent tendencies which could arise from the subjects’ interpretation of the rules. Fortunately evidence already exists on both these points. Firstly the operational difficulty which Johnson-Laird and Tridgell propose, i.e. that it is more difficult to see that a negative is falsified by an affirmative than the other way round, should also operate in regard to a modus tollens inference in which the consequent is negative rather than affirmative. Thus the conclusion not p should be more readily inferred from the premises if p then q and not q than if p then not q and q. This was tested by Evans (1972b), and no evidence was found to support this prediction. On the other hand, evidence from other experiments suggests that subjects do find it difficult to understand a disjunction in which one component is negated. Wason and Johnson-Laird (1969) found that subjects had more difficulty verifying such a disjunctive than one with affirmative components, but it is the introspective data which indicate most strongly the confusion which this sentence causes (see Evans, 1972a). Indeed, Johnson-Laird and Tridgell themselves report that ‘they [the subjects] complained that it was somehow ungrammatical or invalid to assert that “Either John is intelligent or he is not rich” ‘, which does not prevent them from subsequently concluding that ‘. . . our findings show that it is easy to grasp that a negative denies an affirmative, but exceedingly difficult to grasp that an affirmative denies a negative.’ These authors also show a tendency to interpret their results in terms of logical competence. Despite the fact that on the difficult problem no less than 21 of the 24 subjects deviated from the logically expected conclusion, their data are reported
382
J. St. B. T. Evans
in terms of ‘errors’. Although
they give no quantitative
analysis
of what subjects
did
conclude they comment that ‘the main error consisted in stating the opposite of the correct conclusion’. Presumably then most subjects given either p or not q and 4 conclude not p, a very interesting finding which they give little discussion. This finding would seem to me to be of far greater psychological interest than the fact that subjects failed to give the logically correct response. The conclusions of this paper must, unfortunately, prove to be rather negative. Firstly, it would appear that misinterpretations of reasoning data have arisen from a tendency to construe subjects’ behaviour as either conforming to or deviating from the laws of formal logic. This problem is evidently not restricted to the field of deductive reasoning: Cognitive psychologists in all fields could surely find similar misuse of formal systems based on linguistics, mathematics, computer programs and so on. Why do psychologists have this tendency to construe behaviour through the structures of formal models, often classifying behaviour as ‘correct’ and ‘erroneous’? Perhaps we are too easily satisfied with describing behaviour in the most elegant and parsimonious terms, without facing the complexities of genuine psychological explanation. We also have a tendency to think in terms of models which we can readily understand intuitively. Few psychologists today would give serious consideration to Watson’s ‘switchboard’ theory of the brain, yet how crude is today’s most sophisticated computer compared with the mechanism of the human mind? It is far nearer to the switchboard. The discussion of interpretative and operational influences offers a positive, albeit tentative, attempt to give some psychological explanation of subjects’ reasoning behaviour. It does, however, raise another point with disturbing implications for experimental psychology. Are the striking effects of operational variables such as ‘matching bias’ products of the experimental situation itself? We have seen in the case of Wason and Johnson-Laird’s investigations of the selection task that the ‘errors of reasoning’ observed on an abstract task can most parsimoniously be accounted for in terms of matching bias. However, when the materials used are realistic, subjects give the logically correct responses. In any case on a priori grounds relating to man’s proven intelligence one could hardly expect real-life reasoning to be as primitive as observed in these experiments. Thus while it is important to recognise and control for operational or task variables we must be aware of making them or the situations in which they flourish (e.g. abstract tasks) the main topic of our study. To do so would be to confine our interest to the psychology of psychological experiments.3
3. 1 should like and Dr. Veronika
to thank Dr. Peter Coltheart for their
Wason critical
reading
of an earlier
draft
of this paper.
On the problems
of interpreting
reasoning data
383
REFERENCES
Begg, 1. and Denny, P. J. (1969) Empirical reconciliation of atmosphere and conversion interpretations of syllogistic reasoning errors. 1. exp. Psychol., 81, 35 l-354. Chapman, A. C. and Chapman, I. P. (1959) Atmosphere effect reexamined. J. exp. Psychol.,
58, 220-226.
Clark, H. H. (1969a) Linguistic processes in deductive reasoning. Psychol. Rev., 76, 387-404.
Clark, H. H. (1969b). The influence of language in solving three term series problems. J. exp. Psychol., 82, 205-215. Clark, H. H. (1971) More about ‘Adjectives, comparatives and syllogisms’: A reply to Huttenlocher and Higgins. Psychol. Rev., 78, 505-514.
Evans, J. St. B. T. (1972a) Deductive reasoning and linguistic usage (with special reference to negation). Unpublished Ph.D. thesis, University of London. Evans, J. St. B. T. (1972b) Reasoning with negatives. Brit. .I. Psychol., 63, 213-219. Evans, J. St. B. T. (1972~) Interpretation and ‘matching bias’ in a reasoning task. Q. 1. exp. Psychol., 24, 193-199. Evans, J. St. B. T. and Lynch, J. S. (In press) Matching bias in the selection task. Brit. J. Psychol.
Greene, J. M. (1970a) The semantic function of negatives and passives. Brit. J. Psychol., 61, 17-22.
Greene, J. M. (1970b) Syntactic form and semantic function. Q. 1. exp. Psychol., 22, 14-27. Henle, M. (1962) On the relation between logic and thinking. Psychol. Rev., 69, 366-378.
Huttenlocher, J. and Higgins, T. (1971) Adjectives, comparatives and syllogisms. Psychol.
Rev., 487-504.
Inhelder, B. and Piaget, J. (1958) The growth of logical thinking. New York, Basic Books. Johnson-Laird, P. N. (1972) The three term series problem. Cog. 1, 57-82. Johnson-Baird, P. N., Legrenzi, P. and Le-
grenzi, M. S. (1972) Reasoning and a sense of reality. Brit. 1. Psychol., 63, 395-400.
Johnson-Lair-d, P. N. and Tridgell, J. M. (1972) When negation is easier than affirmation. Q. 1. exp. Psychol., 24, 87-91.
Johnson-Laird, P. N. and Wason, P. C. (1970) A theoretical analysis of insight into a reasoning task. Cog. Psychol., 1, 134138. Kelly, G. A. (1955) The psychology of personal constructs. New York, Norton. Legrenzi, P. (1970) Relations between language and reasoning about deductive rules. In G. B. F. D’Arcais and W. J. M. Levelt (Eds.) Advances in psycholinguistics. Amsterdam, London, North Holland. Lunzer, E. A., Harrison, C. and Davey, M. (1972) The four card problem and the generality of formal reasoning. Q. J. exp. Psychol.,
24, 326-329.
Orne, M. (1962) On the social psychology of the psychological experiment. Amer. Psychol., 17, 776-783. Peel, E. A. (1967) A method
of investigating children’s understanding of certain logical connectives used in binary propositional thinking. Brit. J. math. statis. PsychoL, 20, 81-92. Roberge, J. J. (1970) A reexamination of the interpretation of errors in syllogistic reasoning. Psychonom. Sci., 19, 33 l-333. Sells, S. B. (1936) The atmosphere effect: An experimental study of reasoning. Arch. Psychol.
No. 200.
Wason, P. C. (1961) Response to affirmative and negative binary statements. Brit. J. Psychol.,
52, 133-142.
Wason, P. C. (1965) The contexts of plausible denial. J. verb. Learn. verb. Behav., 4, 7-11.
Wason, P. C. (1966) Reasoning. In B. M. Foss (Eds.), New horizons in psychology I. London, Penguin. Wason, P. C. (1968) Reasoning about a rule. Q. J. exp. Psychoi., 20, 273-281.
384
J. St. B. T. Evam
Wason, P. C. (1969) Regression in reasoning? Brit. J. Psychol.,
60, 471-480.
Wason, P. C. and Johnson-Laird, P. N. (1969) Proving a disjunctive rule. Q. J. exp. Psychol.,
21, 14-20.
Wason, P. C. and Johnson-Laird, Psychology
of reasoning:
P. N. (1972) Structure
and
L’auteur pose que la plupart des Ctudes ayant trait ?i I’inftrencc propositionelle ont CtC exagkrtment influenckes par la structure formelle des probltmes et les implications logiques des kponses des sujets. 11s suggkrent, afin de mieux comprendre la base psychologique du comportement des sujets, de distinguer au moins deux types de paramktres. D’une part les paramktres qui ont trait & l’interprktation par les sujets des phrases qui constituent les premisses logiques des problPmes, et d’autre part ceux qui provien-
conterlt. London, Batsford. Wason, P. C. and Shapiro, D. (1971) Natural and contrived experience in a reasoning problem. Q. J. exp. Psychol., 23, 63-71. Woodworth, R. S. and Sells, S. B. (1935) An atmosphere effect in syllogistic reasoning. J. cxp. Psychol., 18, 451-460.
nent de la nature des opkrations mentales ntcessaires B une tache donnke. Dans les exemples prtsentks l’interprktation erronke des rtsultats est due B la trop grande importance accord&e aux implications logiques du comportement des sujets et B un manque de distinction entre les influences interpretatives et opkrationelles. Les auteurs pensent qu’une v&table comprkhension du comportement logique ne peut &tre obtenu que si l’on wnsidkre les donnks d’un grand nombre de taches expkrimentales diffkrentes.
The Ulster
depth interrogation relation to sensory
techniques deprivation
and their research’
T. SHALLICE University College London
Abstract The psychophysiological methods used by the British in Ulster to precede depth interrogation are discussed. The techniques include hooding, masking noise, wallstanding, sleep deprivation and some physical brutality. It is argued that the rationale of the techniques, which was ignored by British Government reports, is to break the resistance of the internee by increasing anxiety in a positive feedback fashion. It is further argued, again contrary to the majority Parker Report, that such techniques may well produce a long-lasting post-traumatic neurosis. One section, utilising the sensory deprivation literature examines why isolation contributes to such stressful effects. The final section argues for more stringent editorial control of papers on sensory deprivation to reduce the chances of an Orwellian use of psychology.
Introduction The use made by the British in Ulster in 1971 of psychophysiological techniques, particularly isolation methods (hooding, etc.) as the preliminary stage of depth interrogation received international publicity and condemnation. Psychologists have a particular responsibility over the use of such techniques for two reasons. Firstly, part of the motivation for the use of this type of depth interrogation is that governments can rely upon a much milder response from public opinion, as imperfectly reflected by the media, to the use of psychological methods compared with the response to physical torture methods, or even the use of drugs, as a means of softening-
1. I should like to thank Dr. C. De Monchaux and Professor P. D. Wall for their advice on technical points. The author’s pre-
sent address is the Psychology Department, National Hospital, Queen Square, London W.C. 1.
Cognition I (4) 385405
386
T. Shallice
up prisoners
for interrogation.
Thus it is important
that psychologists
should expose
the real nature of the techniques. Secondly, the particular techniques used in Ulster are an example of the way all aspects of military operations are losing their craft-like character and being approached from a scientific perspective. The Ulster techniques appear to have been developed utilising a knowledge of the sensory deprivation literature. Hence psychologists have a second and grimmer type of responsibility. After the public outcry that followed the publicising of the use of these techniques, the British Government set up two commissions of enquiry, which each produced a report - the Compton and the Parker reports. With the exception of the minority report of the Parker Commission produced by Lord Gardiner, the reports justified the continuing use of the techniques. By contrast, Gardiner argued that the techniques were both illegal and immoral. The British government did stop the use of these techniques in March 1972 but primarily because it was about to initiate a radical change of policy in Ulster - direct rule -which necessitated concessions being made to Catholic opinion. However the use of these techniques in Ulster was not an isolated aberration on the part of the British. ‘Some or all have played an important part in counterinsurgency operations in Palestine, Malaya, Kenya and Cyprus and more recently in the British Cameroons (1960-61), Brunei (1963), British Guyana (1964) Aden (1964-67), Borneo/Malaysia (196566), the Persian Gulf (1970-71)’ (Parker et al., p. 3). They have clearly been part of standard British intelligence as they were taught orally by the English Intelligence Centre, a little publicised unit of the British army, to members of the Royal Ulster Constabulary in April 1971, four months before internment (Parker er al., p. 12). Thus it seems likely that such techniques will be used again in anti-guerilla warfare where intelligence is at a premium, if not by the British Government then by others operating with similar constraints on the use of more physical tortures. (Questions of the relative effectiveness of physical and mental tortures will not be discussed in this paper.) The majority Parker Report, being more sophisticated than the Compton Report, will be treated as an example of the sort of arguments a government would use to justify the use of such techniques. Within the context of a ‘lesser evil’ argument the basic strategy used in the Report is to mislead the public over the purpose of the techniques and to belittle the long-term harmful effects. ‘The object of all the techniques is to make the detainee from whom information is required, feel that he is in a hostile atmosphere, subject to strict discipline, and that he is completely isolated so that he fears what may happen next. A further object of some of the techniques varying according to local conditions, is one of security and safety.’ Their function in fact is to break the prisoner’s mental resistance by increasing his anxiety and stress
The Ulster depth interrogation techniques
381
until he is traumatised; this will be discussed in the third and fourth sections. The Report continues referring to the medical effects; ‘Evidence we have received is to the effect that, while the techniques may produce some mental disorientation, this is expected to disappear within a matter of hours at the end of the interrogation. It is true that in a small minority of cases some mental effects may persist for up to two months.’ No evidence is in fact cited. In the fifth section it is argued that this considerably underestimate the likelihood of long-term mental scars resulting. The second reason for examining the methods employed in Ulster is that prychological research on sensory deprivation seems to have contributed to their development. Historically one of the motivations for the first sensory deprivation experiment was to understand the nature of brainwashing (Suedfeld, 1969a). In the Cold War hysteria of the early 1950s brainwashing became a headline issue with the ‘conversion’ of American prisoners in Korea to Communism. Such ‘conversions’ are in fact a very complex process, probably basically incomprehensible from within an ‘end of ideology’ ideology. However, it was easy for the public to confuse these socio-psychological methods with the different and mainly psycho-physiological breakdown methods which the K.G.B. used, particularly during the Purges. The first sensory deprivation experiments were an attempt to understand the K.G.B. methods by abstracting one of its basic components and examining it in the laboratory. This aspect of the K.G.B. method - the use of isolation -was probably learned from the secret police of many countries (see Hinkle and Wolff, 1956). The public discussion of K.G.B. methods was probably therefore a discussion of methods known to most secret police departments of that time. In this paper it will be referred to as the K.G.B. method because of the considerable body of literature on these methods as used by the K.G.B. - itself a result of the Cold War. It will be shown in the third section that the Ulster techniques operate according to similar general principles to the K.G.B. ones, but the methods have become considerably more sophisticated; the advancement in technique results from the conscious application of scientific information, particularly that on sensory deprivation. The K.G.B. produced isolation by solitary confinement in featureless cells. In Ulster it was produced by hooding, which removed all visual input, white noise sources to mask auditory input and with the prisoner in a fixed position dressed in a boiler suit, which greatly reduced the change in tactile and kinaesthetic input. The K.G.B. methods are the methods of a ‘craft’ type handed down from one interrogator to his successor. The Ulster methods are those produced by the conscious use of available scientific knowledge, for an attempt was clearly being made to reduce the change in sensory input - a scientific abstraction - to its practical limit. Not surprisingly, psychologists by investigating the nature of brain-washing have improved it.
388
T. Shallice
With the new scientific outlook of the military, further advances in interrogation methods could well be produced by their utilising a technical advance within sensory deprivation research. In the final section it is argued that the likely beneficial consequences of sensory deprivation research are insufficient to outweigh this danger. It is therefore suggested that articles on sensory deprivation should be subject to special editorial scrutiny in an attempt to reduce routine research in the field.
The Ulster techniques On August 9, 1971, internment without trial was introduced in Ulster. At 4:30 a.m. that morning 342 men were arrested under the Special Powers Act. A number of men (twelve in the initially arrested group) were treated in a much more complex way than the others, having been ‘selected for interrogation in depth’ (Compton et al., 1971). After being held for two days at Regional Holding Centres, they were transferred to an ‘interrogation centre’ at 6:30 a.m. on August llth, held there until 15:45, transferred again to Crumlin Road Jail for ‘service of detention and removal orders’ and then returned to the interrogation centre by 19:00 all on the same day. This first visit to Crumlin Road Jail probably occurred in order to avoid legal recognition of the interrogation centre. The men remained at this centre until August 17th, with one exception, when they were transferred to more permanent internment in Crumlin Road Jail. This paper is concerned with the techniques used at the centre, with its rationale and its after-effects. According to the Compton Report, while the prisoner was held at the interrogation centre, he was subjected to the following procedure, while not actually being interrogated: 1. flooding -his head was hooded in a black bag of tightly woven or hessian cloth. 2. Noise - the room in which he was held was filled with a noise described as ‘like the escaping of compressed air’ or ‘the whir of helicopter blades’ - presumably white noise (of 85-87 db [Parker et al., p. 161). 3. Wall-standing - the internee was made to farce a wall with hands high above his head on the wall and legs apart. If he moved or collapsed he was forced or lifted to regain position. Periods of wall-standing were up to 16 hours at a stretch (Parker et al., p. 12). 4. Sleep - none allowed for the first two or three days. 5. Inadequate diet - diet was severely restricted to occasional administration of dry bread and a cup of water. These allegations were substantially accepted by the Compton Report, except for those concerning food and drink. The Report claimed that bread and water were
The Ulster depth interrogation techniques
389
offered at six hourly intervals for the first four days and normal food provided thereafter, and that the men frequently refused the proferred bread and water. The Report did not decide upon one set of allegations. These were allegations of physical beatings to make the men return to the required position. These were denied by the interrogators, but medical examination on August 17th differed from that on August 11th for six of the men. With respect to each of these cases the Report reiterates a variation on the folowing: ‘The evidence we heard threw no light on the origin of the injuries.’ Since they did not hear the evidence of the intemess themselves, who refused to appear before the enquiry, the obvious conclusion is that beatings had taken place for at least six of the men.
The function of the Ulster methuds It will be argued in this section that contrary to the rationale given in the Parker Report, the method used in Ulster was an example of the ‘personality break-down’ method often used by interrogators to precede interrogation itself. First it will be shown how similar the methods used in Ulster were to the methods used by the K.G.B. in the U.S.S.R. in the purges, even though the eventual aim of the latter process was different, namely to obtain a confession, normally false, to which the prisoner would continue to subscribe at a later public trial (see Hinkle and Wolff, 1956). Then the theory of ‘personality break-down’ methods will be discussed. In Russia, as in Ulster, the arrest was normally in the middle of the night. After arrest the prisoner of the K.G.B. went through a routine and totally impersonal introduction process where administrative details were obtained, clothing removed, all with no explanation about the arrest. This lasted a few hours. In Ulster a similar impersonal non-explanatory process was used. The internee did not know internment without trial had been introduced, and the soldier making the arrest was instructed to say only ‘I am arresting you under the powers conferred by the Special Powers Act. I am not required to give any further explanation. I warn you that if you resist arrest, you may have committed an offence.’ The confusing administrative process used in Ulster was outlined in the previous section. In Russia after the introductory process, the prisoner was placed in solitary confinement in a bare cell where he had to follow a precise routine, was allowed no human contact and had nothing to read or to do. The Ulster methods are more severe versions of this isolation technique. In Russia the prisoner was allowed to sleep only at certain hours and then only in a fixed position facing a light and with inadequate blankets. In Ulster sleep was not allowed at all for the first two or three days. In both food was inadequate. The components of the process are therefore
390
T. Shallice
the same: Isolation,
sleep deprivation,
non-specific
threat,
depersonalization,
inade-
quate diet and in many cases occasional physical brutality. Hinkle and Wolff (1956) reviewing the K.G.B. methods say that the prisoner ‘gradually gives up all spontaneous action within his cell and ceases to care about his personal appearance and actions. Finally he sits and stares with a vacant expression, perhaps endlessly twisting a button on his coat. He allows himself to become dirty and dishevelled . . . The slop jar is no longer offensive to him. Ultimately he seems to lose most of the restraints of ordinary behaviour. He may soil himself. . . He follows the orders of the guard with the docility of a trained animal . . . Some prisoners may become delirious and have visual hallucinations. God may appear to such a prisoner and tell him to co-operate with his interrogator . . .However frank psychotic manifestations, other than those of the “prison psychosis” described above are not usual, primarily because those having charge of the prisoners usually break the routine of total isolation when they see that disorganisation of the prisoner’s personality is imminent.’ This breakdown process usually takes from four to six weeks according to Hinkle and Wolff. In Ulster the prisoners were at the interrogation centre for only six days. This difference in time is compensated by the greater intensity of the sleep deprivation and particularly of the isolation methods, to be discussed later. The symptoms of the internees seem to have been similar to those just discussed. Quotes from the depositions made by the internees to the Association for Legal Justice will be made in Section 5. However even the Compton Report is unintentionally informative when referring to one internee: ‘Once Mr. McClean had wet himself whilst at the lavatory and his overalls had been changed. On another occasion he had urinated on his mattress.’ The Compton Report does not account for this somewhat unusual behaviour, but it fits precisely one of Hinkle and Wolff’s comments. There are two major theories of the function of the preinterogation breakdown process - the ‘brain-syndrome’ theory and the stress theory. Each accepts the other as a subsidiary factor. Hinkle (1961) in his systematic account of the physiological state of the interrogation subject argues that its aim is to produce what he calls the ‘brain-syndrome’ an ‘across-the-board impairment of brain function: an impairment of all those aspects of brain function that are commonly tested when the physician undertakes to assess the “mental status” of the patient. A patient exhibiting this syndrome can no longer carry on his usual complex activities, assume his daily responsibilities or cope with interpersonal relations. As its symptoms develop he may become restless, talkative and delirious: ultimately he becomes totally confused and lapses into unconsciousness.’ He considers it a single syndrome, whatever had disturbed the homeostatic organisation of the brain. He discusses in detail three causes - isolation,
The Ulster depth interrogation techniques
391
sleep deprivation and fatigue - all of which were strongly present in Ulster and mentions three others, blood sugar deficiencies, temperature manipulation and altering of blood salinity. Hinkle’s concept seems rather dubious and in any case cannot account very satisfactorily for the effectiveness of the procedure. It fails to take into account the differing functions of differing parts of the brain, for particular procedures would presumably produce differential effects. For example, increasing the dosage of alcohol would roughly produce the effects that Hinkle attributes to the brain syndrome, but it is most unlikely that a process of continuous semi-intoxication would be sufficient to make men talk appropriately. Hinkle’s acount of why the ‘brainsyndrome’ is effective is tenuous: ‘As brain function is impaired, information derived from past experience generally becomes less potent as a guide for action, whereas information derived from the immediate experience, pain, thirst, discomfort and threats to life, becomes more potent.’ Such a statement while intuitively not implausible requires supporting evidence but none is presented. It is unlikely that interrogation effects can be explained just on such a simple physiological level. In fact Hinkle’s phrase ‘threats to life’ perhaps inadvertently highlights the importance of the second factor - stress and anxiety. If there were no stress it is doubtful that the brain syndrome, at least in Hinkle’s generalised form, would be effective, while stress might well be effective if strong enough, even without the ‘brain-syndrome’ being operative. This does not mean that the brain-syndrome aspects of the procedure are irrelevant. The inability to think properly would itself produce stress; it would prevent the prisoner thinking of means to cope with stress and would make him easier to interrogate. It will be argued later that it is an important part of a positive feedback stress-producing process. Sargant (1957) has in my opinion argued very cogently that the breakdown process is primarily due to the effect of stress. In particular his comparison with combat exhaustion seems apposite. Consider the stages recorded by Swank and Marchland (1946) through which a normal soldier would pass when subject to continuous combat. After a period of efficient combat which varies with personality and the severity of the fighting (on average about 50 days), soldiers ‘lost their ability to distinguish the different noises of combat . . . They became easily startled and confused and became tense. They were irritable, frequently “blew their tops”, overresponded to all stimuli . . . This state of hyper-reactivity was followed insidiously by another group of symptoms referred to as “emotional exhaustion”. The men became dull and listless, they became mentally and physically retarded, preoccupied and had increasing difficulty remembering details. This was accompanied by indifference and apathy . . . In such cases bizarre contradictory behaviour could occur.’ This seems a very similar sequence to that outlined as occurring in the K.G.B.
392
T. Shallice
situation, by Hinkle and Wolff, who indeed refer to the end product of the preinterrogation breakdown process as ‘emotional bankruptcy’. Various objections were raised to the sorts of argument Sargant produced about brain-washing (e.g., Schein, Schneier and Barker, 1961), but these were mainly relevant and valid for his conditioning explanations of the conversion process. Evidence that the K.G.B. interrogators did not consciously use Pavlovian methods (Bauer, 1957) refutes some of Sargant’s more ideological statements but is not even evidence against his analogies between the breakdown process and the effects of stress on animals. All the five components mentioned above - isolation, sleep deprivation, nonspecific threat, depersonalization and inadequate diet - are clearly stressors with the possible exception of isolation, to be discussed in detail in the next section. Thus the function of the techniques appears primarily to be to increase stress while depressing cognitive functioning rather than any of the rationalisations given in the Parker Report. The mechanism of this process will be discussed in the next two sections.
The effects of isolation
The isolation characteristics of the Ulster breakdown process are very reminiscent of the sensory deprivation process. Given the vast body of work in the field (a bibliography of the field compiled by Weinstein, Fisher, Richlin and Weinsinger, 1968, contains 45 pages of references), it is not surprising that the general pattern of findings in the long duration sensory deprivation experiments is fairly clear. As many reviews exist (e.g., Schultz, 1965; Zubek, 1969a, 1969b), I will concentrate on the cognitive and particularly the affective aspects of relevance to the interrogation usage. I will ignore differences between sensory deprivation (SD) and perceptual deprivation (PD), since the latter has at least as severe effects (Zubek, 1969b). The most widely studied situation is that of bed confinement, and all references will be to that unless otherwise stated. In this situation roughly two-thirds of subjects last for a previously specified period if that is between three to fourteen days (Zuckerman, 1964; Zubek, 1969b). Of other situations used, one with particular clinical relevance is that of the tank-type respirator (an iron-lung) where Mendelsohn et al. (1960) found that in conditions of constant noise and relatively unstructured visual stimulation (i.e., the end of the iron-lung) none of their subjects could last the required 36 hours, one lasting 30 hours but the rest terminating after between 3.5 and 10.5 hours. With the most severe situation used, the water-tank suspension method devised by Lilly (1956) Shurley (1966) found that only one subject lasted
The Ulster depth interrogation techniques
393
10 hours, the median duration being under 4 hours. Phenomenological, psychological and physiological evidence is roughly compatible with respect to cognitive deterioration. Suedfeld (1969b) mentions seven authors who report subjects’ introspections such as ‘difficulty in thinking and in concentrating’ and provides ample evidence that performance on many standard I.Q. sub-tests deteriorates, although learning and retention are unaffected. Slowing of E.E.G. rhythm occurs (see Zubek, 1969~); for example, Zubek (1964) found that fourteen days of PD slowed the E.E.G. from 0.26 to 3.56 cps per subject (with a mean of 1.63 cps). A large number of experiments indicate the anxiety-provoking nature of the situation, earlier studies being well reviewed by Schultz (1965). Perhaps the most detailed account of the affective symptomatology is given by Smith and Lewty (1959) who confined 20 subjects for as long as they could last in a silent room in which there was a bed; they were allowed to walk about. Thinking became disordered in all 18 volunteers who stayed for more than 10 hours, two complaining of ‘a jumble of thoughts with no rhyme or reason’. In 12 subjects the thought disorder produced fear and panic, one describing it as: ‘I thought this meant I was going mad.’ All had anxiety and often panic attacks either early in the session or more usually before termination. Seven subjects had body-image distortions, e.g., ‘my head is like a spinning cone going away from my body’. Five subjects had nightmares when tension and anxiety were growing, involving suffocation, drowning, killing people, etc. In two subjects there were signs of paranoid states. Reports of nightmares are in fact frequent. For example Zubek, Pushkar, Sansom and Gowing (1961) in a seven-day SD confinement report dreams such as standing on an erupting volcano, being surrounded by ferocious Indians and being in a knifefight with a giant. Paranoid delusions also occur (Freedman, Grunebaum and Greenblatt, 1961). Psychological and physiological studies fully support the strongly anxiety-provoking nature of the situation, although naturally in a more bowdlerised way. Zuckerman et al. (1962) confined 25 student nurses to a tank-type respirator for 7 hours in total darkness and with a masking noise. On the Affect Adjective Check List the experimental group checked the following words more frequently than control groups (one confined in view of the experimenter, the other unconfined): Afraid, desperate, fearful, gloomy, lonely, nervous and panicky. There was no difference between the other two groups. G.S.R. measures, which might be expected to show an increase over time as found in recumbent controls, instead show a decrease (i.e., greater arousal) over time up to the maximum studied of three days (Vernon, McGill, Gulick and Candland, 1961). Biochemical studies have produced a somewhat more complex pattern of results,
394
T. Shallice
some showing an increase in stress-associated chemicals, others not. For instance, Mendelsohn et al. (3 960) in the tank-type respirator experiment mentioned earlier found that adrenalin and noradrenaline levels both rose. However, Zubek and Schutte (1966) found that noradrenaline levels fell during a seven-day bed-confinement PD experiment as did adrenalin levels, except for subjects who terminated after more than two days (‘quitters’). It was evident from a control group that the fall can be attributed to the recumbent position. The adrenalin levels of the late ‘quitters’ rose dramatically from the third day onwards, after being apparently at quite a low level on the first two days. If one accepts a rough association of noradrenaline with anger and adrenalin with fear (see Lazarus, 1966) then the failure of these subjects to show a rise in noradrenaline is compatible with a finding of Smith and Lewty (1960) of a lack of aggressive impulses in bed-confined SD subjects. The seemingly anomalous finding of no rise in adrenalin for non-quitters may be partially explained in terms of the confounding of the biochemical measure by other variables, e.g., being recumbent, but it should be noted that bed-confinement is the only SD situation where subjects have lasted for long periods. Another experiment did show biochemical effects even with non-quitters in bed-confinement situations. Zuckerman et al. (1966) found an increase in 17 keto-steroids - a hormone which is a major part of adrenocortical output - during only an eight-hour confinement, although in a less wired-up repeat no rise was found (Zuckerman, Pcrsky, Link and Basu, 1968). The effects of sensory deprivation, confinement and social deprivation on anxiety and stress are difficult to partial out. However Zubek and MacNeill (1967) reported that compared with confinement in a coffin-like box with no PD for the same period, one week’s PD produced greater effects on scales of reported visual sensations (RVSs i.e., hallucinations), loss of contact with reality, changes in body image, reminiscence and vivid memories, sexual preoccupation, temporal disorientation and positive attitude to experimenters. Similarly Zuckerman et al. (1968) found increases in primary-process type responses and in anxiety in their SD group. How are these findings to be explained? The area was at one time described as ‘facts without a theory’ (Hebb, 1961) but quite a number of explanations have been put forwards. Although psychoanalytic (e.g., Rapaport, 1958) and information-processing (e.g., Bruner, 1961) theories have been suggested, the majority of workers in the area seem to prefer a form of reductionist optimum-arousal theory (e.g., Hebb, 1955; Schultz, 1965). Zubek (3969b), probably the most respected sensory deprivation researcher, prefers, of these, Zuckerman’s (1969b) ‘Optimal Level of Stimulation’ theory, on the grounds that it deals with certain difficulties for most arousal theories, namely some types of cognitive improvement (e.g., retention), high autonomic arousal with decreased cortical arousal and the large individual differences.
The Ulster depth interrogdion
techniques
395
In my opinion, Zuckerman avoids these difficulties by vague ad hoc assumptions, which does not augur too well for the other reductionist theories. Consider his explanation of the vital stress effects: ‘In the absence of directed cognitive activity but with diffuse stimulation from ARAS, excitation spreads along diverse pathways in associative areas of the cortex leading to an increase in ‘primary process’ thinking. The ‘lack of control’ over cognition is threatening to some persons and may result in anxiety.’ In such an explanation the role of neurophysiology seems to be merely that of providing a respectable language for psychological intuitions. To reject physiological levels of explanation in accounting for sensory deprivation phenomena would clearly be foolhardy. For instance, Zubek and Bross (1972) have results which indicate that the non-firing of nerve cells has a similar effect on the cells with which they synapse as does lesioning the former cells. However it is also obvious that a solely reductionist theory would be unsatisfactory. A complete theory would have to combine dialectically physiological and phenomenological levels of explanation and probably the information-processing level and psychoanalytic factors as well. Since psychology does not have the epistemological base upon which to construct such a multilevel theory, the rest of the section remains theoretically primitive. What is vital for understanding of the Ulster situation is to consider the phenomenological level, namely how the subject conceives of what is happening. It is apparent from published work how important this is in the sensory deprivation situation. Using the E.E.G. as a measure of the more cognitive (as opposed to affective) aspects of SD effects, Saunders and Zubek (1967) found that after seven days, E.E.G. frequency had slowed by 11 % in a group expecting to remain only seven days, but for a group expecting to remain fourteen days, it had only slowed by 5 %. While it is not clear that self-selection had been properly ruled out, similar results have been reported by Zubek, Shepard and Milstein (1970) and by Lebedinsky, Levinsky and Nefedov (1964) (m a ten-day or two-month social isolation period). Thus there is a strong suggestion from data as ‘hard’ as E.E.G. that the effects of SD depend critically upon how the subject conceives of the situation. (This point had been made originally by Jackson and Pollard, 1962, in their set-expectancy theory,) When one turns to the affective results, the paucity of the reductionist approach becomes even more obvious. For instance something as simple as the subject’s expected confirnement time is again an important factor. In most of the bed-confinement studies, roughly two-thirds of the subjects stayed for expected durations up to fourteen days. One exception is the study of Smith and Lewty (1959) who just asked subjects to last as long as they could. They had average quitting times of 29.24 hours for men and 48.70 hours for women and no subject lasted for as long as four days. Again using an unspecified endurance duration? Cohen, Silverman,
396
T. Shallice
Bressler and Shavonian (1961) found that a mere two-hour period of conventional PD was sufficient to produce confusional states and various fears including that of insanity. If we turn to people undergoing PD in a non-experimental situation, where the situation would be phenomenologically very different, the stressful nature of PD becomes even more apparent. Neurotic patients undergoing supposedly therapeutic PD have had psychotic episodes (Azima and Cramer-Azima, 1,957). Moreover such episodes are occasionally but consistently reported after eye operations which temporally remove vision (Jackson, 1969), although it is possible that in some cases organic damage is also involved. It should be noted that if an interaction between organic damage and PD is the important causative factor, this exists in the Ulster situation if one views the ‘brain-syndrome’ as temporary organic impairment. In order to understand the relevance of the stressful nature of SD to the Ulster method, it is necessary to consider the way stress develops. A suggested outline of the events leading to quitting will therefore be presented. Low cognitive arousal in a totally unstructured environment increases the chance of a disturbing dream, distortion of the body image or hallucination which in turn may well increase anxiety which could in any case be increased by reflection on disordered thought or even mere boredom. The lack of extraneous demands in the situation prevents attention being redirectcd, and the disordered nature of the cognitive processes lessens the chance of the anxiety being rationally defended against. The increased anxiety itself increases the probability of disturbing subjective events and of bodily discomfort, and so a positive feedback spiral of increasing stress begins which can only end by the subject leaving the situation. Noone to my knowledge has attempted to provide evidence for such a positive feedback process, but circumstantial evidence exists. Smith and Lewty (1959) say: ‘Our observations show a fairly clear-cut chain of events. At first volunteers show a tendency to sleep [obviously not possible in the Ulster situation] . . . Then follows a period of growing agitation, tension, and restlessness. Disturbed thinking, particularly obsessional, occurs about this stage, and most subjects experience panic which makes them leave the room. . . A 20-year-old girl showed none of the usual premonitary signs, then very quickly became restless and developed the idea that she was locked in the room. This she knew was untrue but the idea persisted and increased in intensity together with the feeling of distress. The miniature and experimental obsessive-ruminative state developed over an hour or so and the ensuing anxiety and panic was only relieved when she pressed the plunger to let herself out.’ Zubek and Schutte’s (1967) steep rise in the adrenalin levels of late quitters is also relevant. Late quitters show a rapidly increasing adrenalin level only over the last two days of their stay. Zubek and Schutte suggest that these subjects
The Ulster depth interrogation techniques
397
fight unsuccessfully against the rising stress, unlike early quitters who leave at the first sign of discomfort. Since non-quitters in their bed-confinement situation do not show a rise, the process presumably never starts with them. There is no evidence that it can be stopped once underway. The idea that unpleasant subjective sensations are more likely to occur in situations of higher anxiety is supported by the predominant view that RVSs (reported visual sensations) tend to occur in medium or high arousal states (Zuckerman, 1969a). For instance, Zuckerman and Hopkins (1966) found that subjects giving RVSs scored significantly more highly on anxiety, depression and hostility scales.
The after-effects
of this pre-interrogation
process
In the Ulster situation the internees had a thick black bag over their heads, were subject to a loud masking noise, had to remain in a fatiguing and painful fixed position while dressed in a boiler suit. Thus there was virtually no variation in visual, auditory or tactile input - a clear PD situation. Moreover they were strongly physically restrained, a situation that one knows exaggerates PD effects from the results of tank-type respirator experiments such as that of Mendelsohn et al. (1960) in which only one subject lasted for longer than 11 hours. In Ulster subjects were ‘at the wall’ for durations of 43, 40, 30, 29, 20, 15, 14, 13 and 9 hours. Sleep was prevented and food was inadequate. Thus cognitive functioning would be impaired. Pain would be present both from beatings and from the use of the ‘stoika’ position at the wall. Finally anxiety must have been at a high level for the internees even before sensory deprivation began, especially as noone knew that internment without trial was to be introduced, let alone that they were to be arrested and subjected to the depersonalization and disorientation of the arrest and initial imprisonment process. Thus one would expect the positive feedback process, discussed in the previous section, to operate starting from an initially high level of stress with no need because of objective danger for chance ‘id’ factors to initiate it (what could reasonably be considered a paranoid delusion under such objective circumstances?) and with cognitive functioning much impaired so that rational defences would be impossible. Clearly very high levels of stress would be reached rapidly. All the reports about the internees’ state support this conclusion. In their depositions to the Association for Legal Justice Auld, for instance, says: ‘Before this I heard strange noises, screams and my only desire was to end all this pain and confusion by killing myself. This I tried to do in my thoughts by striking my head on a pipe but without success.’ Another, Hannaway, said, ‘I prayed for God to take my life.’
398
T. Shallice
Dr. P. P. O’Malley,
a psychiatrist
who saw three of the men ten days after inter-
rogation was interviewed by Wade (1972) for Science. He said that he estimated all three men had developed a psychosis within 24 hours after the start of the interrogation. The psychosis consisted of loss of sense of time, perceptual disturbances leading to visual and auditory hallucinations, profound apprehension and depression, and delusional beliefs. As an example of the latter, one man said he heard voices singing evangelistic hymns and saw it was a choir conducted by Ian Paisley, a Protestant leader. He believed Paisley was about to lead a general slaughter of the Catholics. One of the commonplaces of military psychiatry is that all men have a breakingpoint when stress becomes too great and too persistant. The classic situations where this occurs is after prolonged combat or after a brief, highly traumatic incident, which naturally occurs especially frequently in wartime (see e.g., Military neuropsychiatry, 1946). By comparison with the magnitude and duration of stress, personality variables are relatively unimportant (Hocking, 1970). The after-effects of such highly traumatic experience are fairly well-known. Laughlin (1967) says: ‘Posttraumatic symptoms in psychiatry can run the gamut of clinical manifestations. On the other hand, we also recognise that anxiety, hygeiaphrontic (i.e., hyperchondriacal) and conversion (i.e., hysterical) reactions are the most frequent ones bearing a relationship to trauma. Phobias, depressions, emotional fatigue and the obsessivecompulsive reactions are less frequently observed.’ Anxiety can produce gastrointestinal, cardiovascular and genito-urinary symptoms, tremors and sleep disturbance as well as its subjective reactions. Moreover Hocking (1970) has argued from much evidence that such symptoms may well be very long-lasting, if not permanent. From the limited reports so far available, discussed earlier, it seems that the interrogation methods used in Ulster must have been traumatic. One would therefore anticipate that long-term post-traumatic neuroses could well result. So far little information is available on this point. Dr. O’Malley (see Wade, 1972) diagnosed that two of the men he saw would probably recover. With one, though, he thought there was ‘a chance of permanent mental damage’. Other reports (see Wade, 1972) refer to one man who ‘shakes continually and finds it hard to articulate sentences’ and who ‘could not be alone at any time’. Another is reported as ‘shuddering spasmodically and complains of violent headaches, insomnia and nightmares when he does get to sleep’. Neither of these is the man of whom Dr. O’Malley made his pessimistic diagnosis. I have no information on his present state. The British Government, six months after the events, refused permission for a British Society for Social Responsibility in Science team of two psychiatrists and a neurologist (Drs. O’Malley, Storr and Wall) to interview the still-interned men. Further investigations are being organised by the Eire Government.
The Ulster depth interrogation techniques
399
Conclusions It can be concluded that the Ulster pre-interrogation methods had as their aim to so increase levels of anxiety and stress while decreasing the efficiency of cognitive functioning that the prisoners could be broken. In common with other methods of brain-washing and torture, it is likely to leave long-lasting mental scars. As the attempted public relations exercises of the Compton and majority Parker Reports demonstrate, these are factors that governments will attempt to hide, especially as the apparently unbrutal nature of the techniques is presumably one reason why they have been used by a government such as the British, relatively sensitive to public opinion. Moreover, as the lack of clearly visible after-effects makes allegations of the use of such techniques much more difficult to substantiate than for physical tortures, there may well be an increasing use of such techniques in future. It is easy to condemn governments for the use of such techniques. More important is to assess the responsibility of psychologists as a group. Clearly the only area of psychological research that could have helped in the development of these techniques is sensory deprivation research, since the general principles of anxiety-production and cognitive disorganisation are traditional interrogation methods. However, if one compares the Ulster techniques with K.G.B. ones, the isolation methods used seem to be much more scientifically conceived; the hood, white-noise, the unvarying position and the boiler-suit all fit an attempt to produce as unchanging a sensory input as practical in the circumstances. They have clearly been produced by people who have absorbed the sensory deprivation literature. The way that the techniques were taught to the Ulster police by officers of the English Intelligence Centre strongly suggests that the use of such methods is now organised utilising any scientific information available. There is no reason to suppose that the British Army is alone in this interest. Indeed the book by Biderman and Zimmer (1961) devoted to research on interrogation methods, was sponsored by the U.S. Air Force, ostensibly ‘because of their interest in problems which face the prisoner of war’. Research on sensory deprivation could well lead to further refinements in interrogation techniques since the magnitude of sensory deprivation effects are clearly dependent on the precise conditions and on the personality of the subject, and an experiment could well hit on more effective procedures even though it had another aim. For instance, it is only good fortune that the Lilly-Shurley immersion technique would be inconvenient, expensive, difficult to rationalise away, and liable to lead to suicide by drowning if used as an interrogation technique; it would be a really potent stressor. Thus pure research in this area could well lead to an increase in the effectiveness of torture. Even granted the highly implausible premise that one’s own government would not misuse these techniques, it would be a man of
400
T. Shallice
rare political persuasion who would argue that Orwellian techniques are best made available for all governments; open publication of sensory deprivation research increases this danger. Given that sensory deprivation research probably has and possibly could aid in the development of interrogation techniques, it seems appropriate to take stock of such research, to see what its positive contribution has been and is likely to be. Since the initial Montreal experiments, an enormous amount of research has been done, but it seems to have contributed little of general theoretical interest to the rest of psychology with the exception of its impetus for arousal theory in the 1950s (e.g., Hebb, 1955). Zubek and Bross’s (1972) recent work may be another exception but there are not many. Why then has all this work been done? One important cause is that military agencies have pumped a considerable amount of money into the research, for obvious reasons. However another vital reason seems to be the sociological phenomenon first discussed by Tulving and Madigan (1970) in the context of verbal learning research, ‘the functional autonomy of methods’, describing it by, ‘yesterday’s methods have become today’s objects of study’. This phenomenon probably results from the combination of empiricist philosophy and publication pressures characteristic of modern psychological research. Within such a context sensory deprivation, an intuitively interesting method which allows endless possibilities for variation because of its complexity, would naturally attract a lot of researchers. Thus a research field develops whose methods and problems are internally generated. For workers within it, its relevance to other fields of science or its applications become of secondary interest. It becomes self-perpetuating and semi-autonomous, dependent only on external agencies for financial support and psychology as a whole for its academic respectability. Yet does the continuing existence of such a field mean that it needs no external justification? Consider Goldberger’s (1970) relatively favourable review in Science of Zubek’s (1969a) book: ‘As the present volume makes abundantly clear, the original aim (i.e., to discover the effects of sensory curtailment) met with obstacles from a variety of methodological sources, so that the interpretation of findings was difficult . . . Moreover with increasing experience it became evident to most investigators that another very troublesome objection to unambiguous findings lay in the fact that a subject undergoing sensory deprivation is in a totally altered life situation that affects his self-systems, defences, fantasies, motivation and cognitive and interpersonal strategies . . . Despite these obstacles, however, it would be fair to say that the bulk of evidence gathered over the last ten years in the main supports the original claims.’ With any knowledge of the effects of solitary confinement, for example Charles Dickens’ ‘I hold this slow and daily tampering with the mysteries
The Ulster depth interrogation techniques
401
of the brain, to be immeasurably worse than any torture of the body’, the general nature of the original claims should not have been too surprising. The succeeding body of studies have mainly demonstrated that the situation is not easily investigated by precise experimental methods. Such a record hardly suggests that the field will produce advances of general interest and does not justify the ever-present danger that some technical advance within the field could lead to an improvement of depth interrogation methods. While a knowledge of the gross effects of sensory deprivation is relevant in certain clinical fields (e.g., eye surgery), medical advances would seems to be most likely to result from relatively direct clinical investigations, given the unique conditions of the subject and the particular medical constraints. As Jackson (1969) says: ‘Although clinical sensory deprivation covers a lot of ground and offers many possibilities for worthwhile investigation the unique characteristics of different groups and settings far outnumber their similarities.’ Although virtually any technique which increases SD effects with normal subjects could have relevance for interrogation methods, its chance of being clinically relevant seems much less. Consider, for example, the effects of restraint, known to increase SD effects. This knowledge is utilised in the wall-standing technique, but it would not seem reasonable to allow patients who have just received eye surgery to walk around, thus showing it is much less easy to utilise this knowledge medically. Thus work of the ‘functional autonomy of method’ type seems less likely to have socially beneficial side-effects than socially harmful sideeffects. Such an assessment of a research field can only be very provisional, especially as public discussion of such issues is rare. More important, in a field which seems unlikely to lead to general theoretical advances and whose primary relevance is therefore applied, judgment of the social consequences of the work is entirely controlled by individual research workers and individual funding agencies. In the present social system of research, research workers do not have to provide any public and debatable justification of the likely social consequences of their work. Technical competence, being much easier to assess than relevance, dominates the publication process. Ideally, articles in academic journals on such a primarily applied area, where potential applications appear more likely to be socially damaging, should contain a justification of the significance of the work whose points of reference lie outside the field itself, e.g., potential clinical application, but may lie in other parts of science. This justification should be critically scrutinised by editors and referees. It would be specially relevant for research funded by military agencies. Since the publication system is the backbone of the academic career structure, this would discourage routine paper-accumulation in the area and in this case end the ‘functional autonomy’ of the method.
402
T. Shallice
It could be argued that restricting open publication of such research would effectively damage the sort of work carried out in this paper, namely that of exposing the true nature of such methods as depth interrogation. This is a minor matter compared with limiting the effectiveness of such methods themselves, the aim of forcing more of the research to be secret. In such cases failure to discover is preferable to neutralisation of the discovery as the history of atomic weapons should have taught us. Alternatively, it could be argued that such a proposal would produce a distortion of unfettered pure research. ‘Pure research’ has always been vitally if indirectly dependent for its choice of problems upon the society in which it operates (see Bernal, 1954). The idea that science if left to itself will inevitably be beneficial, a somewhat archaic argument used by scientists to restrict external interference, is a fantasy in this era of government financed science (see Greenberg, 1967; Ravetz, 1971). Within the overall social context, for scientists themselves to attempt to reorganise their community to favour socially valuable and hinder socially deleterious research is to make the crucial questions of what type of research should take place less easily controllable by external agencies whose policies are frequently determined by commercial or military considerations. Since the scientific process is so complex and because scientists themselves play such a large economic role today, a limited program of this sort could not be easily prevented by particular nation states. It has been argued to me that it is utopian to suppose that the scientific community is capable of reforming itself, that it will inevitably act as an unconscious machine for fulfilling the requirements of the dominant class or to use a term, possibly more familiar in the U.S.A., the military-industrial complex. Yet the objective long-term interests of the scientific community are in the rationally planned, socially conscious application of science, not the way it is used by capitalism (see Commoner, 1971). Moreover a failure to reform would mean the scientific community losing one of its most vital assets. The trend that has existed since the 1940s would continue with scientists ceasing to be seen as the carriers of progress and increasingly being seen as the agents of the dominant class. The response to such clearly ideological work as that of Jensen and Herrnstein could become generalised to the whole of science.
REFERENCES
Azima, H. and Cramer-Azima, F. J. (1957) Studies on perceptual isolation. Dis. nerv. Cyst. (Mon. Suppl.), 18, 80-85. Bauer, R. A. (1957) Brainwashing: Psychology or demonology. J. sot. Issues, 13, 41-47.
Bernal, J. D. (1954) Science in history. London, Watts. Biderman, A. D. and Zimmer, H. (1961) The
manipulation
of human
behavior.
New York, Wiley. Bruner, J. S. (1961) The cognitive
conse-
The
quences of early sensory deprivation. In P. Solomon, P. E. Kubzansky, P. H. Leiderman, J. Mendelson and D. Wexler (Eds.), Sensory deprivation. Cambridge, Mass., Harvard University Press. Cohen, S. I., Silverman, A. J., Bressler, G. and Shmavonian, B. (1961) Problems in isolation studies. In P. Solomon et al., (Eds.), Sensory deprivation. Cambridge, Mass., Harvard University Press. Commoner, B. (1971) The closing circle. New York, Knopf. Compton, E., Fay, E. S. and Gibson, R. (1971) Report of the enquiry into allegations against the security forces of physical brutality in Northern Ireland arising out of events on the 9th August, 1971. London, H.M.S.O.
Freedman, S. J., Grunebaum, H. V. and Greenblatt, M. (1961) Perceptual and cognitive changes in sensory deprivation. In P. Solomon et al. (Eds.), Sensory deprivation. Cambridge, Mass., Harvard University Press. Goldberger, L. (1970) In the absence of stimuli. Science, 168, 709-711. Greenberg, D. S. (1968) The politics of pure science. New York, New American Library. Hebb, D. 0. (1955) Drives and the C.N.S. (Conceptual Nervous System). Psychol. Rev.,
62, 243-254.
Hebb, D. 0. (1961) Discussion. Sensory Deprivation: Facts in search of a theory. In P. Solomon et al. (Eds.), Sensory deprivation. Cambridge, Mass., Harvard University Press. Hinkle, L. E. (1961) The physiological state of the interrogation subject as it affects brain function. In A. D. Biderman and H. Zimmer (Eds.), The manipulation of human behavior. New York, Wiley. Hinkle, L. E. and Wolff, H. G. (1956) Communist interrogation and indoctrination of ‘enemies of the sate’. A.M.A. Arch. Neural. Psychiut., 76, 115-174. Hocking, F. (1970) Extreme environmental stress and its significance for psychopathology. Am. 1. Psychother., 24, 4-26. Jackson, C. W. (1969) Clinical sensory deprivation: A review of hospitalised eye-
Ulster
depth
interrogation
techniques
403
surgery patients. In J. P. Zubek (Ed.), Sensory deprivation: Fifteen years of research. New York, Appleton-CenturyCTOftS.
Jackson, C. W. and Pollard, J. C. (1962) Sensory deprivation and suggestion: A theoretical approach. Behov. Sci., 7, 332-343.
Laughlin, H. P. (1967) The neuroses. London, Butterworth. Lazarus, R. S. (1966) Psychological stress and the coping process. New York, McGraw-Hill. Lebedinsky, A. V., Levinsky, S. V. and Nefedov, Y. G. (1964) General principles concerning (the reaction of the organism to the complex factors existing in spacecraft cabins. Paper presented at the XV International Aeronautical Congress, Warsaw. Translated from Russian by NASA, TTF-273. Lilly, J. (1956) Mental effects of reduction of ordinary levels of physical stimuli on intact healthy persons. Psychiat. Res. Rept., 5, l-9.
Mendelson, J. H., Kubzansky, P. E., Leiderman, P. H., Wexler, D., Dutoit, D. and Solomon, P. (1960) Catecholamine excretion and behavior during sensory deprivation. Arch. gen. Psychiat., 2, 147-155.
Military nerv.
neuropsychiatry ment.
Dis.,
25,
(1946) Ass. Res. Baltimore: Wil-
liams and Wilkins. Parker, Lord, Boyd-Carpenter, J. A. and Gardiner, Lord (1972) Report of the committee of Privy Counsellors appointed to consider authorised procedures for the interrogation of persons suspected of terrorism. London,
H.M.S.O. Rapaport, D. (1958) The theory of ego-autonomy: A generalisation. Bull. Menninger Clinic, 22, 13-35.
Ravetz, J. R. (1971) Scientific knowledge and its social problems. Oxford, Clarendon. Sargant, W. (1957) Battle for the mind. London, Heinemann. Saunders, M. G. and Zubek, J. P. (1967) EEG changes in perceptual and sensory deprivation. Electroenceph. Clin. Neu-
404
T. Shallice
rophysiol.,
Suppl.
25, 246-257.
Schein, E. H., Schneider, 1. and Barker, C. H. (1961) Coercive persuasion. New York, W. W. Norton. Schultz, D. P. (1965) Sensory restriction: Effects on behavior. New York, Academic Press. Shurley, J. T. (1966) Stress and adaptation as related to sensory/perceptual isolation research. Military Med., 131, 254258. Smith, S. and Lewty, W. (1959) Perceptual isolation using a silent room. Lancer, No. 2, 342-345. Suedfeld, P. (1969a) Introduction and historical background. In J. P. Zubek (Ed.), Sensory research.
deprivation: Fifteen years of New York, Appleton-Century-
Crofts. Suedfeld, P. (1969b) Changes in intellectual performance and susceptibility to influence. In J. P. Zubek (Ed.), Sensory deprivation: Fifteen years of research. New York, Appleton-Century-Crofts.
Swank, R. L. and Marchand, E. (1946) Combat neurosis: Development of combat exhaustion. A.M.A. Arch. Neurol. Psychiat., 55, 236-247.
tion:
Fifteen
years
Sensory research.
deprivation: Fifteen years of New York, Appleton-Century-
Crofts. Zubek, J. P. and Brass, M. (1972) Depression and later enhancement of the critical flicker frequency during prolonged monocular deprivation. Science, 176, 10451047. Zubek, J. P. and Macneill, N. (1967) Perceptual deprivation phenomena: Role of the recumbent position. 1. abnorm. Psychol., 72, 147-150. Zubek, J. P., Pushkar, D., Milstein, S. and Shephard, J. M. (1961) Perceptual changes after prolonged sensory isolation (darkness and silence). Can. 1. Psychol., 15, 83-101. Zubek, J. P. and Schutte, W. (1966) Urinary excretion of adrenalin and noradrenaline during prolonged perceptual deprivation. J. abnorm. Psychol., 71, 328334.
Tulving, E. and Madigan, S. A. (1970) Memory and verbal learning. Ann. Rev. Psychol., 21, 437-484. Vernon, J., McGill, T. E., Gulick, W. L. and Candland, D. K. (1961) The effect of human isolation on some perceptual and motor skills. In P. Solomon eral., (Eds.), Sensory deprivation. Cambridge, Mass., Harvard University Press. Wade, N. (1972) Technology in Ulster: Rubber bullets hit home brainwashing backfires. Science, 176, 1102-l 105. Weinstein, S., Fisher, L., Richlin, M. and Weinsinger, M. (1968) Bibliography of sensory and perceptual deprivation, isolation and related areas. Percept. mot. Skills, 26, 1119-l 163. Zubek, J. P. (1964) Behavioural and EEG changes after 14 days of perceptual deprivation. Psychon. Sci., 1, 57-58. Zubek, J. P., Ed. (1969a) Sensory deprivaYork:
Zubek, J. P. (1969b) Behavioural and physiological effects of prolonged sensory and perceptual deprivation. Nato symposium on man in enclosed space. Rome. Zubek, J. P. (1969c) Physiological and biochemical effects. In J. P. Zubek (Ed.),
of
research.
Appleton-Century-Crofts.
New
Zubek, J. P., Shephard, J. M. and Milsten. S. (1970) EEG changes after 1, 4 and 7 days of sensory deprivation: A cross sectional approach. Psychon. Sci., 19, 67-68. Zuckerman, M. (1964) Perceptual isolation as a stress situation: A review. Arch. gen. Psychiut., 11, 255-276. Zuckerman, M. (1969a) Hallucinations, reported sensations and images. In J. P. Zubek (Ed.), Sensory deprivation: Fifteen years of research. New York, Appleton-century-crofts. Zuckerman, M. (1969b) Theoretical formulations: I. In J. P. Zubek (Ed.), Sensory deprivation: New York,
Zuckerman, M., and Miller, hallucinatory lation and
Fifteen
years
of research.
Appleton-Century-Crofts. Albright, R. J., Marks, C. S. G. L. (1962) Stress and effects of perceptual isoconfinement. Psychol. Mo-
nogr., 76 (549).
Zuckerman,
M. and Hopkins,
T. R. (1966)
The Ulster depth interrogation techniques
Hallucinations or dreams? A study of arousal level and reported visual sensations during sensory deprivation. Percept. mot. Skills, 22, 447-459.
Zuckerman, M., Persky, H., Hopkins, T. R., Murtaugh, T., Basu, G. K. and Schilling, M. (1966) Comparison of the stress effects of perceptual and social isolation.
Arch.
gen.
Psychiat.,
405
14, 356-365.
Zuckerman, M., Persky, H., Link, K. E. and Basu, G. K. (1968) Experimental and subject factors determining responses to sensory deprivation, social isolation, and confinement. J. abnorm. Psychol., 73, 153-194.
Rksumt
En Ulster les Britanniques utilisent des mCthodes psychophysiologiques pour preparer les interrogatoires prolong&s. Ces techniques englobent le port de la cagoule, des bruits de fond, des stations debout prolong&s, la privation de sommeil, et des brutalites physiques. La raison d’&tre de ces techniques, dont ne font pas mention les rapports du Gouvernement Britannique, est de briser la resistance de l’inteme en augmentant son anxied par des mtthodes de feedback positif. L’argumentation soutenue va B l’encontre de la plus grande partie du rapport Parker, et sou-
tient que de telles techniques peuvent trbs bien causer des n&roses post-traumatiques de longue duree. Dans une premiere partie, en s’appuyant sur la literature ayant trait aux privations sensorielles, l’auteur a examine les raisons pour lesquelles I’isolement contribue a de tels effets de stress. La derniere partie plaide en faveur dun controle plus strict de la publication d’articles sur la privation sensorielle de facon a reduire les risques dune utilisation Orwellienne de la psychologie.
Discussions
Comments
on Herrnstein’s
response
NOAM CHOMSKY Massachusett’s
Institute of Technology
In my paper ‘Psychology and ideology’ [Cognition 1 (1) 11-461 I considered Herrnstein’s argument that a genetic component in IQ is producing a stable hereditary meritocracy and showed that the argument requires tacit premises which happen to incorporate basic elements of the prevailing ideology and unsubstantiated behaviorist doctrine. It is necessary not only to adopt his explicit premise that people labor only for gain but also the further premise that the gain sought is transmittable and cumulative, the obvious candidate being wealth. Once errors are cleared away, Herrnstein’s reply adds supporting evidence to this conclusion and reveals still more clearly the assumptions about the human species that are implicit in his general conception. Herrnstein objects: (A) That I was wrong to describe the assumption that people labor only for material gain in transmittable reward (say, wealth) as ‘[his] assumption’; and (B) that no such assumption is ‘required by any of [his] conclusions’. Neither objection stands. As to (A), I did not attribute the assumption to Herrnstein; rather, my point was that the assumption was unacknowledged, and given his reply we may now add ‘rejected’, but that once the hidden premise is made clear the argument loses all interest. Contrary to his objection (B), the assumption in question is indeed required for his conclusion that ‘there will be a hereditary meritocracy’. The same assumption is required for his ‘extrapolation’ that any viable society will be stratified in ‘hereditary social classes’ based on inherited mental capacity. Let us isolate for examination exactly what is at issue. There is no doubt that if some complex of characteristics C, partially heritable, is the factor that leads to ‘success’, then ‘success’ will partially reflect inherited differences. Furthermore, it is uncontroversial that there are vast differences in wealth, power and other ‘rewards’ in industrial societies (there is, furthermore, no detectable move towards equalization in the past several generations). When pressed, Herrnstein retreats to these truisms, but of course his article would never have been published unless it went beyond Cognition 1 (4), pp. 407-418
408
Noam Chomsky
them,
as it did in two respects.
First,
Herrnstein
asserted
that mental
capacity
as
measured by IQ is a crucial component of the complex C. Furthermore, he claims to have identified the source of existing inequalities, and since he believes this factor to be essential for a viable society, he ‘extrapolates’ to his conclusion that a hereditary meritocracy is virtually inescapable. Central to his argument is the thesis that ability ‘expresses itself in labor only for gain’, that ‘labor flows towards the rewards’, and that there is no way ‘to end the blight of differential rewards’. Herrnstein does not qualify this assumption in his response. What he denies is that the gain in question must be transmittable and cumulative (say, wealth) for his argument to go through. Let us take up in turn the points at issue. Consider Herrnstein’s claim that IQ is a major factor in producing the ‘gain’ which alone motivates labor. As I noted, Herrnstein presents no serious evidence that IQ is a major factor in determining social reward. But others have investigated the matter. Bowles and Gintis (1972) conclude from a careful analysis of the data that IQ is a minor factor and the heritable component in IQ an insignificant factor in determining income: ‘A perfect equalization of IQ’s across social classes would reduce the intergenerational transmission of economic status by a negligible amount.’ Furthermore, when IQ, social class, and education are evaluated for their independent contribution to the determination of income, ‘IQ is by far the least important’.’ Jencks et al. give as their ‘best estimate’ that there is ‘about 3 percent less income inequality in genetically homogenous subpopulations than in the entire American population’.2 Thus empirical investigations of such data as exist indicate that IQ is a minor component of the complex C which brings economic reward, and the heritable element in IQ, a negligible component. Herrnstein presents no evidence that prestige, acclaim, or any other ‘social reward’ accrues to mental capacity in general, nor is there any to my knowledge (consider the annual list of the 10 most admired men and women). Thus if mental capacity is measured by IQ and success by income (or, so far as we know, prestige, etc.), Herrnstein’s factual premise (2) that success requires mental capacity has only limited validity, and his conclusion that success reflects inherited mental capacity has virtually no force. These observations suffice to dismiss Herrnstein’s argument. However, my concern was not its empirical inadequacies but rather its ideological assumptions and 1. Samuel Bowles and Herbert Gintis (July 1972) I.Q. in the U.S. class structure, mimeographed, Harvard University. They also present an important and, I believe, persuasive discussion of the social and ideological factors that lead Herrnstein and others to stress false claims in this matter. 2. Christopher Jencks et al. (1972) Znequali-
ly: A reassessment of the effect of family and schoolirr,g in America, Basic Books, p. 221. Jencks et al. also carry out a detailed study of heritability of IQ (Appendix A), suggesting that Hermstein accepts an estimate of heritability that is far higher than what the data warrant.
Comments on Herrnstein’s response
409
the further question why any interest has been aroused in work so plainly devoid of substance. Therefore, after noting that IQ was very likely only a minor element (and its inherited component, a negligible element) in determining measurable ‘success’, I disregarded the issue and granted Herrnstein’s factual claims, for the sake of argument, a fact which apparently led him to believe that I accepted these claims as correct. Let us now turn to the second question. Still assuming, with Hermstein, that ability expresses itself in labor only for gain and that this ability is partially heritable, let us ask what conclusion can be drawn if the gain that motivates effort is nontransmittable, say, prestige and acclaim. Herrnstein takes issue with my observation that the conclusions in this case are innocuous. I did not explain the point, assuming it to be obvious, but perhaps a comment is in order, given Herrnstein’s failure to understand it. Consider two persons with greater than average ability, who attain thereby an increment R of reward beyond the average. By hypothesis, their child is likely to have higher than average ability (though less so than the parents because of regression toward the mean, as Herrnstein correctly points out) and thus will be expected to attain, by virtue of his own ability, an increment R’ of reward beyond the average, where R’ is less than R. Suppose that reward is prestige and acclaim. Then the children of talented (hence, we assume, rewarded) parents will in general be less talented and, by assumption, less acclaimed. Conceivably, one might argue that ‘prestige’ is itself transmittable and cumulative (like wealth). But this claim (which Hermstein does not make) would be most dubious. One might just as easily argue that in a competitive society, the discrepancy between R’ and R might further diminish the child’s total increment (‘he didn’t live up to expectations’). Noting further that the heritable component in IQ is a negligible component in C, and that there is no reason to suppose C itself to be a significant (or any) factor in choosing mates, we see that any tendency for perpetuation of ‘prestige’ along family lines further diminishes. On these assumptions, there will be no significant tendency for reward to be concentrated in a ‘hereditary meritocracy’. Suppose, however, that reward is transmittable wealth (and the power that flows from wealth). Then the child’s total increment is R’ f R, + Rz + RX, where R, is the portion of R transmitted to the child, R2 is the additional wealth generated by Rr itself (say, by investment), and R3 is the increment attained by the child beyond R’ by virtue of the initial advantage afforded him by possession of R,. If Rr + R2 + R3 is greater than R-R’, one can deduce a tendency towards ‘hereditary meritocracy’ on appropriate assumptions about selection of mates. As I noted in the original article, Herrnstein does continually fall into the assumption that wealth is the reward that motivates labor, as when he writes that if the social ladder is
410
Noam Clzomsky
tapered steeply, the obvious way to rescue the people at the bottom is ‘to increase the aggregate wealth of society so that there is more room at the top’ - which is untrue if ‘social standing’ is a matter of prestige and acclaim (and is incorrect for different reasons if it is wealth, since redistribution is an equally obvious strategy). I hope it is now clear that if we accept all of Herrnstein’s assumptions but take the reward that alone motivates labor to be prestige and acclaim, then we cannot deduce that ‘there will be a hereditary meritocracy’, nor will this result be increasingly likely given ‘the successful realization of contemporary political and social goals’, nor can we ‘extrapolate’ to the conclusion that in any viable society a ‘virtually hereditary meritocracy’ will appear. To reach Herrnstein’s major and ‘most troubling’ conclusions we must add the tacit premise that the motivating reward is transmittable gain, such as wealth (though conceivably one might invent some other new premise to resurrect the argument - I will disregard this possibility in ensuing discussion). Contrary to Herrnstein’s assertion that the assumption that people labor only for transmittable (material) gain is not ‘required by any of [his] conclusions’, we see that it is critical for his only nontrivial conclusions. By assuming that prestige and acclaim suffice as the motivating reward for labor, we can distinguish two conclusions that Herrnstein continually confuses in presenting his ‘syllogism’. In his response, Herrnstein argues initially that from his premises (1) that mental capacities are partially inherited and (2) that success in our society calls for those mental capacities it follows that (3) ‘success in our society reflects inherited differences between people’. In the next paragraph he states that by accepting (1) and (2) one is ‘inexorably carried on’ to the conclusion (4) ‘that society will stratify itself increasingly by genetic factors as it divests itself of the barriers commonly held to be unfair’. And shortly after, he claims that from (1) and (2) it follows directly that (5) ‘there will be a hereditary mirtocracy. In fact, (3) does follow from (1) and (2); 3 taking prestige as ‘success’ in the ‘syllogism’, it follows from (1) and (2) that prestige partially reflects inherited differences. But (5) does not follow from (1) and (2) and cannot be deduced unless we add the assumption that people labor for transmittable gain, which measures ‘success’. 3. To be precise, what (1) and (2) imply is only (3’): Success in our society requires a characteristic which is partially inherited. Further premises are needed for the conclusion that Hermstein apparently intended in his vaguely worded (3), namely, that successful people are genetically distinguishable in the population at large. Assuming (1) and (2). we can conclude nothing about genetic distinctiveness of successful people. Suppose,
say, that the heritable component in ‘mental capacity’ is negatively correlated with some characteristic P that far outweighs it as a factor in success. Then it might result that successful people are deficient in the heritable component of ‘mental capacity’, while (1) and (2) remain true. I am indebted to Ned Block for clarification on this point.
Comments on Herrnstein’s
response
411
As to (4), it does not follow from (1) and (2) if we understand by the phrase ‘will stratify itself increasingly by genetic factors’ something like ‘will stratify itself in a hereditary meritocracy’, as presumably intended. Now the conclusions that Herrnstein and his readers regarded as important were (5) and (4) under the suggested interpretation, namely, the two conclusions that do not follow from his premises. What was ‘most troubling’ was his claim that ‘the growth of a virtually hereditary meritocracy will arise out of the successful realization of contemporary political and social goals’ and that ‘social classes [will] not only continue but become ever more solidly built on inborn differences’, namely IQ. But, as I noted, no significant social consequences follow from the fact that the children of two Olympic swimmers are likely to receive acclaim for their swimming ability (though less so, on the average, than their parents). Hermstein’s claims about existing societies (as well as his ‘extrapolation’) require that the reward that alone motivates labor must be transmittable gain (and other assumptions about the extent of transmittability and choice of mates). Thus the major point in his response is clearly false. At this stage of the discussion, Herrnstein shifts his ground, introducing another assumption which is as revealing as it is unwarranted. He argues that if prestige were the reward that motivated labor, then there would be ‘a society stratified by a mortal competition for prestige’. How does he reach this conclusion? Evidently, prestige differs from wealth not only in that it is not transmittable and cumulative in the same way (or at all, as already noted), but also in that it is not in short supply. By granting more of this ‘reward’ to one individual, we do not correspondingly deprive another. Still assuming that individuals labor only for gain, then if the reward is prestige and acclaim, high performance could be assured generally by granting each individual prestige to the extent that he achieves in accordance with his abilities, whatever his task. But now consider some individual who will work only if his reward in prestige is not only greater than what he would attain by not working or working less well but also greater than the prestige given to others for their accomplishments. Such an individual might find himself in a ‘mortal competition for prestige’ and might suffer ‘sadness and regret’ (as Herrnstein predicts) or even ‘painful psychic deprivations’ if others are successful - say, if someone else writes an outstanding novel or makes a scientific discovery or does a fine job of carpentry and is acclaimed for his achievement. Rather than take pleasure in this fact, such a person will be pained by it. Hermstein assumes, without argument, that this psychic malady is characteristic of the human race. By hypothesis, individuals labor only for ‘differential reward’ in prestige. But Hermstein confuses two senses of this term. Under sense 1, the phrase ‘differential prestige’ refers to the increment an individual receives beyond others; under sense 2, it refers to the increment he obtains beyond what he would receive were he not
412
Noam Chomsky
to work or were he not to work in a way commensurate with his talents. Herrnstein speaks of the consequences of ‘attaching different outcomes for different jobs [sense 11, or for jobs done well or poorly [where a single individual is involved, sense 21’. But under sense 2, there will be no ‘mortal competition for prestige’, no pain if others are successful, and no tendency for prestige to accrue to talent. The ‘deprivation’ that makes ‘differential prestige’ an operative reward will be the lack of the prestige that would be obtained by working or working well. Herrnstein reaches the opposite conclusion in each of these respects because he slips unconsciously into sense 1 of ‘differential prestige’, thus assuming that only reward beyond one’s fellows can motivate labor and that individuals are not pained (indeed, may be rewarded) by noting the ‘sadness and regret’ or ‘painful psychic deprivations’ that they cause for others by gaining ‘differential prestige’ themselves. This assumption is even more extraordinary than the explicit assumptions in his original article; e.g., if bakers and lumberjacks were to get the top salaries and the top social approval, then the most talented would strive to become bakers and lumberjacks rather than, say, engineers, surgeons and research scientists. To summarize, if prestige and acclaim suffice as motivating reward, then contrary to Hermstein’s belief, we cannot conclude that there will be a tendency towards social stratification in a hereditary ‘meritocracy’. Furthermore, Hermstein’s belief that there must then be a ‘mortal competition for prestige’ follows only if we assume that differential prestige in sense 2 does not suffice and that an individual will labor only for the pleasure of being acclaimed beyond his fellows (thus causing them ‘sadness and regret’ or perhaps ‘painful psychic deprivations’). Continuing, let us assume that there is a complex of characteristics C4 partially heritable and a factor in choosing mates, that brings transmittable and cumulative social reward (say, wealth and power) sufficient to induce a tendency towards a hereditary ‘meritocracy’. One who finds this ‘troubling’ might simply conclude that we should change the pattern of reward. We now turn to Herrnstein’s ‘extra4. From the observation that IQ is only a minor factor in determining income, it follows that one must seek other characteristics that play the role of C in what remains of Hermstein’s syllogism, if income is taken as index of ‘success’. Jencks etal. (1972) point out that there is little empirical evidence on this matter. I noted a number of factors that Seem plausible candidates, e.g., ruthlessness, avarice, lack of concern for others, etc., and pointed out that the term ‘meritocracy’ seems to beg some interesting questions. It is a measure of Hermstein’s uncritical acceptance
of prevailing ideology that he takes this suggestion to be refuted by the fact that ‘the data’ show that those who succeed are ‘buoyant, energetic, independent, healthy ones’. Evidently, there would be no reason for surprise if in a competitive society with an ideological commitment to labor only for material gain those who are ruthless, avaricious, etc., and thus well constituted for a war of all against all, are regarded by conventional standards as ‘healthy’, or if they prove to be ‘buoyant, energetic, independent’.
Comments on Herrnstein’s
response
413
polation’ to the conclusion that this decision is impossible because no viable society can function otherwise. A society in which everyone has the opportunity to find interesting work suited to his talents - what Herrnstein predictably calls a ‘beehive society’ - cannot survive. Herrnstein raises two objections to the ‘beehive society’. First, no one will be willing to do such work as ‘coal mining, ditch digging, schooling, garbage collecting, etc.’ He considers it an ‘outlandish’ claim that people might voluntarily choose to be, say, teachers or students, unless given ‘differential reward’ in his sense (sense 1). Suppose that he is correct and that in the ‘beehive society’ no one would willingly choose to do what Herrnstein apparently regards as the dirty work of society - no problem in the society he depicts, of course, because the untalented are driven to such work by need. By assumption, the talented would find interesting work in the ‘beehive society’, but the dirty work would not get done. In my original discussion, I mentioned several possible ways to deal with this problem, which Herrnstein ignores in his reply. The obvious suggestion is that in the ‘beehive society’ onerous tasks should be shared equally. But suppose that extrinsic motivation is necessary to guarantee this and that prestige and respect do not suffice. Then a differential reward of material gain must be provided to ensure that those who do not have the talent or inclination for work that is intrinsically satisfying (or that brings acclaim) will undertake the dirty work. Herrnstein’s syllogism now ‘starts cranking away’ to give a result precisely opposite to the one he foresees, namely, material reward will accrue to those lacking in ‘mental capacity’, and, on his further assumptions about heritability and selection of mates, there will be a long-term tendency towards social stratification in a hereditary ‘anti-meritocracy’. I emphasize again that I do not believe any of this, since I see no reason to doubt that in a decent (i.e., ‘beehive’) society onerous work will be shared out of a sense of social responsibility or perhaps because of the respect afforded to achievement or socially useful work. But on Herrnstein’s assumptions, a hereditary antimeritocracy is quite a plausible outcome. Contrary to his beliefs, then, his syllogism ‘cranks away’ to give virtually any long-term result, depending on one’s implicit ideological commitments. In short, the ‘syllogism’ itself is once again seen to be empty of significant consequences, Herrnstein then turns to his second objection: He argues that ‘the burden [of proof] falls squarely on [me] to prove’ that the ‘beehive society’ is possible. This is a transparent evasion. My point was that Herrnstein’s claims, even admitting his fallacious factual assertions, rest entirely on the assumption that there can be no such decent society, that his argument collapses unless we add tacit and entirely unsubstantiated premises borrowed from behaviorist doctrine and capitalist ideology. Accordingly, I made no effort to prove that these tacit assumptions are false but rather pointed out that without them there is nothing left of his nontrivial con-
414
Noam Chomsky
elusions. Since he is the one who is claiming that society is moving (and must move) in a certain direction, clearly the burden of proof is on him to show that the assumptions required for his argument are plausible. It will not do to evade the issue by asking his critic to prove that the tacit assumptions are wrong. In another context, I would, indeed, argue that a decent (‘beehive’) society is possible, increasingly so as technological progress eliminates the need for onerous and stultifying work and provides means for democratic decision-making and for freely undertaken productive labor under conditions of voluntary association. But this is another matter entirely. Herrnstein’s final recourse, once again, is that history proves . . . But as 1 noted in the original article, this argument in itself has all the cogency of an 18th century argument that capitalist democracy is impossible, as history proves. For the argument to have any force, a reason must be adduced to explain the alleged fact that in existing societies ability ‘expresses itself in labor only for gain’, further, for gain in transmittable and cumulative reward - a necessary assumption for Herrnstein’s argument, as we have seen. One might argue, say, that it is a fact about human nature that humans will work only for such gain, that they will vegetate rather than seek an interesting outlet for their abilities, that they will never sacrifice for ideals or to help others; perfect nonsense, I am sure. The point is that for the argument from history to have any force as a projection for the future, one must be able to explain the historical facts in terms of some factors that are expected to endure. Since Herrnstein presents nothing even plausible in this regard, this argument goes the way of the others he has offered. It remains a political decision, not a conclusion of science or ‘reason’ or history, to accept forms of social organization that permit accumulation of material reward, decision-making power, control of productive resources. Those who are ‘troubled’ by the consequences of permitting free play to the complex of characteristics C that brings reward in a competitive society will therefore argue that social conditions be changed to prevent accumulation of wealth and power by those who have these characteristics. Herrnstein offers no hint of a counter-argument. Rather, his uncritical and apparently unconscious acceptance of prevailing ideology in its most vulgar form leads him to conclude that things could not be different. It is striking that Herrnstein can conceive of only one alternative to existing society. The only alternative that occurs to him is a society in which the leaders of the state order citizens to do this or that ‘for the sake of the state’. This is what he takes to be the characteristic of ‘socialist states’. On his weird assumptions about human nature, it may be that this is the only alternative to capitalism. I need hardly emphasize that it is not the socialist alternative; nor do I (or other libertarian socialists) believe that a socialist society exists today, though the germs of socialism,
Comments on Herrnstein’s
response
415
democracy and freedom exist in some measure in various societies. Herrnstein’s response contains numerous other errors.5 For example, he believes that I ‘keep dismissing’ the ‘Skinnerian principles of reward and punishment’. On the contrary, as I have repeatedly pointed out, I exclude from my criticism the actual scientific work (e.g., on partial reinforcement) which has no bearing on any of the issues raised in the work I was discussing. As to the trivialities and tautologies that are invoked to support such argument as there is, I merely urge that they be labeled for what they are; and more generally, that Skinner’s general approach to explanatory theory, models, ‘inner states’, and so on, be recognized for what it is, namely, a rejection of the general approach of the sciences and even of engineering practice. Herrnstein, like some of his colleagues, interprets this reaction as a ‘contemptuous dismissal’ of some scientific material. And, like others of his persuasion, he does not attempt to show some error of fact or reasoning, or to meet the simple challenge: Produce some nontrivial hypothesis on the ‘causation of behavior’ that has empirical support and has any demonstrable social consequences or any bearing on the positions that are presented. Perhaps this is enough, without belaboring the point further, to show that Herrnstein’s argument is indeed an intellectual shambles. But I was interested in showing that it is equally dismal from a moral point of view. Let us now turn to Herrnstein’s treatment of this matter. I observed that Herrnstein’s argument will surely be used by racists to justify discrimination and, more generally, by the privileged to justify their privilege. Herrnstein responds that he does not see why his argument ‘will have that unwholesome consequence’ (of being exploited by the racist) and pleads that he ‘was not proposing to study . . . a genetic flaw in any race’. This seems rather disingenuous. He begins his seven-paragraph discussion of IQ and race by noting that ‘it is the relation between heritability and racial differences that raises the hackles’ and then asks whether ‘the well-established, roughly fifteen-point black-white difference in I.Q. . . . arises in the environment or the genes’, concluding finally that ‘a neutral commentator . . . would have to say that the case is simply not settled’ and that the ‘fundamental issue’ is ‘whether inquiry shall (again) be shut off because someone thinks society is best left in ignorance’. I take it that Herrnstein does understand why his argument will be used to justify privilege in general - at least, he does not question this. With regard to the matter 5. Along with at least one improvement over the original. He now says that high rewards to those who serve ‘the wealthy few’ may harm society, whereas his original claim was that society was ‘husbanding its intellectual resources’ with the existing ‘gradient of occu-
pations’ which ‘is a natural measure of value and scarcity’ and makes ‘practical sense’. His remarks about ‘structural inadequacies’, a ‘consensus of social utility’, and so on, suggest that he does not see the true force of the point, however.
4 16
Noam Chomsky
of race, it is surely plain that a racist would be quite pleased with the statements just quoted. In a racist society, no one interprets Herrnstein as saying that ‘a neutral commentator would have to say that the case of White inferiority to Blacks is not settled’ (as indeed it is not). And surely there is a far more ‘fundamental issue’ than the one he identifies, namely that the scientist, like anyone else, is responsible for the foreseeable consequences of his acts. Suppose that even opening certain questions for inquiry is likely to have malicious social consequences (as in the example in question). Then the scientist has the clear moral responsibility to show that the importance of his inquiry outweighs these malicious consequences. It would take a degree of moral imbecility to fail to perceive even that there is a conflict of values. In the case in question, the conflict is easily resolved, if only because the possible correlation between partially heritable traits is of little scientific interest (in the present state of our understanding), and were someone interested in pursuing this matter as a scientist, he would certainly select traits more amenable to study than race and IQ. Furthermore there are no social consequences except under the assumptions of a racist society. What calls for explanation, then, is the zeal and intensity with which some investigators pursue scientifically trivial and socially malicious investigations. Herrnstein contests none of these observations, so I will elaborate them no further. To illustrate my point I offered two analogies, one real, one hypothetical. The real example is the racist anthropology of the 19th century, and the second, the case of a hypothetical psychologist in Hitler Germany who tried to show that Jews had a genetically-determined tendency towards usury, unperturbed by the fact that even opening this question for inquiry would provide ammunition for Goebbels and Rosenberg. Herrnstein attempts to ‘refute’ the second analogy, with reasoning even more defective than that already illustrated. Herrnstein observes that ‘America is not Hitler’s Germany’, that the hypothetical scientist should have left Germany, and that Goebbels and Rosenberg would have stopped the research. If I had given the hypothetical psychologist the name ‘Hans’, Herrnstein could, with equal logic, have argued that he should have changed his name to ‘Peter’. Obviously, the only relevant question is whether the analogy brings out the point at issue, namely, the conflict of values that should arise for a scientist who opens a question for inquiry with the knowledge that by doing so he may reinforce some of the most despicable features of his society. The point of the analogy, which Herrnstein evades, is that it would have been a poor excuse for the hypothetical psychologist (or the far-from-hypothetical racist anthropologist) to argue, with Herrnstein, that the question is not settled and the ‘fundamental issue’ is ‘whether inquiry shall (again) be shut off. . .’ The reason is the one I have already stated: The inescapable conflict of values, which Herrnstein never faces.
Comments on Herrnstein’s
response
417
In a final misrepresentation of the issue, Herrnstein states that contrary to what he takes to be my intention, he concludes that we should encourage ‘more research on people, not less’. But what I asserted was that ‘Of course, scientific curiosity should be encouraged (though fallacious argument and investigation of silly questions should not), but it is not an absolute value’; and where a conflict of values arises, it can not simply be swept under the rug. Herrnstein objects to what he perceives as a demand that he ‘subject scientific findings to the test of political suitability’. As I noted in ‘Psychology and ideology’, this is the typical maneuver of those who are attempting to evade the moral issue and to place the onus of ‘antiintellectualism’ on those who ask the reasonable question: How do you justify your research, given its likely social consequences? Herrnstein fails utterly to understand the simple, elementary point that the scientist is responsible for the foreseeable consequences of his acts and therefore must justify his pursuit of research with predictable malicious consequences (and in this case, no scientific merit). I had assumed that Herrnstein was simply thoughtless in overlooking this trivial point and misidentifying the ‘fundamental issue’ so grossly. That he seems so incapable of perceiving the point even when it is spelled out in detail seems to me most remarkable. Finally, a word on the question of why there is such interest in work so clearly lacking in substance. Here we can only speculate, and the natural question to raise is: What is the social import of the conclusions presented? In ‘Psychology and ideology’, I mentioned the social import of 19th century racist anthropology and pointed out that ‘the tendencies in our society that lead toward submission to authoritarian rule may prepare individuals for a doctrine that can be interpreted as justifying it’. In the case of Herrnstein’s presentation, apart from the fact that his observations with regard to the study of race and IQ will be welcomed by the racist (whatever the consequences of the investigation - which are sure to be uncertain, confused, and to leave open the possibility that . . .), the hidden premise that individuals will labor only for material gain and ‘differential reward’ beyond their fellows appears to give scientific credibility to prevailing ideology. And furthermore, the privileged will naturally delight in the conclusion that a hereditary meritocracy results as society wisely husbands its intellectual resources, that it may even be inevitable if society is to function properly. In short, the underprivileged and dispossessed must accept, even welcome, their fate, since things could not really be otherwise. Even the most benign social policy will only accentuate the problem, and they themselves gain by the fact that the social order is prejudiced against them. Hermstein’s argument thus is a contribution to what has elsewhere been called ‘the new assault on equality’.6 6. For discussion, see the May/June 1972 issue of Social Policy and the references cited
in footnotes
1 and 2.
418
Noam Chomsky
This is a particularly important matter in a period when one of the standard and most effective techniques of social control may diminish in effectiveness or disappear, namely, the belief that as the economy grows every individual share will grow so that it is irrational, for culturally-determined economic man, to urge revolution, reform, or redistribution, with all of their attendant uncertainties and dislocation. Walter Heller, Chairman of the Council of Economic Advisers under the liberal American administrations of the early 196Os, expressed the point this way: ‘When the cost of fulfilling a people’s aspirations can be met out of a growing horn of plenty - instead of robbing Peter to pay Paul - ideological roadblocks melt away, and consensus replaces conflict.’ 7 But the social costs of the ‘growing horn of plenty’ are now becoming apparent, as is the fact that laws of physics and chemistry impose certain limits on growth that may not be too far distant for the industrial societies, where the problem is most severe. Noting many uncertainties, still the problem cannot be simply dismissed, and it troubles economic and political elites as well as many scientists. But if the technique of social control based on the assumption of limitless growth loses its efficacy, some new device must be constructed to ensure that privilege is not threatened. What could serve better than the theory that the poor and weak must accept and welcome their status, which results from the wise decision to reward the talented in a ‘hereditary meritocracy? I note again, as in the original article, that questions of fact and questions of ideological commitment are in principle separable. If there were any merit to arguments such as those presented by Skinner and Hermstein, one could not reject them on the ground that their consequences are unpleasant. Since the arguments are entirely lacking in merit, however, we turn rather to the questions I have just raised, and I think that the speculation just offered is reasonable one.
7. New Dimerlsions of Political Economy, (1967), Cambridge Mass., p. 12, cited by Richard B. Du Boff and Edward S. Herman (August 1972) The New Economics: Hand-
maiden of inspired truth. The Revirr of Rudical Political Economics, 4 (4), an important discussion of economics and ideology.
Comments
on Professor
Layzer’s
‘Science
or superstition,
R. J. HERRNSTEIN Harvard University
Time and journal space are too dear to warrant adding significantly to what Professor Layzer has already consumed with his ‘Science or superstition’. I will briefly note a few of his errors, leaving it to those who would to ferret out the others. To his political tub-thumping, I have no response (nor, for that matter, any reaction). 1) On heritability in general. The heritability, narrowly defined, of traits showing dominance - like eye color - is less than 1.0, even though non-genetic factors may contribute no variance. The reason is that the narrow definition of heritability concerns intergenerational correlations, which may be quite small even for purely genetic traits, if the genetic factors interact non-additively. In contrast, the broad definition of heritability includes those non-additive genetic components in its calculation of the index. For some purposes, one needs one definition, for others, the other. It would, I agree, be a help if word usage in this field were standardized, but Professor Layzer has not helped at all by depicting these issues as if they were mere confusions or inconsistencies. The various formulas for estimating heritabilities are usually set forth with explanation. Such is certainly the case with Burt and Howard (1956), whom Professor Layzer archly but vacuously denigrates or Jinks and Fulker (1970) whom he omits. 2) On the heritability of IQ Professor Layzer asserts italicly: ‘The heritability of such scores [i.e., tests of reading] tells us nothing about the educability of the children being tested’ (p. 270). That, bluntly, is wrong. It tells us what fraction of the variance of such scores is genetic and, therefore, by subtraction from 1.0, what fraction can be attributed to environmental (i.e., non-genetic) sources. Practically speaking, if a group of children are subjected to various sorts of environments that might reasonably be expected to affect reading, but the heritability comes out high, we have learned a lot about educability. We have learned that our expectations about those environments in relation to the tested performance were false. Possibly, some other environments might make a difference. The index of heritability tells us, indeed, nothing about environments in which the subjects have
420
R. J. Herrnstein
not lived. Nevertheless, for those like Professor Layzer and me who would have guessed that the variety of common environments in our society must be effectively gigantic for the nurturing of cognitive performance, it comes as a jolt to find large heritabilities for intelligence test scores. Professor Layzer cushions himself against the jolt by invoking covariance. He points out that if genetic and environmental factors covary (i.e., are positively correlated), we might be tricked into attributing to the genetic source that which really belongs to both, thereby inflating it. From that sensible possibility, he rashly leaps to the conclusion (or seems to) that such covariance has, in fact, inflated heritability estimates for IQ. What he overlooks are the studies that bear directly on covariance of the sort he postulates. Interested readers should consult Burks (1928), Honzik (1970), Bouchard (in press), and, of course, Jensen (1969, 1970) if they want to know whether the separation of genetic from family-environment contributors to IQ materially reduces the natural-parent vs. child correlation. What they will find is that the correlation is reduced about as much as would be predicted for a trait whose heritability is in the vicinity of .8. 3) On deviation scales. Professor Layzer heaps scorn on the IQ because it is a measure of relative standing. (Have astronomers, which is what Professor Layzer is, given up stellar magnitudes?) What is wrong with relative standing? I have heard it argued persuasively (to me) that all measurement - physical or otherwise - is based on comparison. In any event, Professor Layzer apparently does not know that the estimation of heritability requires the use of deviations from population means in units of variation. Even if we measured intelligence the way Professor Layzer thinks we should (or, more accurately, the way he thinks he thinks we should, since any scale he comes up with will be a deviation scale against some reference), we would have to convert it into something like an IQ to estimate the heritability. 4) On the range of environments in twin studies. As Professor Layzer knows, the correlation between the IQs of identical twins raised in different homes bears some relation to its heritability. If the correlation is high, and if the homes for twin pairs are not environmentally similar, then the genetic factor looms large. This is not the place to review, once again, the findings in such studies (see Jensen, 1970; Vandenberg, 1971), except for one point that Professor Layzer focusses on. In the largest of these twin studies (Burt, 1966), the subjects were mostly London schoolchildren, from which Professor Layzer concludes that ‘the range of educational experience in this study was considerably narrower than it would have been for a representative sample of English schoolchildren’ (p. 291). Let us grant him his assertion. Presumably, then, the between-twin correlation - which reflects the genetic identity - is spuriously inflated because the environmental diversity in the
Comments on Professor Layzer’s science or superstition’
421
sample is restricted. But then, should not the total variation of IQ in the sample also be reduced? Apparently without noticing it, Professor Layzer has stumbled on a shrewd test of the environmental factor. It goes like this: a) The twins encounter a limited range of environments (spuriously enhancing the correlation of their IQs; b) Therefore, the total variation in IQ in the sample should be small compared to the population as a whole. On second thought, perhaps Professor Layzer did notice, for the test works against him. The standard deviation of IQ in Burt’s study was 14.8; the standard deviation in the English population as a whole is 15. Now, it behooves one to work back through the foregoing test to see where it leads. Incidentally, and finally, Professor Layzer mentions a number of times that Burt’s assessments were ‘subjective’ or ‘semi-objective’. A thoughtful reader would do well to see what Burt and his associates (1956, 1957, 1970) have to say on the subject. For now, I note simply that in Burt’s studies, the heritability for unadjusted, objective scores was about .77 (1958, 1970). His adjusted scores gave still higher estimates of heritability for interesting and plausible reasons, worth the attention of earnest scholars (but not to be found in Professor Layzer’s addition to this troubled field). REFERENCES Bouchard, T. J., Jr. (In press) Genetic factors in intelligence. In A. R. Kaplan (Ed.), Human behavior genetics. Springfield, C. C. Thomas. Burks, B. S. (1928) The relative influence of nature and nurture upon mental development; a comparative study of foster parent-foster child resemblance and true parent-true child resemblance. Yearbook of the National Society for the Study of Education, 27 (l), 219-316. Burt, C. (1958) The inheritance of mental ability. Am. Psychol., 13, 1-15. Burt, C. (1966) The genetic determination of differences in intelligence: A study of monozygotic twins reared together and apart. Err?. J. Psychol., 57, 137-153. Burt, C. (1970) The genetics of intelligence. In W. B. Dockrell (Ed.), On intelligence. London, Methuen. Burt. C. and Howard, M. (1956) The multifactorial theory of inheritance and its application to intelligence. Brit. 3. stat.
Psychol., 9, 95-131. Burt, C. and Howard, M. (1957) Heredity and intelligence: A reply to criticisms. &it. J. stat. Psychol., 10, 33-63. Honzik, M. P. (1957) Developmental studies of parent-child resemblance in intelligence. Child Devel., 28, 215-228. Jensen, A. R. (1969) How much can we boost I.Q. and scholastic achievement? Harvard educ. Rev.,
39, 1-123.
Jensen, A. R. (1970) I.Q.‘s of identical twins reared apart. Beh. Gen., 1, 133-146. Jinks, J. L. and Fulker, D. W. (1970) Comparison of the biometrical genetical, MAVA, and classical approaches to the analysis of human behavior. Psychol. Bull., 73, 311-349. Vandenberg, S. G. (1971) What do we know today about the inheritance of intelligence and how do we know it? In R. Cancro (Ed.), Intelligence: Genetic and environmental influences. New York, Gnme and Stratton.
A rejoinder
to Professor
Herrnstein’s
comments
DAVID LAYZER Harvard University
Professor Herrnstein has failed to address himself to any of my principal arguments: (1) That heritability theory does not apply to all metric characters but only to those for which environmental variations do not appreciably affect the character’s response to further environmental variations - a condition likely to be met only when the environmental variations are small or when the character is insensitive to them; (2) that because cognitive skills develop through complex and as yet little understood interactive processes involving many kinds of genetic and environmental factors, it would be unduly optimistic to expect such skills - even if they could be adequately measured - to fall into the rather narrow category of traits for which heritability is a meaningful concept; (3) that IQ tests are not measurements of cognitive skills, still less of cognitive ‘capacities’, that they are not in fact measurements at all, in any sense sanctioned by widely accepted scientific usage; (4) that although some of the published twin studies afford useful qualitative insights, heritability estimates derived from them are meaningless because (a) the statistical samples employed are strongly biased, (b) the relevant environmental factors have not been isolated much less measured, (c) IQ test scores are not measurements of a phenotype character, and (d) the statistical analyses in question are technically unsound; (5) that the conclusions drawn by Jensen, Herrnstein and others concerning (a) the educability of children with low IQs and (b) systematic differences between ethnic and socioeconomic groups have, therefore, no valid basis; (6) that the contention advanced by Jensen, Herrnstein, Eysenck, Shockley and others, that such statistical analyses constitute ‘scientific research into the genetic basis of differences in human intelligence’ rests on a mistaken conception of scientific research; (7) that the widespread tendency to take such ‘research’ seriously reflects (a) a belief that it is better to have measured something - never mind what or why than never to have measured at all, (b) a linked belief in the maxim ‘mind the mathematics and the meanings will mind themselves’, and (c) certain aspects of the social and political climate prevailing in the United States and Great Britain. Cognition
I (4), pp. 423426
424
David Layzer
Professor Herrnstein has evidently decided (perhaps rightly) that the best strategy available to him for dealing with the arguments offered in support of these points is to ignore them. Apart from some exercises in projection, mind-reading and innuendo, his critical remarks are directed largely at straw men. (i) Professor Herrnstein points out that writers on population genetics often find it convenient to distinguish between two kinds of heritability and adds that I have ‘not helped at all by depicting these issues as if they were mere confusions or inconsistencies.’ ‘These issues’ are not in fact mentioned in my article, which defines and uses one kind of heritability only. The definition I used is the one chiefly of interest in human genetics - in the rather limited circumstances where the concept applies at all - because it determines the sensitivity of phenotypic characters to environmental variation. The other kind of heritability (heritability in the narrow sense) is of special interest to animal geneticists because it determines the rate at which changes in a given character can be achieved through selective breeding. (ii) My discussion of environmental differences in the twin studies does not, as Professor Herrnstein says it does in his fourth numbered paragraph, ‘focus’ on the banal but surely noncontroversial remark he quotes. His criticism is directed not against my actual argument but against one I might have gone on to develop. He sketches a scenario in which ‘without noticing it, Professor Layzer . . . stumble[s] on a shrewd test of the environmental factor’ - which alas! turns out to be a booby trap. (Curses!) Leaving aside the question of why Professor Hermstein chose to create this edifying fantasy while many genuine arguments awaited his critical scalpel, let us take a closer look at his ‘shrewd test of the environmental factor’. Brief as it is, Professor Hermstein’s description of the test manages to incorporate one major unstated and demonstrably incorrect theoretical inference and two unstated and probably incorrect factual assumptions. The first of the unstated assumptions is that Burt standardized his tests and testing procedures (which included personal interviews with teachers and, when necessary, repeated individual retesting) on a representative sample of English school-children. If he did not do so - and there is no mention of such an enterprise in his paper - the fact that ‘the standard deviation of IQ in the English population as a whole is 15’ is a red herring; the standard deviation of IQ is 15 in any population, whether of English school-children, chimpanzees or earthworms. The second unstated assumption is that the only relevant environmental differences between school-children in London and elsewhere in England relate to schooling. The unstated and demonstrably false inference is that narrowing the range of educational experience must diminish the variance in IQ if it increases the between-twin correlation (assuming, for the sake of the argument, that all other relevant environmental factors are held constant). This
A rejoinder to Professor Herrnstein’s
comments
425
would be a valid inference if IQ differences within a population were composed additively of genetic and environmental contributions. But the main technical point of my article is that there is every reason to reject this assumption! Suppose that a school relies heavily on positive feedback, providing opportunity, incentive and reward in proportion to achievement. Such a school tends to amplify initial inequalities, whatever their provenance. The larger the amplification factor, the larger the between-twin correlation and the variance. Now consider two hypothetical schools, both employing positive feedback but having different value systems. For example, one school might reward mathematical ability, aggressive behavior and dark skin; the other, verbal ability, cooperative behavior and fair skin. The IQ means and variances of children attending the two schools might well be identical and the between-twin correlation for split MZ pairs in the same school would be high. But the between-pair correlation for pairs split between the two schools would presumably be much lower. This example is of course unrealistic (differences in school experience may not even be a major factor in cognitive inequality), but it serves to illustrate the main point I tried to make in my discussion of the twin studies: That, because relevant environmental differences between separated twins in these studies are statistically dissimilar to differences between randomly paired individuals drawn from the population at large, and because interaction effects are crucial, the measured between-twin correlations can have no quantitative significance. (iii) Professor Hermstein’s accusation that my article ‘denigrates’ the work of Burt and Howard, and the suggestion (in his final paragraph) that I have somehow misrepresented Burt’s views by describing his assessments of intelligence as subjective or semi-objective, merit attention here only insofar as they serve to divert attention from the fundamental inconsistency pointed out in my article between the Jensen-Hermstein view of IQ as an objective measure of intelligence and the views of Burt, perhaps the foremost psychometrician of his day, whose work forms the mainstay of Jensen’s and Herrnstein’s arguments. In my paper Burt speaks for himself, in a long quotation that includes the statement, ‘differences in this hypothetical ability [intelligence] cannot be directly measured’. Burt also believed, and acted on the belief, that subjective assessment is indispensable for arriving at valid and reliable indices of intelligence. Thus Burt’s conception of intelligence is the antithesis of the operational view expounded by Jensen and Herrnstein - a point that these authors have omitted to mention in their writings. (iv) I must confess that I can make no sense whatever of Professor Herrnstein’s remarks on ‘deviation scales’. It is true that measurement depends on comparison, but it is not true that every numerical assignment based on comparison constitutes a measurement in the sense that scientists use that term. In particular it is not true
426
David Layzer
in general that rank ordering is equivalent to measurement. Stellar magnitude is not, as Professor Herrnstein seems to think, a measure of relative standing. It is a logarithmic measure of emitted (or received) power in some designated band of frequencies. In principle one could define stellar magnitude in the way IQ is defined. That is. one could arrange the stars in a given population in order of increasing brightness and then arrive at a magnitude scale by forcing the stellar frequency distribution to approximate a standard error curve with a mean of 100 and a standard deviation of 15. Since observed frequency distributions of stellar magnitudes do not in fact resemble such curves. this procedure would invariably yield a definition of magnitude having no quantitative or physical significance. Tn my article I argued that, for somewhat similar reasons, the IQ scale is devoid of quantitative significance. Professor Herrnstein asserts that converting a set of measurements into a set of deviations from the mean, measured in units of the standard deviation of the set, automatically converts the measurements into ‘something like an IQ. This is patently absurd. What is distinctive and essential about the definition of IQ - and fatal to its scientific pretensions - is not that the zero-point and the unit of measurement are arbitrarily prescribed but that the form of the frequency distribution is arbitrarily prescribed. As I discussed at length in my article, we have no rational grounds for supposing that the frequency distribution. in any given population, of any scientifically valid measure of a specific cognitive skill bears any resemblance to a normal frequency distribution. Finally, T come to that brief portion of Professor Herrnstein’s response that seems to have some relevance to my arguments. Professor Hermstein rejects my assertion that the heritability of test scores tells us nothing about the educability of the children being tested. The high estimates of heritability obtained by Jensen and others teach us, he says, that our previously held views on the extent to which environmental differences affect copnitive performance must have been false. Tt does not seem to have occurred to Professor Herrnstein that there could be an alternative explanation for the discrepancy, namely. that our previous expectations. based on experience and common sense, were in fact valid. while the statistical results. based on defective mathematics, biology and psychology, are devoid of significance. That thesis is what my article seeks to establish. So far Professor Herrnstein has failed to confront it on a substantive level.
The IQ controversy:
ARTHUR
A reply to Layzer
R. JENSEN
University of California,
Berkeley
The publication of David Layzer’s ‘Science or superstition?’ in the previous issue of this journal brings to 120 the total number of articles and books provoked by my article in the Harvard education review (Jensen, 1969a). A bibliography of these items appears in my recent book Genetics and education (Jensen, 1972~) the Preface to which also provides a chronicle of my involvement in ‘the IQ controversy’. This book, together with my more recently published Educability and group differences (Jensen, 1973) actually give my detailed answers to practically all of the questions, criticisms, and issues raised in Layzer’s article. I therefore urge readers who wish to gain a greater understanding of these matters, and of my own position concerning their educational implications, to delve into these books and see for themselves just what I am saying - completely and in context. This is important because it is a common feature of so many of the criticisms of my position that they have had to misrepresent it and distort it, at times in ridiculous ways, in order to criticize it with the appearance of discrediting my main arguments. In this, Layzer’s critique is no exception. Whatever the acumen that Layzer, as a physical scientist, might be able to bring to bear in this field if he were not so politically or ideologically involved, his article makes it all too obvious that he is very one-sided in the exercise of his critical judgment. In discussing the rationale and findings of studies which point to the strong involvement of genetic factors in the distribution of mental ability, Layzer assumes the posture of an extreme methodological blue-nose. Yet he shows total suspension of his critical powers when dealing with studies which he perceives (at times wrongly) as lending support to his ‘environmentalist’ and ‘anti-hereditarian’ attitudes. I am not arguing with Layzer’s political and social egalitarianism. But, I am saying that genetic equality of human abilities is an altogether untenable belief in view of the evidence we already possess, as untenable as the geocentric theory in astronomy or the doctrine of special creation in biology. I find nothing in Layzer’s article that contradicts my main conclusions regarding the inheritance of mental Cognition I /4), pp.
427-452
428
Arthur
R. Jensen
ability,
and at times Layzer himself seems to acknowledge the importance of genetic that genetic factors can play an imfactors (e.g., ‘this result clearly indicates portant role in the development of cognitive skills’). Yet, for some reason, Layzer aims to denigrate or discredit evidence which attempts to achieve a greater precision in our knowledge of the relative influences of genetic and environmental factors (and their interaction) in the causation of human differences. The evidence we now possess leads me to the conclusion that in existing populations genetic factors significantly outweigh environmental influences in the distribution of human intelligence. The fact that the evidence is based on ‘IQ tests’ should not be construed to mean that such tests rank-order individuals much differently than if they were ordered in terms of parents’, teachers’, employers’ or the ‘man-in-the-street’s’ more subjective criteria for judging intelligence or ‘brains’. If the concept of intelligence or the IQ were merely psychological esoterica, we can be fairly sure there would be no ‘IQ controversy’.
Misrepresentations Like so many of my critics, Layzer falsely attributes non sequiturs and absurdities to me and then boldly attacks them. Is it necessary to distort what I have actually said in order to find fault with it? For example, Layzer writes, ‘such studies show, according to Jensen, that IQ differences are approximately 90 % genetic in origin’. I have never stated any figure as high as 90 % as an average. What in fact I have said in numerous articles is that in accounting for the causes of the differences among persons in IQ, genetic factors outweigh environmental influences by about 2 to 1. This is quite different from the 9 to 1 ratio implied by Layzer’s figure of 90 %. If the broad heritability of IQ is about 80 (as my review of the evidence in 1969 led me to conclude), then the proportion of genetic to nongenetic (or environmental) variance is in the ratio of 4 to 1. Since the variance is derived from squared differences, the relative contributions of genetic and environmental effects to the actual differences in IQ would be in the ratio of 1/4 to 1/l, or 2 to 1. Arguments as to whether the broad heritability (h2) of IQ is .80 or some other value is another matter. Since by the definitional nature of h2 there can be no one ‘true’ or constant value of h2 for intelligence (or any other metric trait), it is fatuous to argue whether h2 is .60 or .80 or .90 or some other value. But competent analyses of existing evidence find that most obtained values of broad h2 fall in the range from .60 to .90 for various mental tests and various populations. Layzer notes that I pointed out (Jensen, 1969b, p. 50) that the median correlations of .75 between the IQs of identical twins reared apart and .24 between IQs of unrelated children reared together are quite consistent with one another if the
The IQ controversy:
A reply to Layzer
429
heritability of intelligence is .75, since the correlation of .75 (an estimate of the genetic variance) plus the correlation of .24 (an estimate of environmental variance) totals .99, which comes very close to 1.00 or 100 % of the phenotypic variance. Actually, the theoretical discrepancy should be even more than .Ol, and probably closer to .lO, since the correlation between unrelated children reared together does not include the within-families component of variance and the observed correlation is most likely inflated slightly by some degree of genetic correlation due to selective placement by the adoption agencies. In any case, sampling error for these correlations would exceed .Ol. Yet Layzer claims that I attribute this left-over .Ol to genotype X environment interaction! I have never drawn any such unwarranted and absurd conclusion. It is a figment of Layzer’s, perhaps invented for the sake of criticism. Or perhaps he got it from Light and Smith (1969, p. 496) who also falsely attributed this absurdity to me and then proceeded to make it the keystone of an elaborate hypothetical analysis intended to demolish my incorrectly represented position. (The Light and Smith analysis, incidentally, proved fatally fallacious on other grounds as well; see Shockley, 1971a, 1971b; Light and Smith, 1971.) Layzer complains that I ignore sex differences and color differences as sources of IQ variance, when in fact I have written a major research paper on sex differences and their interaction with racial differences (Jensen, 1971) and I have also explicated methods for studying the contribution of the social aspects of skin color to racial IQ differences (Jensen, 1969b, p. 241; 1970, pp. 150-1.51; 1973, pp. 222227). Layzer quotes the British geneticist Waddington to the effect that ‘. . . if one takes some particular phenotypic character such as body weight or milk yield, one of the first steps in an analysis of its genetic basis should be to try to break down the underlying physiological systems into a number of more of less independent factors’. Layzer comments: ‘These views contrast sharply with those of Jensen . . .’ In reply, let me simply quote what I have written about the genetic analysis of mental abilities: ‘A heritability study may be regarded as a Geiger counter with which one scans the territory in order to find the spot one can most profitably begin to dig for ore. Characteristics with low heritability are less likely to yield pay dirt. The reason, of course, is that all we have to work with, at least at the beginning of our investigation, is variance, and if what we are interested in is genetical analysis, we would like to know that some substantial proportion of the trait variance we are concerned with is attributable to genetic factors. So we should not belittle heritability studies; but they should be regarded as only the beginning rather than as the goal of our efforts in genetical analysis . . . a test score usually represents an amalgam of a number of psychological processes in each of which there are imperfectly correlated and genetically conditioned individual differences. Thus our
430
Arthur
R. Jensert
aim should be to fractionate our ability measurements so as to get at smaller and more unitary components of ability. This is the province of the differential psychologist, but it requires also the methods of experimental psychology. Factor analysis alone is not the answer’ (Jensen, 1972b, p. 243). I go on to explain that factor analysis, as it has generally been used, has revealed common factors among tests which are already so complex as to permit nothing more than the fitting of a crude polygenic model, and 1 suggest greater experimental ‘fractionation’ of abilities in order to permit a more fine-grained and penetrating behavior-genetic analysis of them (Jensen, 1972b). Is this so very different from the views stated by Waddington? It is a fact that the heritability of a complex behavior or processes (like performance on an IQ test, or milk production in cows, or physical growth) can be determined without knowing anything about the underlying mechanisms. But as scientists, of course, we wish to attain an ever more complete understanding of the phenomenon. A substantial heritability index tells us that an understanding of individual differences in the trait in question must be sought in the organism’s internal, genetically conditioned biochemical and physiological processes as well as in the external environmental influences acting throughout its development. Misconceptions
about heritability
Part of Layzer’s criticisms of the heritability concept seems to be based on the disillusionment of his own assumptions as to what it means, or at least the meaning he thinks others give to it. But heritability was never intended to have some of these meanings, and I do not know any fully informed persons who have held the beliefs about heritability that Layzer so deplores. Heritability (hZ) is a technical term in genetics which refers to the proportion of the population variance in a phenotypic characteristic or measurement that is attributable to genetic variation. Narrow heritability includes only the additive part of the genetic variance, i.e., the part which ‘breeds true’ and largely accounts for the resemblance between parents and children. Psychologists are more interested in broad heritability, which includes all the genetic variation, the additive portion plus variance due to interaction between genes at the same loci on the chromosomes (called dominance) and interactions among genes at different loci (called epistasis). Broad heritability, which is what we are concerned with here, is sometimes referred to as the coejjicient of genetic determination to distinguish it from narrow heritability. Broad heritability can take any value from 0 to 1. It is not a constant, but differs for different traits, different measurements, and in different populations. Its value can be estimated by a number of methods in quantitative genetics. Like any population statistic, it is subject to measurement error and sampling error. Since it is based
The IQ controversy:
A reply to Layzer
431
essentially on the analysis of variance, it can tell us nothing at all about the causes of the particular value assumed by the grand mean of the population. It only analyzes the variance (or squared deviations) about the grand mean. And it tells us what proportion of this total variance is genetic variance and what proportion is nongenetic, i.e., due to environmental factors of all kinds, to interaction and covariance of genetic and nongenetic factors, and to errors of measurement. Most estimates of the broad heritability of IQ in the European and North American populations on which we have reasonably good data fall in the range from .60 to .90, and most of these estimates are in the range from .70 to .80. We could never determine heritability by studying a single individual, since heritability deals with differences among individuals. Each individual’s own development from the moment of conception is the inextricable product of his genes and environment; both are as necessary as are length and width in determining the area of a rectangle. But it is possible to analyze the phenotypic variance of a population sample into additive components of variance attributable to the additive effects of genetic factors and environmental factors, to their nonadditive effects (called G X E interaction) and to the covariance (or correlation) between the genetic and environmental factors. The quantitative methods for doing this are explicated in textbooks on quantitative and population genetics. The most recent and sophisticated application of these techniques to psychological test data has been made by the British geneticists Jinks and Fulker (1970). A major point in Layzer’s argument is that IQ scores, in principle, are unsuitable for this type of analysis, since they do not constitute an absolute scale (which is distinguished by a true zero point and equal intervals) like height and weight. An IQ is actually a normalized standard score, which indicates an individual’s standing, represented on a scale of deviates of the normal curve (multiplied by some arbitrary constant, such as 15 for the standard deviation of IQ), in relation to some reference population. In the case of the most widely used standardized tests, the reference population (or ‘norms’, as they are called) is a representative sample (usually nationwide) of individuals of the same age as the individual whose IQ we wish to determine. Thus there is no true zero point, and the IQ scale can be regarded as an interval scale only if we make the assumption that intelligence should have an approximately normal distribution in the population. Such measurements can meaningfully be subjected to the statistical techniques of analysis of variance, regression analysis, and correlation analysis - the principal tools of quantitative genetics. Since essentially all we are analyzing are the squared deviations from the mean, measurements on an absolute scale are quite unnecessary. (True, an absolute scale for intelligence would be a great advantage for some other purposes, such as studying the form of the growth curve of intelligence, but it is
432
Arthur R. Jensen
not a necessary condition for quantitative-genetic analysis.) Absolute scale characteristics such as height and weight are often converted to deviation scores within age groups, thus making their scale properties essentially the same as the IQ scale, prior to a genetic analysis of these physical traits. Some such transformation of the original measurements is needed to take care of sex and age differences when the genetic analysis involves correlations between parents and children, siblings, cousins, etc. One might argue that deviation measurements are permissible when the original measurements on which they are based is an absolute scale, as in the case of height and weight. Granted, the measurements may be more reliable, or more unidimensional, but those are quite different and separate issues. The fact is that the methods of quantitative genetics work as well for deviation scales as for absolute scales. If this were not so, it is doubtful that some of the world’s leading quantitative geneticists, such as Sewell Wright and John F. Jinks, would have undertaken genetical analyses of IQ scores. Textbooks on quantitative genetics do not limit the application of these methods only to characteristics that can be measured on an absolute scale. Many geneticists have written about the genetics of intelligence and found no fault with the scale properties of the IQ as regards its suitability for heritability analysis. Thus, it seems to me, a major pillar of Layzer’s critique simply crumbles. Experience with transformations of physical measurements to a scale equivalent to the IQ scale has shown me that when the sample size is reasonably large (i.e., 50 or more) and the measures are fairly continuous and unimodal, the difference between correlations (or results of analysis of variance) based on the original (absolute) measurements and on the deviation scores (equivalent to IQs) is practically nil. Now, if two or more studies are based on different tests in which the deviation scores are based on significantly different normative populations, you may average the components of variation (expressed as proportions of total variance) revealed by the genetical analysis, but what you cannot do is translate these proportions back into any scalar quantities. That is, the proportions of genetic and environmental variance can no longer be expressed in terms of number of IQ points difference (on average) between individuals attributable to genetic or to environmental factors (or to any other components). To do that, one would need to make the assumption that the tests and norms were the same in the two or more studies that had been averaged. Whether such an assumption is tenable is an empirical issue. But there is nothing wrong with averaging proportions of variance as a way of summarizing the central tendency of a number of studies, as long as it is realized that they are only proportions of variance and are not misinterpreted as scalar quantities.
The ZQ controversy:
A reply to Layzer
433
Heritability and teachability
The fact that IQ has high heritability surely does not mean that individuals cannot learn much. Even if learning ability had 100 % heritability it would not mean that individuals cannot learn, and therefore the demonstration of learning or the improvement of performance, with or without specific instruction or intervention by a teacher, says absolutely nothing about heritability. But knowing that learning ability has high heritability does tell us this: If a number of individuals are all given equal opportunity - the same background, the same conditions, and the same amount of time -for learning something, they will still differ from one another in their rates of learning and consequently in the amount they learn per unit of time spent in learning. That is the meaning of heritability. It does not say that individuals camrot learn or improve with instruction and practice. It says that given equal conditions, individuals will differ from one another, not because of differences in the external conditions but because of differences in the internal environment which is conditioned by genetic factors. ‘Teachability’ presumably means the ability to learn under conditions of instruction by a teacher. If this is the case, then it is true that heritability has nothing to do with teachability. But was this ever really the question? Has anyone questioned the fact that all school-children are teachable? The important question has concerned differences in teachability - differences both among individuals and among subgroups of the population. And with reference to the question of differences, the concept of heritability is indeed a relevant and empirically answerable question. We have heard it said that ‘teachability is not inversely related to heritability’. Such a statement simply ignores the central fact that heritability deals with differences. The degree to which equal conditions of teaching or instruction will diminish individual differences in achievement is inversely related to the heritability of the ‘teachability’ of the subject in question, and various school subjects probably differ considerably in heritability. The fact that scholastic achievement shows lower heritability than IQ means that more of the variance in scholastic achievement is attributable to nongenetic factors than is the case for IQ. Consequently, we can hypothesize what the sources of the environmental variance in scholastic achievement are, and possibly we can manipulate them. For example, it might be hypothesized that one source of environmental variance in reading achievement is whether or not the child’s parents read to him between the ages of 3 and 4, and we can obviously test this hypothesis experimentally. Much of the psychological research on the environmental correlates of scholastic achievement has been of this nature. The proportion of variance indicated by l-h*, if small, does in fact mean that the sources of environmental variance are skimpy under the conditions that prevailed in the population in which h* was estimated. It means that the already existing variations in environmental
434
Arthur
R. Jensen
(or instructional) conditions are not a potent source of phyotypic variance, so that making the best variations available to everyone will do relatively little to reduce individual differences. This is not to say that as yet undiscovered environmental manipulations or forms of intervention in the learning or developmental process cannot, in principle, markedly reduce individual differences in a trait which under ordinary conditions has very high heritability. By the same token, low heritability does not guarantee that most of the nongenetic sources of variance can be manipulated systematically. A multitude of uncontrollable, fortuitous microenvironmental events may constitute the largest source of phyotypic variance in some traits. The heritability of individual differences and of group differences in scholastic performance in the total population are therefore relevant if we are at all interested in the causes of these differences. To say that heritability is trivial or irrelevant is to say also that the complement of heritability, l-h*, or the proportion of variance attributable to nongenetic or environmental factors is also trivial. To dismiss the question of heritability is to dismiss concern with the causes of educational differences and their implications for educational practices. As I read it, what most educators, government officials, and writers in the popular press who discuss the present problems of education are in fact referring to is not primarily dissatisfaction with some absolute level of achievement, but rather with the large group differences in educational attainments that show up so conspicuously in our educational system the achievement gaps between the affluent and the poor, the lower-class and the middle-class, one race and another, the majority and the minority, the urban and the suburban, and so on. Educational differences, not absolute level of performance, are the main cause of concern. Whether we like to admit it or not, the problem of achievement differences today is where the action is, where the billions of dollars of educational funds are being poured in, where the heat is on, and where the schools are being torn apart. Are we not trying to understand more about the causes of these differences? It is mistaken to argue that heritability has no implications for the probable effects of environmental intervention. Since 1-h; (hz is h2 corrected for attenuation) is the proportion of trait variance attributable to nongenetic or environmental factors, the square root of this value times the SD of the ‘true score’ trait measurement gives the SD of the effect of existing environmental variations on the particular trait. For IQ this is about six points; that is to say, a shift of one SD in the sum total of whatever nongenetic influences contribute to environmental variance (i.e., l-h& will shift the IQ about six points. Thus, the magnitude of change in a trait effected by changing the allocation of the existing environmental sources of variance in that trait is logically related to its heritability. This applies, of course, only to existing sources of environmental variance in the population, which is all that can be estimated by 1-h:. It can
The IQ controversy:
A reply to Layzer
435
have no relavance to speculations about as yet nonexistent environmental influences or entirely new combinations of already existing environmental factors. With respect to IQ, I believe Carl Bereiter (1970) stated the situation quite correctly: ‘What a high heritability ratio implies, therefore, is that changes within the existing range of environmental conditions can have substantial effects on the mean level of IQ in the population but they are unlikely to have much effects on the spread of individual differences in IQ within that population. If one is concerned with relative standing of individuals within the population, the prospects for doing anything about this through existing educational means are thus not good. Even with a massive redistribution of environmental conditions, one would expect to find the lower quarter of the IQ distribution to be about as far removed from the upper quarter as before’ (p. 288). Genotype
X environment
interaction
Layzer makes much of the possibility of interaction of genetic and environmental factors. Interaction in this case means that the population variance of the phenotypic measurements is composed in whole or in some part of the nonadditive effects of genetic and environmental factors. The existing models of heritability analysis take such interaction into account and are capable of estimating the proportion of variance attributable to such interaction. With respect to IQ, the fact is that this interaction component is either nonexistent or so insignificant as to be undetectable in the existing data. If it were of substantial magnitude, it would easily show up with the present methods of analysis, which are quite capable of detecting other forms of interaction, such as dominance. In reading Layzer, one might easily get the impression that there is a lot of G X E interaction but that our models are unsuited to detecting it. Not so (see Jinks and Fulker, 1970). The fact that the genetic model for heritability is an additive model (as is all analysis of variance) does not mean that all of the components are forced into being either G or E; some of the components can be nonadditive functions of G and E. True, geneticists usually try to account for as much of the total variance as possible in terms of the strictly additive effects of G and E and will often make some scale transformation of their original measurements in order to minimize or eliminate the nonadditive components of variance. But this is unnecessary for IQ scores, which show little or no G X E interaction; the additive model fits IQ data about as well as it fits data on physical characteristics like height and weight. One of the impressive facts about genetical analyses of the IQ is how much it behaves like measures of continuous physical traits. Experimentally, psychologists have not discovered any teaching methods of environmental manipulations which interact with IQ in such a way as to wipe out differences in learning between individuals differing in IQ. The search for aptitude X
436
Arthur R. Jensen
training interactions, or AT1 for short, has become a popular area for research in educational psychology. What AT1 means, simply, is that no single instructional method is best for everyone, and that optimal performance will result only by matching a diversity of instructional methods with the diversity of individual’s aptitudes. If Bill and John are both taught by method A and Bill does much better than John, perhaps there is a different teaching method, B, that will permit John to learn as fast as Bill. That is the hope of AT1 researchers. The only trouble so far has been that when you find a method B which boosts John’s performance a little, it usually does so even more for Bill. Bracht (1970) recently reviewed a large number of studies in the ATI field which met certain methodological and statistical criteria to permit rigorous evaluation, and he found that out of 90 studies that were specifically designed to yield aptitude X treatment interactions of the kind that would solve the performance difference between Bill and John, only five yielded such an A X T interaction, and none of these aptitude differences was of the IQ variety - they were presonological variables unrelated to intelligence. Bracht says a number of interesting and important things: ‘When a variety of treatment stimuli, especially conditions not controlled by the experimenter, are able to influence performance on the dependent variable, it is unlikely that a personological variable can be found to produce a disordinal interaction with the alternative treatments . . . Success on a combination of heterogeneous treatment tasks is predicted best by measures of general ability [i.e., IQ tests], and the degree of prediction is about equally high for alternative treatments’ (p. 636). ‘The degree of task complexity may be a major factor in the occurrence of ATI. Although the treatment tasks for most of the 90 studies were classified as controlled, the treatments were generally relatively complex tasks. Conversely, four of the five experiments with disordinal interactions [ATI] were more similar to the basic learning tasks of the research laboratory’ (p. 637). ‘Despite the large number of comparative experiments with intelligence as a personological variable, no evidence was found to suggest that the IQ score and similar measures of general ability are useful variables for differentiating alternative treatments for subjects in a homogeneous age group. These measures correlate substantially with achievement in most schoolrelated tasks and hence are not likely to correlate differentially with performance in alternative treatments of complex achievement-oriented tasks’ (p. 638). Such findings give little cause for optimism in finding new teaching methods that will overcome the large scholastic achievement differences that are so evident in our schools and are highly related to IQ.
The ZQ controversy:
A reply to Layzer
437
Heritability and group differences
I have often been falsely accused of claiming that the high heritability of IQ inevitably means that the mean differences in IQ between social class groups and racial groups must be due to genetic factors. I have never made this incorrect inference. What I have said is this: While it is true, indeed axiomatic, that heritability within groups cannot establish heritability between group means, high within group heritability increases the a priori likelihood that the between groups heritability is greater than zero. In nature, characteristics that vary genetically among individuals within a population also generally vary genetically between different breeding populations of the same species. Among the genetically conditioned traits known to vary between major racial groups are body size and proportions, cranial size and cephalic index, pigmentation of the hair, skin, and eyes, hair form and distribution on the body, number of vertebrae, fingerprints, bone density, basic metabolic rate, sweating, fissural patterns on the chewing surface of the teeth, numerous blood groups, various chronic diseases, frequency of dizygotic (but not monozygotic) twinning, male/ female birth ratio, ability to taste phenylthiocarbomide, length of gestation period, and degree of physical maturity at birth (as indicated by degree of ossification of cartilage). In light of all these differences, Spuhler and Lindzey (1967) have remarked ‘. . . it seems to us surprising that one would accept present findings in regard to the existence of genetic anatomical, physiological, and epidemiological differences between the races . . . and still expect to find no meaningful differences in behavior between races’ (p. 413). The high within-groups heritability of certain behavioral traits, such as intelligence, adds weight to this statement by Spuhler and Lindzey. In fact, it is quite erroneous to say there is no relationship whatsoever between heritability within groups and heritability between group means. Jay Lush, a pioneer in quantitative genetics, has shown the formal relationship between these two heritabilities (1968, p. 312), and it has been recently introduced into the discussion of racial differences by another geneticist, John C. DeFries (1972). This formulation of the relationship between heritability between group means (hg) and heritability within groups (h$) is as follows : pcu
h2 w
where:
(1-r)
P
(l--j)<
hi is the heritability between group means. h$ is the average heritability within groups. r is the intraclass correlation among phenotypes within groups (or the square of the point biserial correlation between the quantized racial dichotomy and the trait measurement). p is the intraclass correlation among genotypes within groups, i.e., the within-group genetic correlation for the trait in question.
438
Arthur
R. Jensen
Since we do not know p, the formula is not presently of practical use in determining the heritability of mean group differences. But it does show that if, for a given trait, the genetic correlation among persons within groups is greater than zero, the between group heritability is a monotonically increasing function of within-groups heritability. And the probability that a phenotypic mean difference between two groups is in the same direction as the genotypic mean difference is greater than the probability that the phenotypic and genotypic differences are in opposite directions. I know no principles of genetics that would rule out the possibility of determining the heritability of differences between group means. If this question is unresolvable in the sense, say, that perpetual motion is impossible, Layzer should be able to spell out the laws of nature that make it so. To say it is possible in principle, however, is not to say it is easy in practice. The methods would have to differ from those used for determining the heritability of individual differences, just as the method for determining the temperature of a distant star must differ from that for measuring the temperature in the kitchen stove. The science of astronomy would never have advanced beyond star gazing if astronomers had applied as little imagination and ingenuity to solving problems in their field as Layzer seems to insist be applied in genetics. The storm of criticisms and ideological protests directed at me has been a result of my expressing serious doubts that the observed mean IQ difference between whites and blacks in the U.S., and between social class groups, is entirely explainable in terms of culture-bias in tests, unequal educational opportunities, social discrimination. and other environmental influences. My position is that there is now sufficient evidence seriously to question the 100 percent environmental theories of the whiteblack intelligence difference. Are there any responsible scientists today who claim that this position can be ruled out on the basis of evidence or ruled out n priori by any principle of genetics? How many scientists today express little or no doubt that all of the racial IQ difference is attributable to environment? And on what evidence do those who claim no doubt base their certainty? I have not found any 100 % environmental theory which can explain the facts or which stands up when its major premises are critically examined in the light of evidence. Therefore, I regard this issue scientifically as an open question which can be eventually answered in a scientific sense only if we are willing to consider all reasonable hypotheses. It is a reasonable and potentially testable hypothesis that genetic factors are involved in the average white-black IQ difference. My study of the research evidence bearing on this question at present leads me to believe that a preponderance of the evidence is more consistent with a genetic hypothesis, which, of course, does not exclude the influence of environment (Jensen, 1973).
The IQ controversy: A reply to Layzer
439
Means versus medians Layzer complains about my use of the median (i.e., the middle value in a distribution) for summarizing the correlations obtained in numerous studies of various kinship groups (Jensen, 1969a, Table 2, p. 49). I had noted that these median correlations between IQ8 for various degrees of kinship come very close to the values one should expect from a polygenic model of inheritance, and they are the basis for the conclusion that genetic factors predominate as a cause of IQ differences in the populations in which these kinship correlations were obtained. Layzer believes that the median is not the proper statistic for indicating the central tendency of a number of correlations obtained in various studies. He suggests that instead of the median one should use the mean of the various obtained correlations - a weighted mean, with each correlation entering into it weighted inversely to its probable error. This is a correct and standard way for combining statistics, and in general I agree with it. Weighting the correlations by their standard error, SE, (or the probable error, which is .67 SE) surely makes sense, since the SE indicates the precision or reliability of the sample estimates of the population value, and of course we would want to give more weight to the more reliable values. On the other hand, an argument can be made in this particular case for using the median instead of a weighted mean. The median, of course, is least affected by extreme or atypical values. Since the kinship correlations reported in the literature are based on a variety of tests, some of which are scholastic achievement tests or tests of very narrow and special abilities rather than tests of general intelligence, a decision has to be made concerning which tests to include in the collection of correlations of which we wish to represent the central tendency. Errors of judgment on this point would little affect the median but could markedly affect the weighted mean, particularly if the correlation for an atypical test or population were based on a very large size sample. As an example, one of the largest sets of twin data ever collected consists of a nationwide sample of monozygotic (MZ) and dizygotic (DZ) twins who as high school seniors entered the National Merit Scholarship competition and took the set of scholastic achievement tests which serve as part of the basis, along with high school grades, for picking the winners. In the first place, tests of scholastic achievement generally show much higher correlations between any children reared together (even when they are genetically unrelated) than do intelligence tests, and the difference between MZ and DZ twin correlations is much less for scholastic tests than for IQ tests. This difference in the case of the National Merit Scholarship data is further diminished by the fact that poor students do not enter the competition, and since DZ twins are less likely to be alike than MZ twins, there will be more instances where only one member of a DZ twin pair will get into the National Merit Scholarship screening process than in the case
440
Arthur R. Jensen
of MZ twins. This differential selection bias in the MZ and DZ twin samples makes the twin correlations (and consequently any estimates of heritability derived from them) atypical. The probable error, however, is smaller than for any other study, so a weighted mean including the National Merit Scholarship correlations could be quite distorted. The median would be much less distorted. Also when it comes to weighting the various entries, one could make a case for weighting correlations in terms of more than just their probable error. Why not weight the correlations or heritabilities derived from any given study in terms of the degree to which the particular test used in the study is loaded with g, that is, the extent to which the test, when factor analyzed with other tests of intelligence, shares common variance with the other tests? Tests that have less in common with what we are calling intelligence (viz., the g or general factor common to all complex mental tests) would be given less weight in the composite weighted mean. Why not assign weights according to the representativeness of the sample? Should a heritability estimate based on college graduates be given as much weight as one based on a representative sample of elementary school children? Should we weight in terms of the degree to which the various sample means and variances approach the population values for the tests used in the various studies? How about differentially weighting studies that differ in the degree to which they meet certain assumptions that underlie the methods for estimating heritability, such as equality of the total variances in both the MZ and DZ twin samples? We can see that there can be many other criteria for weighting besides just the probable error of the obtained correlations. I do not advocate such elaborate weighting, because I believe it can introduce too much subjectivity and, since many of the weights themselves are subject to error, would tend to lower our confidence in the composite. (Weighting by the SE alone, however, does not have this drawback.) All things considered, therefore, I feel that with these data there is apt to be less risk of distortion in the median than for any other measure. Anyway, it should be interesting to see how much difference it would make if we used weighted means instead of medians. Layzer’s readers may have been led to believe that the weighted means would give a quite different picture from that provided by the medians. I have obtained both medians and weighted means of all the reported kinship correlations that I can find in my reprint files which are based on some kind of general intelligence test.’ (I have excluded purely scholastic achievement tests.) The individual correlations were weighted by their standard errors. (In accord with the standard statistical procedure for averaging correlations coefficients,
1. It would take up too much space to list here all of the published sources of these kinship correlations. The writer will provide
the list of references, anyone requesting it.
keyed to Table
1, to
3282
1181
385
6
6
9
__
I
_
-
-
-~
“~
_
_
.25
.21
.47
.87 .74 55 .53
Median correlation
.32
.23
.47
.87 .76 .55 .55
-.
Weighted* mean correlation --
.174
-
I
^
-
,_
-
.
-
^.
I
-
__
_
.lOO
,076
.088 .lOl .075 .080
correlations
SD of N
the transformed correlation, i.e., 1/N-3; the weighted mean Z then was transformed back to the correlation coefficient, r. ** From Burt (1966).
1626 122 1384 1935
19 9 19 7
Number (N) of correlations
* The separate correlation coefficients, r, entering into each mean were transformed by Fisher’s Z transformation and were then weighted inversely by the standard error of
Monozygotic twins Reared together Reared apart Dyzgotic twins reared together Full siblings reared together Parents and children reared by their natural parents Adopting parents and adopted children Unrelated children reared together
Degree of relatedness
Number of subject pairs _
-~-
>-
.051
.029
,017
.024 .091 .026 .022
VIII
-.07
.96 .94 47 SO
.24
.93 .88 .59 .57
Correlations for SE of mean physical measures** correlation Height Weight -.-
Table 1. Comparison of median correlation and weighted mean correlation from various kinship studies
442
Arthur R. Jensen
r, these were first transformed
to Fisher’s Z, then weighted by the inverse of the SE, then averaged, and finally, transformed from Z back to r.) Table 1 shows the results. We see that the medians and weighted means are quite similar (in fact, they correlate .995). Also shown is the standard deviation (SD) of the correlations from the various studies. If all studies represented samples from the same population (of persons and tests), we should expect the SD of the obtained correlations to be close to the standard error of the mean correlation. The fact that the SD is slightly larger than the SE indicates that the correlations obtained in the various studies are more variable than we should expect if all had used the same test and had sampled from the same population.2 The last two columns in Table 1 show the correlations for physical measures. It can be seen that they follow the same pattern as the correlations for the mental tests. The fit of the data to the values expected according to a simple polygenic model is remarkably close. It suggests that the heritabilities of measures of intelligence and of body weight are very similar and consistent with a broad heritability of between .70 and .SO. I do not know of any strictly environmental theories that can explain this pattern of correlations as well as does the polygenic model of inheritance. These correlations substantiate the conclusion of a greater genetic than environmental determination of individual differences in IQ. No geneticist who has studied such evidence has, to my knowledge, drawn the opposite conclusion, and I have searched all the up-to-date textbooks of genetics that deal with this subject. In regard to this evidence, Layzer himself writes, ‘these findings show that IQ is strongly influenced by both genetic and environmental factors’. Though the genetic factors are in fact predominant, even if they were not, they eventually would become so as we achieved more and more equality of environmental conditions, cultural and educational opportunities, and the like, thereby reducing the environmental variance. Twin differences and environmental differences
Layzer points out that the IQ differences between MZ twins reared apart show a fairly substantial correlation with ratings of the amount of difference in the environments in which they grew up. This point seems to be made with the idea that it somehow contradicts the high heritability of IQ as indicated by the high correlation 2. Since the SE of the weighted mean correlations shown in Table 1 is the SE for the total number of kinship pairs, the SE for any one of the studies entering into the weighted mean would of course be considerably larger. A rough approximation to the average SE
for single studies would be given by the value of SE in Table 1 multiplied by VN, where N is the number of correlations. These values of SE/j/N, interestingly, differ but little from the SD of the N correlations.
The IQ controversy: A reply to Lnyzer
443
between MZ twins reared apart. Since both members of a pair of MZ twins possess exactly the same complement of genes, any difference between them must of necessity be due to nongenetic causes. Thus it should not be at all surprising that the magnitude of the difference between their IQs is correlated with differences in the environmental conditions to which they were exposed. This fact in no way alters the fact that the nongenetic variance is quite small (about .20 to .30 of the total IQ variance). Moreover, by no means all of this nongenetic variance is attributable to what we ordinarily think of as ‘environmental.’ I have written in detail on this point (Jensen, 1972). Had Layzer carefully read my article, I doubt that he would have used the correlation between MZ twins’ IQ differences and their environmental differences as part of his argument because this evidence actually weakens the case for the importance of social-psychological factors as a cause of IQ differences. Information processing versus ZQ Layzer would prefer a measure of information processing capacity instead of the traditional IQ. Quantity of information can be measured on an absolute scale in terms of bits. (A bit [for binary digit] is the amount of information necessary to resolve two equally probable alternatives; it is equivalent to the minimum number of binary questions [answerable with Yes or No] needed to reduce uncertainty to zero. The number of bits is the logarithm, to the base 2, of the number of alternatives.) Such absolute measures have certain decided advantages in scientific research. Psychologists surely would welcome an instrument that measured a person’s information processing on an absolute scale. Intelligence tests that involve problem solving and judgment are most likely measures of information processing capacity. The only trouble is that the items or problems that comprise such tests are so complex that we have no way, at least at present, of directly quantifying their informational content. The item difficulty of, say, Raven’s Progressive Matrices (a nonverbal reasoning test) is probably highly correlated with the number of bits of information contained in the items. If we could determine the bits for every Raven item, it would be a boon to research in differential and developmental psychology. But would it change any of our main conclusions about the heritability of individual differences in g (which the Raven test largely measures)? I doubt it. I believe that information processing capacity is the essence of g, the general intelligence factor. Layzer points to Piaget’s conceptions of mental development and intelligence as being consistent with his idea of information processing. I agree. But let it be noted that Piaget has devised various special tests with which to study this information processing capacity, and when these Piagetian tests are given to large samples of children and are factor analyzed along with conventional tests of intelligence (e.g.,
444
Arthur
R. Jensen
the Stanford-Binet, the Wechsler tests, Raven’s matrices, Kohs block designs, etc.), the Piagetian tests show high correlations with the other tests and are most highly loaded on the g factor; they tap little if any other source of variance not found in the conventional tests (Vernon, 1956). Moreover, the Piagetian tests show about the same magnitude of average differences between social classes and racial groups in California school children as are found with conventional IQ tests (Tuddenham, 1970). When laboratory techniques have been specially devised to permit the actual measurement of information processing capacity in terms of time per bits, as in highly precise measures of visual information processing and of choice reaction time to differing amounts of information, quite striking social class and Negro-white differences have been found in the expected direction (Bosco, 1970; Noble, 1969; studies reviewed by Jensen, 3973, pp. 322-329). The current denigration of the standard intelligence tests is a part of the attempt to minimize the significance of the evidence for a substantial genetic component in the variance on such tests; the scores on these tests are known to be correlated with educationally, occupationally, and socially significant criteria to about the same degree in different racial groups in the U.S. Contrary to the popular mythology in this field, it is very difficult to find any objective evidence of culture bias that could account for social class and racial differences in performance on current standard tests of intelligence, even those, like the Peabody Picture Vocabulary Test (PPVT), which give the appearance of being highly culture-loaded. They may be culture-loaded, but there is no evidence we have been able to find that the culture-loading is what differentially affects the performance of Negro and white children. Difference in mean score cannot be a criterion of culture bias. One must seek other evidence. We have examined several types of evidence of culture-bias in the PPVT and Raven’s Progressive Matrices. These studies have involved very large samples of Negro and white children in several California school districts. We find that the rank order of the percent, p, passing each item is virtually the same for Negroes and whites. The correlations between the p values of Negroes and whites on these tests are all above .95, averaging .97. In this respect, the two racial groups are more alike than are boys and girls within each race. In other words, the cultural biases in the test are more apparent with respect to sex differences than with respect to race differences. (The sexes do not differ appreciably in mean score, however, while the racial groups differ about one standard deviation, or 15 IQ points, on the average.) The matrix of item intercorrelations and the factor structure of these tests is not significantly different for white and Negro samples when these are roughly matched
The IQ controversy: A reply to Layzer
445
for mental age or total score. These properties of the data, for example, do not in the least distinguish between 4th grade white children and 6th grade Negro children. Yet they distinguish between 5th grade and 6th grade Negro children, and between 5th grade and 6th grade white children. A culture-bias hypothesis would predict greater Negro-white differences than adjacent grade differences in item intercorrelations. The findings, on the other hand, are more consistent with a developmental lag hypothesis. In multiple-choice tests, such as the PPVT and Raven, there is no systematic or significant racial difference in the choice of distracters on those items that are answered ‘wrong’. A special scoring key was made up so as to score as correct whatever response is given by the largest number of children in the Negro sample. When the tests are scored by this key, the Negro sample still averages lower than the white sample. Scales based on subgroups of items which discriminate either least between Negroes and whites or discriminate most are correlated with each other over .90 (approximately the reliability of the test), showing that the two types of items are measuring the same ability. The intelligence tests show essentially the same size of correlation with scholastic achievement in Negro and white samples. When scholastic achievement is ‘predicted by a multiple regression equation comprised of several intelligence tests, adding race (white vs. Negro) to the multiple prediction equation does not increase the multiple correlation with scholastic achievement. The predictive validity of the IQ test is the same for Negroes and whites. Negroes and whites with the same IQ perform equally well in school. In short, none of our analyses reveals any racial differences other than the number of items gotten right. There seems to be no good reason to believe that these tests behave any differently for Negroes than for whites. The sibling correlations on 16 ability tests were examined in large Negro and white samples. They are very similar, as indicated by a correlation of .71 between the sibling correlations on each test for Negroes and whites. The average difference between siblings on each test does not differ significantly for Negroes and whites. When estimates of the heritability (i.e., the proportion of genetic variance in test scores) of the various tests are correlated with the magnitude of the mean whiteNegro difference on the tests, the correlation is positive (.80 for whites, .61 for Negroes). In other words, those tests which are least sensitive to environmental influences (i.e., high heritability) in general show the largest white-Negro differences, and those tests which are most sensitive to environmental influences (i.e., low he& tability) show the smallest Negro-white differences. This outcome is just the opposite of what one would expect from a culture-bias or environmental hypothesis of the
446
Arthur R. Jensen
cause of the racial difference. This study has been repeated by other investigators using a different set of tests, and the results are essentially the same, i.e., a strong positive correlation between tests’ heritability and the magnitude of the white-Negro difference (for details see Jensen, 1973). Those who claim culture bias in current widely used tests, it seems to me, are obligated to produce some objective evidence that such bias in fact exists. I have found no evidence that it does, at least in the well-known tests we have studied. Misinterpretation
of Skodak and Skeels
Layzer cites the famous study by Skodak and Skeels (1949) as if it contradicted my position regarding the heritability of IQ. For readers who might be misled into believing that the findings of this study are inconsistent with a genetic theory of intelligence and with the evidence on heritability, a brief review of it is in order. Layzer’s use of the Skodak and Skeels study is typical. The study is often held up by ‘environmentalists’ as an example of evidence which supposedly contradicts the high heritability of intelligence. The fact that the adopted children turned out to have considerably higher IQs than their biological mothers is thought to constitute a disproof of the conclusion from many heritability studies that genetic factors are more important than environmental factors (in the ratio of about 2 to 1) in the causation of individual differences in IQ. If about 80 percent of the IQ variance is attributable to genetic factors, the 20 percent of the variance due to environmental differences can be thought of as a more or less normal distribution of all the effects of environment on IQ, including prenatal and postnatal influences. This distribution of environmental effects would have a standard deviation of about 7 IQ points, since the total variance of IQ in the population is 15’ = 225 and the 20 percent of this which is attributable to environment is .20 (225) = 45, the square root of which gives SD = 6.71. Is there anything in the Skodak and Skeels data that would contradict this conclusion? Skodak and Skeels based their study on 100 children born to mothers with rather low IQs (a range from 53 to 128, with a mean of 85.7, SD of 15.8). The children were adopted into what Skodak and Skeels described as exceptionally good, upper-middle class families selected by the adoption agency for their superior qualities. Of the 100 true mothers, 63 were given the 1916 form of the Stanford-Binet IQ test at the time of the adoption. Their children, who had been reared in adoptive homes, were given the same test as adolescents. The correlation between the mothers’ and children’s IQs was .38. Layzer notes that the IQs of the adopted children average about 20 points higher than the IQs of their true mothers. However, the difference between the mothers’ and children’s TQs is not really the re!evant question. It is on this point that the interpretation of this study has so often
The IQ controversy:
A reply to Lnyzer
447
misleading. What we really want to know is, how much do the children differ from the IQs we would predict from a genetic model?3 Using a simple model (provided by Crow, 1971, p. 157) which assumes that the children represent a random selection of the offspring of mothers with a mean IQ of 85.7 and that the children are reared in a random sample of homes in the general population, the children’s average predicted IQ should be 95. In fact, however, their average IQ turns out to be 106, or 11 points higher than the predicted IQ. If 20 percent of the IQ variance is environmental, and if one standard deviation of environmental influence is equivalent to about 7 IQ points, then it might be said the Skodak and Skeels children were reared in environments which averaged 11/7ths or about 1.6 standard deviations above the average environment of randomly selected families in the population. This would be about what one should expect if the adoption agency placed children only in homes they judged to be about one standard deviation or more above the average of the general population in the desirability of the environment they could provide. From what Skodak and Skeels say in their description of the adoptive families, they were at least one standard deviation above the general average in socioeconomic status and were probably even higher in other qualities deemed desirable in adoptive parents. So an eleven-point IQ gain over the average environment falls well within what we should expect, even if environmental factors contribute only 20 % of the IQ variance. In other words, this 11 points is well within the reaction range of phyotypic IQ, given a broad heritability of .80. But this 11 IQ points of apparent gain is more likely to be an overestimate to some extent, since these children, it should be remembered, were selected by the agency as suitable for adoption. They were not a random selection of children born to low IQ mothers. Many such children are never put out for adoption. (Most of the children were illegitimate, and as indicated in Leahy’s, 1935, study, illegitimate children who become adopted have a higher average IQ than illegitimate children in general or than legitimate children placed for adoption.) Even so, it is interesting that Skodak and Skeels found that the 11 adopted children whose true mothers had IQs below 70 averaged 25 points lower than the 8 adopted children whose true mothers had IQs above 105. There are also certain technical, methodological deficiencies of the been
3. This genetic prediction is sometimes made incorrectly by basing it on all 100 children, while actually we can make a prediction only for the 63 children whose true mothers’ IQs were known. The model assumes (a) test reliability of .90; (b) an ‘age attenuation’ of .95 (due to the fact that the mothers and children are widely separated in age and the correlation between the IQs of the same per-
sons tested that many years apart is .95 after correction for immediate test-retest unreliability); (c) narrow heritability of .71 (the estimate of Jinks and Fulker, 1970, p. 342); the narrow heritability is used when predicting offsprings’ values from parents’ values; (d) random mating (since the mothers were unmarried and nothing is known about the IQs of the true fathers).
448
Arthur R. Jensen
Skodak and Skeels study which make the basic data questionable; these deficiencies were trenchantly pointed out many years ago in critiques by Terman (1940, pp. 462467) and McNemar (1940). But the Skodak and Skeels study, such as it is, can be seen to be not at all inconsistent with a heritability of 80 for intelligence. To assume that the same 11-point IQ gain over the predicted value would have occurred if the biological mothers had been Negro instead of being white (but with exactly the same IQs) would be an unwarranted inference. It is unwarranted because according to the genetic model or prediction equation the Negro children would regress toward the Negro population mean IQ of about 85, rather than toward the white mean IQ of 100. Thus the predicted IQ gain of the adopted Negro children under the same environmental conditions would be some 10 IQ points less than the 11 IQ points gain for white children. If the Negro-white population difference in IQ is largely genetic, then a genetic model with dominance will predict regression of individual IQs to different population means for Negro and white children. The truth or falsity of this prediction is what we would like to know. The study of crossracial adoptions might help to elucidate the matter. Since cross-racial adoptions are not hard to find, it is interesting that the environmentalists who go on citing the rather weak Skodak and Skeels study have never investigated similar data based on Negro children. It would be considerably more relevant. A one-sided critical stance As I noted earlier, Layzer suspends his critical judgment when citing those studies which he apparently believes support his position. There are many examples of this in his article, but at least three instances merit some comment, since they involve relatively recent publications which readers may not have had the chance to evaluate for themselves. Layzer, being as methodologically puritan as possible in judging the evidence for the heritability of IQ, states ‘. . . measurements unaccompanied by error estimates have no scientific value’. If he had determined the error estimates of the data so which he was referring, he would have found the inferences based thereon to be highly significant (e.g., Jensen, 1967; 1972, pp. 294-306; Jinks and Fulker, 1970). On the other hand, all the points Layzer refers to in Starr-Salapatek’s study were presented by Starr-Salapatek (1971) without any error estimates or tests of statistical significance. Furthermore, when the proper error estimates are made, it turns out that all the ‘evidence’ in Starr-Salapatek’s study regarding the comparative heritabilities of intelligence test scores in white and Negro samples, and in lower- and middle-class groups, is completely lacking in significance. The study has been subjected to a detailed examination by two leading quantitative geneticists (Eaves and
The IQ controversy: A reply to Layzer
449
1972). Here is what they conclude about this study: ‘On purely theoretical grounds, therefore, we suggest that this particular experimental design, with the small samples available, could not be expected to lead to the conclusions which were drawn and indeed could only be drawn from it by omitting proper tests of significance.’ So large are the standard errors in Starr-Salapatek’s study that, as Eaves and Jinks point out, ‘. . . the data cannot even support the well-established conclusion that there is a genetical component of individual differences in intelligence’. Also, ‘. . . there is no evidence that the size of any heritable component depends on race or social advantage. This finding contradicts the main conclusion of Dr. ScarrSalapatek’s analysis which is based on a comparison of the numerical values of the correlations’. Finally, ‘there is certainly no evidence in Starr-Salapatek’s studies that the proportion of genetical variation in either verbal or nonverbal IQ depends on race or social class’. Layzer refers to Heber’s Milwaukee Project as an example of the extreme plasticity of intelligence. He quotes Scar-r-Salapatek’s description of it, to the effect that IQs of ghetto children born to especially low-IQ mothers showed an enormous IQ gain of 37 points over a control group as a result of environmental intervention. It is unlikely that Layzer has critically examined this study, for there have been no published reports of it except for stories in the popular press, and the authors of the study have apparently not been willing to make technical reports of it available to other researchers in this field who have requested it, myself included.4 However, Professor Ellis B. Page, an expert in psychometrics and research methodology, managed to secure a detailed report of this study from the agency which funded it (and which no longer makes the report available). Page has subjected this report of the Milwaukee Study to detailed scrutiny (1972) and his findings should be of interest to anyone who, with Layzer, claims to insist upon methodological purity. Page’s critique certainly leaves one with a markedly different impression of the study, and with a much greater skepticism, than is prompted by the sensationally optimistic reports appearing in the popular press. Page concludes: ‘The Milwaukee Project, then, is here viewed as deficient on three counts: biased selection of treatment groups, contamination of criterion tests; and failure to specify the treatments. Any one of these would largely invalidate a study. Together, they destroy it.’ Layzer uncritically refers to an ‘incisive critique of Jensen’s [HER] article’ by Deutsch (1969). In this ‘critique’ Deutsch claimed that some 17 errors were turned up in a casual perusal of my article (p. 524) and elsewhere he claimed that my article contained ‘. . . fifty-three major errors or misinterpretations, all of them uniJinks,
4. Since this article went to press, I have, received a copy of the report from Dr. Heber’s
office,
some
nine
months
after
requesting
it.
450
Arthur R. Jensen
dimensional and all of them anti-black’. This claim is baseless and defamatory. It took 22 months of repeated prodding by the American Psychological Association’s Committee on Scientific and Professional Ethics and Conduct to extract an itemized list of these 53 purported ‘errors’ from Deutsch. In view of all the efforts by ideological environmentalists to discredit my HER article, one wonders why Deutsch’s list of 53 ‘major errors’ has not gotten beyond the Ethics Committee and found its way into print. Considering the extreme pressure Deutsch was under from the Ethics Committee either to make a retraction of his defamatory claim or to produce a list of the ‘53 errors’, it is most instructive, and I might add most flattering to my HER article, to see what Deutsch’s list of ‘53 errors’ actually consists of. It utterly fails to support his claim. I wish it were published, but since it is not, I will gladly send it to all who request it. Readers can judge for themselves the quality of Deutsch’s ‘incisive critique’, to use Layzer’s words. One could go on noting other deficiencies in Layzer’s critique, but many of his points are long since discredited arguments that would be apparent to most readers familiar with this literature; most of the issues are treated in more general terms in my other writings (see References). As to Layzer’s ideological-political brand of environmentalism, I will make no comment here. My own position concerning the broader educational, societal, and ethical aspects of the genetical study of human differences has been amply expressed in numerous other articles (for a complete bibliography, see Jensen, 1972c, pp. 365-369). The overwhelming fact is that the scientific world no longer presents a consensus of environmentalism to the public,s and articles such as Layzer’s will do nothing to restore the appearance of consensus which Layzer and his likes are so disturbed to see undone.
5. See the resolution
on behavior
and
gene-
tics in the Amer.
Psychol.,
1972,27,
660-661.
The ZQ controversy:
451
A reply to Layzer
REFERENCES Bereiter;C.(l970) .Genetics.and.educability: Educational implications of the Jensen debate. In J. Hellmuth (Ed.), Disadvantaged child. Vol. 3. Compensatory education: A national debate. New York, Brunner-Mazel. Pp. 279-299. Bosco, J. J. (1970) Social class and the processing of visual information. Final Report, Project No. 9-E-041, Contract No. OEG-5-9-325041-0034(010). office of Education, U.S. Dept. of Health, Education and Welfare. Bracht, G. H. (1970) Experimental factors related to aptitude-treatment interactions. Rev. educ. Res., 40, 627-645. Burt, C. (1966) The genetic determination of differences in intelligence: A study of monozygotic twins reared together and apart. Brit. J. Psychol., 57, 137-53. Crow, J. F. (1970) Do genetic factors contribute to poverty? In Allen, V. L. (Ed.), Psychological factors in poverty. Chicago, Markham. Pp. 147-160. DeFries, J. C. (1972) Quantitative aspects of genetics and environment in the determination of behavior. In Lee Ehrman, G. S. Omenn, and E. W. Cospari (Eds.), Genetics, environment and behavior: Zmplications for educational policy. New
York, Academic Press. Pp. 5-16. Deutsch, M. (1969) Happenings on the way back to the forum: Social science, IQ, and race differences revisited. Harv. educ. Rev.,
39, 523-557.
Eaves, L. J., and Jinks, J. L. (1972) Insignificance for differences in heritability of IQ between races and social classes. Nature,
240,
84-87.
Jensen, A. R. (1967) Estimation of the limits of heritability of traits by comparison of monzygotic and dizygotic twins. Proceedings of National Science, 58, 149-157.
Academy
of
Jensen, A. R. (1969a) How much can we boost IQ and scholastic achievement? Harv. educ. Rev., 39, 1-123.
Jensen, A. R. (1969b) Reducing the heredityenvironment uncertainty, Harv. educ. Rev., 39, 449-483.
Jensen, A. R. (1970) Can we and should we study race differences? In J. Hellmuth (Ed.), Disadvantaged child, Vol. 3. Compensatory debate. New
education:
York,
A
national
Bnmner/Mazel.
Pp. 124-157. Jensen, A. R. (1971) The racexsexxability interaction. In Cancro, R. (Ed.), Contributions to intelligence. New York, Grune & Stratton. Pp. 107-161. Jensen, A. R. (1972a) The causes of twin differences in IQ. Phi Delta Kappan, 53, 420-421.
Jensen, A. R. (19 72b) Discussion of Tobath’s paper. In Lee Ehrman, G. S. Omenn, and E. W. Caspari (Eds.), Genetics, environment and behavior: Zmplications for educational policy. New
York, Academic Press. Jensen, A. R., (1972~) Educability and group differences. London, Methuen (New York, Harper and Row, 1973). Jensen, A. R. (1973) Educability and group differences. New York, Harper and Row (London, Methuen). Jinks, J. L., and Fulker, D. W. (1970) Comparison of the biometrical genetical, MAVA, and classical approaches to the analysis of human behavior. Psychol. Bull., 73, 311-349.
Leahy, Alice M. (1935) Nature-nurture and intelligence. Gen. Psychol. Mono., 17, 241-305. Light, R. J., and Smith, P. V. (1969) Social allocation models of intelligence: A methodological inquiry. Harv. educ. Rev., 39, 484-501.
Light, R. J., and Smith, P. V. (1971) Statistical issues in social allocation models of intelligence: A review and a response. Rev. educ. Res., 41, 351-367. Lush, J. L. (1968) Genetic unknowns and animal breeding a century after Mendel. Transactions of the Kansas Academy Science, 71, 309-3 14.
of
McNemar, Q. (1940) A critical examination of the University of Iowa studies of environmental influences upon the IQ. Psychol. Bull., 37, 63-92.
452
Arthur
R. Jensen
Noble, C. E. (1969) Race, reality, and experimental psychology. Perspectives in biology and medicine,
13, 10-30.
Page, E. B. (1972) Miracle in Milwaukee: Raising the IQ. Educ. Res., 1, 8-16. Starr-Salapatek, S. (1971) Race, social class, and IQ. Science, 174, 1285-1295. Shockley, W. (1971a) Negro IQ deficit: Failure of a ‘Malicious Coincidence’ model warrants new research proposals. Rev. educ. Res., 41, 227-48. Shockley, W. (1971b) Models, mathematics, and the moral obligation to diagnose the origin of Negro IQ deficits. Rev. educ. Res., 41, 369-377. Skodak, Marie, and Skeels, H. M. (1949) A Final follow-up study of one hundred adopted children. .I. gen. Psychol., 75, 8.5-12.5.
Spuhler, J. N., and Lindzey, G. (1967) Racial differences in behavior. In J. Hirsch (Ed.), Behavior-genetic analysis. New York, McGraw-Hill. Pp. 366-414. Terman, L. M. (1940) Personal reactions of the Yearbook Committee. In Whipple, G. M. (Ed.) Intelligence: Its nature and nurture, 39th Yearbook of the National Society for the Study of Education, Part I. Pp. 460-467. Tuddenham, R. D. (1970) A ‘Piagetian’ test of cognitive development. In Dockrell, B. (Ed.), On intelligence. Toronto, Ontario Institute for Studies in Education. Pp. 49-70. Vernon, P. E. (1965) Environmental handicaps and intellectual development: Part II and Part III. Brit. 1. educ. Psychol., 35, l-22.
Jensen’s
reply:
The sounds
of silence
DAVID LAYZER Harvard University
‘Is there any point to which you would wish to draw my attention?’ ‘To the curious incident of the dog in the night-time.’ The dog did nothing in the nighttime.’ ‘That was the curious incident,’ remarked Sherlock Holmes. Silver
Blaze
What I found most interesting about Jensen’s response to my article were its omissions. Reading Jensen out of context, one is stunned by the avalanche of irrelevancies; reading him in context, one is deafened by the silences. What are the key issues and the positions taken on them by Jensen and his critics? The following brief summary may help to refocus the discussion. 1. Jensen maintains that differences between blacks and whites in average performance on I.Q. tests probably have a genetic basis. His critics argue that there are no scientific
grounds for such an inference.
2. Jensen has suggested that children with low 1.Q.s may be genetically incapable of abstract thought. His critics point out that there is no evidence to support this notion and much to contradict
it.
3. Jensen concludes from published statistical analyses of I.Q. test scores that differences in genetic endowment account for about 80 % of the variance in these scores. His critics argue that this is an unwarranted inference from untrustworthy
data.
4. Jensen asserts that more and better statistical studies will lead to an improved understanding of inequalities in achievement between individuals and between Cognition 1 (4), 453-473
454
David Layzer
groups. His critics argue that such studies have little if any scientific value and are, in any case, largely irrelevant to the issues they nominally address. Let us see how Jensen deals with these issues. I.Q. differences
between blacks and whites
‘I have often been falsely accused of claiming that a high heritability of I.Q. inevitably means that mean differences in I.Q. between social class groups and racial grous must be due to genetic factors,’ writes Jensen on page 437. Nine hundred words later we discover wherein the accusation is false: ‘My study of the research evidence bearing on this question leads me to believe that a preponderance of the evidence is more consistent with a genetic hypothesis, which, of course, does not exclude the influence of environment.’ The reasoning offered in support of this belief is simple, plausible and - as we shall see - fallacious. Jensen argues that, because genetic factors could account for the reported I.Q. differences between blacks and whites, and because no statistical evidence has been presented to show that they are produced by anything else, we can reasonably assume that they are so produced. Why? Because ‘the probability that a phenotypic mean difference between two groups is in the same direction as a genotypic mean difference is greater than the probability that the phenotypic and genotypic differences are in opposite directions’ (p. 438). But this rule is nonsense. Suppose, for example, that average differences in skin color have been noted between two geographically separated populations. ‘Jensen’s rule’ would tell us that the observed differences are probably genetic in origin. But before we could draw this inference, we would need to know something about possible systematic nongenetic differences (e.g., in exposure to sunlight) and the effects of their interaction with possible systematic genetic differences (e.g., in pigmentation). When systematic nongenetic effects are known to be present, but their magnitude is completely unknown, we have no valid basis for drawing even a tentative conclusion about genetic differences. Among the relevant systematic differences affecting performance on I.Q. tests are cultural differences and differences in psychological environment. Both influence the development of cognitive skills in complex ways over a long period of time. So far, no one has found a way to estimate or eliminate the effects of cultural differences on I.Q. test scores. So-called culture-free tests are not the answer because there is no known way of separating ‘culture-free’ from ‘culture-bound’ aspects of cognitive development. The developmental effects of difference in psychological climate - differences in the way black and white children perceive themselves and are perceived by others - seem even less susceptible to quantitative as-
Jensen’s reply: The sounds of silence
455
sessment. Yet until such assessments have been made and their effects eliminated in the construction of the tests, no inference concerning possible genetic differences between blacks and whites can legitimately be drawn from I.Q. data. How does Jensen deal with this flaw in his argument? So far as I can tell, he simply ignores it. In so doing, he violates a cardinal rule of empirical research. All scientific measurements are subject to two qualitatively distinct kinds of errors: Random and systematic. Random errors introduce a kind of ‘noise’ whose magnitude can usually be estimated reasonably accurately in well-designed experiments and observations. Systematic errors, on the other hand, contribute not to the ‘noise’ but to the ‘signal’. That is what makes them so insidious. An experiment or observation intended to measure a particular quantity may in fact be measuring something quite different. Experience has shown that the only adequate way to deal with systematic errors is to eliminate them. This of course requires that they be well understood. No competent scientist simply ignores systematic errors he knows are there. An astronomical example, which offers some parallels to the problem of interpreting black-white I.Q. differences, illustrates this point. In certain hot stars the spectral lines of helium are abnormally weak. The spectra of these stars are also peculiar in other ways, but it is not known how these peculiarities relate to the abnormal weaknesses of the helium lines, or even whether they have anything to do with it. An astronomer who adopted ‘Jensen’s rule’ might argue that ‘a preponderance of the evidence is more consistent with the hypothesis’ that helium is deficient in these stars. But so far astronomers have resisted this temptation. The reason is not that they are, in Jensen’s words, methodological bluenoses, but that training and experience have taught them to avoid boobytraps of this kind. Some of my colleagues in the social sciences have suggested to me that the physical scientist’s rigorous approach to data analysis is inappropriate in the social sciences. Soft data, they say, require soft handling. I suggest that the exact opposite is true: The softer the data, the more rigorously we must analyze them in order to avoid drawing spurious conclusions. The irrelevance
of heritability
Several of Jensen’s critics, myself among them, have argued that the heritability of I.Q. (assuming this to be a meaningful concept) has no direct bearing on the social and educational issues to which Jensen’s article addresses itself. We assert that there is no logical conflict between the hereditarian premise - that I.Q. differences result largely from genetic differences - and the liberal premise - that appropriate social, cultural and educational changes can bring about changes in
456
David Layzer
the general level of cognitive competence as well as in the magnitude of the inequalities. Jensen’s original arguments to the contrary relied heavily on the metaphysical premise - that I.Q. tests measure an inherent capacity for abstract thought. From this it would follow that children with low I.Q.‘s can never acquire ‘higher’ cognitive skills. Jensen actually recommended that such children should be taught mainly by rote and that they should not be encouraged to aspire to occupations that require the exercise of higher cognitive skills (i.e. middle-class occupations). To my knowledge, students of cognitive development have found no evidence whatever to support the notion that a substantial proportion of normal children lack the capacity to acquire higher cognitive skills. In the light of present knowledge, Jensen’s metaphysical premise and the practical recommendations based upon it can only be described as gratuitous and irresponsible. Jensen also uses a second, entirely different, line of argument to discredit the liberal premise. This proceeds from the Utopian (or Leibniz-Pangloss) premise ‘that, despite all the criticisms that can be easily leveled at the educational system, the traditional forms of instruction have actually worked quite well for the majority of children’ (Jensen 1969, p. 7). Is this claim justified? It has been estimated that more than half of American adults have not attained the formal-operational stage of cognitive development (see, e.g., Kohlberg and Mayer 1972). If this finding is accurate, we must conclude either that half the American population lacks an innate capacity for formal-operational thought or else that traditional forms of instruction have worked rather badly. Thus, the Utopian premise, which denies the second possibility, is essentially a disguised form of the metaphysical premise, which affirms the first. In his reply, Jensen develops still a third argument, based on the relativistic premise - that ‘educational differences, not absolute level of performance, are the main cause of concern’ (p. 434). Statistical analyses of I.Q. test scores are relevant, Jensen says, because they help us to understand the causes of such inequalities. This argument has at least the merit of internal consistency. However, it does not stand up under examination. To begin with, as Christopher Jencks and his collaborators (1972) have shown, differences in cognitive skills and educational attainment explain only a small fraction of differences in wealth and economic status within contemporary American society. So an understanding of the causes of cognitive inequality can contribute only marginally to an understanding of socioeconomic inequality. But there is an even more basic flaw in the relativistic premise. Adequate participation in a democratic and highly technological society clearly demands minimum absolute levels of intellectual and moral development. Defining these levels
Jensen’s reply: The sounds of silence
457
and ensuring that they are regularly attained should, therefore, be a primary goal of education. Jensen argues that efforts to raise general levels of individual development through improved educational techniques would actually increase rather than diminish inequality. Whether or not this is a valid point - and it seems to me that it is not - it is surely irrelevant. If general levels of individual development can be raised substantially - if, for example, the proportion of American adults who attain the formal-operational stage of cognitive development can be increased from 50 to 95 % - does it matter that the best and the brightest will in the process become still better and brighter? (They are, after all, unlikely to become wealthier and more powerful.) The relativistic theory of cognitive inequality goes hand in hand with a relativistic view of poverty, which asserts that people consider themselves to be poor not because they lack money but because they are at the bottom of the socio-economic ladder. But poverty in the United States is not merely a matter of having less money than your neighbor. More important, it is a matter of not being able to find regular work, of being sick and not being able to get medical care, of growing up hungry and malnourished in filthy and dangerous surroundings, of being despised and feared by one’s more fortunate neighbors. It seems absurd to suppose that these social evils are automatic consequences of economic - still less of cognitive or educational - inequality. Now, even if differences in scholastic achievement are weakly correlated with differences in income, it remains true that in a highly technological society most nonmenial jobs demand skills and attitudes that many American children fail to develop. So the problem of raising general developmental levels does have a direct bearing on the problem of poverty, though of course it is not the only relevant consideration. Z.Q and heritability
Although the heritability of I.Q. does not bear directly on educability, on differences between socioeconomic and ethnic groups, or on the causes of poverty, the conclusion that IQ. has a heritability of about 80 % nevertheless has important implications. It suggests that under present conditions differences in intelligence are largely hereditary. If environmental inequalities were wholly eliminated, the spread in I.Q. (as measured by its standard deviation) would be reduced by only 10 %. These figures provide little incentive for reducing environmental and educational inequalities between inner-city slums and the affluent suburbs, or even for improving the quality of education in the affluent suburbs. But how seriously should we take the ‘scientific finding’ on which they are based?
458
David Layzer
One can dispute Jensen’s figure of 80 % (see below). But there is a more fundamental issue than the technical competence of Jensen’s heritability estimate: Is ‘heritability of I.Q.’ a meaningful concept? My article argued that it is not - that textbook recipes for estimating heritability cannot meaningfully be applied to I.Q. as they are applied to traits like height and weight. I.Q. scores are not measurements in the same sense as height and weight - or indeed in any useful sense at all. Every scientific measurement involves a quantitative - not merely a qualitative - comparison with some fixed but arbitrary standard. This makes possible quantitative comparisons between measurements of the same variable. By contrast, I.Q. scores contain qualitative information only. They merely rank-order individuals in a population. They tell us that A scored higher than B who scored higher than C, but the statement that the I.Q. difference between B and A is (say) twice that between C and B has no quantitative significance because the I.Q. scale is determined in an arbitrary manner. It could just as easily have been chosen so that the first difference was half the second. In fact, the I.Q. scale is constructed so as to make the frequency distribution of test scores approximate a normal error curve. For reasons discussed in my paper, the choice of a normal error curve has no particular theoretical significance in the present context. Any other frequency distribution would have done as well. Thus it is meaningless to compare I.Q. differences quantitatively. Since I.Q. has no quantitative significance to begin with, statistical analyses of I.Q. test scores can yield no meaningful quantitative information. This was the basis for my assertion that ‘heritability of I.Q.’ is an empty concept. How does Jensen deal with this argument in his Reply? He simply ignores it, addressing himself instead to an objection that he invents in order to refute: ‘A major point in Layzer’s argument is that I.Q. scores, in principle, are unsuitable for this type of analysis, since they do not constitute an absolute scale . . . like height and weight [but instead are measured in units of the standard deviation from the population mean as zero-point].’ This would indeed be an odd and vulnerable argument, since, as every natural scientist knows, the units and zero-point of any measurement may be freely specified, and the use of standard deviation as a unit and population mean as a zero-point is commonplace in statistical discussions, What is not arbitrary or conventional is the frequency distribution of the measurements in a given population. For true measurements, this may not be arbitrarily prescribed beforehand, as is done for I.Q. I find Jensen’s apparent inability to grasp this simple but basic fact about scientific measurements inexplicable. There is another equally important reason why I.Q. test scores cannot be assimilated to measurements of a metric character: They are not measurements of anything. The I.Q. test is like a black box that registers a definite number when
Jensen’s reply: The sounds of silence
459
it is connected to an appropriate subject, but whose inner construction is unknown. For reasons explained in my paper, scientific measurements cannot be made with black boxes. I know of no special dispensation that would exempt psychometrics from this rule. Jensen does offer one further argument in defense of I.Q. (p. 432). It must be meaningful to apply heritability analysis to I.Q. test scores, he writes, because, if it were not, ‘it is doubtful that some of the world’s leading quantitative geneticists . . . would have undertaken [such] analyses . . .’ The objections I raised, he continues, are not to be found either in ‘textbooks on quantitative genetics’ or in the papers of ‘many geneticists [who] have written about the genetics of intelligence.’ ‘Thus,’ Jensen triumphantly concludes, ‘a major pillar of Layzer’s critique simply crumbles.’ This passage of Jensen’s reply may hold particular interest for historians of science, illustrating as it does the survival of a type of argument usually thought to have become extinct around the middle of the seventeenth century. Heritability in general
The above objections could in principle be met by tests designed to measure specific cognitive skills in a quantitative way. It is therefore of interest to ask whether heritability would be a meaningful concept as applied to data from such tests. In my paper I pointed out that the applicability of heritability analysis has a much narrower domain than has often been supposed (even by textbook writers). Heritability analysis presupposes that variations of a trait within a population may be split up into a purely genetic component, a purely environmental (i.e. nongenetic) component, and a small correction term. Mathematically, this formula represents phenotypic variations by the first-order terms in a Taylor expansion about appropriately chosen mean values of the genotypic and environmental variables (see below). The correction term, assumed small, represents the combined effects of second- and higher-order terms in the Taylor expansion. Now, a first-order approximation is normally valid only for sufficiently small variations. How small is ‘sufficiently small’ must be decided in individual cases. But a useful qualitative criterion can be formulated: The first-order representation is approximately valid for genetic and environmental variations that do not appreciably alter a subject’s response to further genetic or environmental variations. Now, observational studies of cognitive development, especially those of Piaget and his collaborators, make it seem highly doubtful that any meaningful measure of intelligence could have the required properties. In the first place, genetic and environmental factors interact in an exceedingly complicated way during cognitive development. In the second place, development proceeds through sequences of
460
David Layzer
qualitatively distinct stages. Even if a meaningful quantitative measure of progress within a given stage could be devised, it seems hopeless - and indeed pointless to devise a quantitative scale embracing several distinct stages. For what is important about stages of cognitive development is precisely the fact that they are qualitatively, and not merely quantitatively, different. Thus, it seems to me that as soon as we begin to talk about the heritability of intelligence we are implicitly introducing certain assumptions about how intelligence develops. Not only are these assumptions gratuitous, but they seem to be incompatible with a considerable body of relevant observational evidence. These considerations, discussed at length in my paper, suggest that conventional heritability theory provides an inappropriate framework for discussing the role of genetic factors in cognitive development. That genetic differences are strongly implicated in observed behavioral differences is borne out by the most commonplace observations. But before we can hope to quantify the relationship, we must understand it qualitatively. And there is no reason to suppose that even then a quantitative description will be possible.
Effects of intervention The question whether environmental changes can substantially accelerate or decelerate cognitive development involves no difficult methodological or theoretical questions such as the validity of heritability theory or the epistemological status of I.Q. tests. It is a straightforward, empirical question. Moreover, it is obviously much easier to answer than the relation but distinct question, whether specific kinds of intervention produce substantial, long-term gains in tested I.Q. As I emphasized in my paper, intervention studies that have given negative results are not strictly relevant to the first question. On the other hand, studies that have given positive results are relevant even when, for methodological reasons, they do not have clearcut implications for the long-term effectiveness of specific kinds of intervention. Bronfenbrenner (1972) has reviewed a number of recent studies of early intervention. He finds that while disadvantaged children enrolled in highly structured, cognitively-oriented, pre-school programs consistently register substantial gains in I.Q., these gains gradually disappear after the children enter school, and concludes that ‘even the best curriculum cannot immunize a child against developmental decline, once he is cast back into a consistently impoverished environment.’ By contrast, home-based intervention programs, in which mothers are trained to interact with their children in ways expected to promote cognitive development, have been found to produce substantial and enduring gains in cognitive performance. ‘The earlier and more intensely mother and child were stimulated to engage in
Jensen’s reply: l’lte saunas
oJ
stlence
4t)l
communication around a common activity, the greater and more enduring was the gain in I.Q. achieved by the child.’ (Bronfenbrenner 1972, p. 48). To illustrate the importance of mother-child interactions in promoting cognitive and emotional development, Bronfenbrenner cites a study @keels 1966) in which thirteen subnormal children were ‘adopted’ by the inmates of a state institution for mentally retarded females. The mean IQ. of the children rose from 64 to 92, and the IQ.? of eleven of the children who were afterwards legally adopted rose subsequently to a mean of 101. The evidence presented by Bronfenbrenner is entirely consistent with the evidence and interpretations cited in my paper. Page’s criticisms of Heber’s study (based, by Jensen’s account, on a report not intended for publication or dissemination) are at best only marginally relevant to the point I intended to support in citing Heber’s study. For example, Heber’s alleged ‘failure to specify treatments’ in his unpublished report, an omission that Page considers serious enough to ‘largely invalidate the study,’ is obviously irrelevant to the question whether some forms of environmental intervention can substantially accelerate the cognitive development of underprivileged children. I think this is also true of the other objections quoted by Jensen. Whatever the study’s methodological shortcomings may eventually turn out to be, the reported 37-point gain in average I.Q. of the experimental group will not easily be reconciled with the notion that I.Q. is an index of innate mental capacity. Statistical considerations
In his statistical remarks, as elsewhere, Jensen dwells inordinately on minor points while giving short shrift to, or totally ignoring, those that really matter. Thus, while he elaborately defends his use of medians, he wholly ignores the real point of my remark that properly weighted means are more appropriate in theoretical contexts. Jensen says that he agrees with this remark in principle, but adds that in the particular case of I.Q. correlations, medians are more appropriate because they give less weight to ‘atypical’ studies with small probable errors. But why not assign low weight to such ‘atypical’ studies ? ‘I do not advocate such elaborate weighting,’ he writes, ‘because I believe it can introduce too much subjectivity and . . . would tend to lower our confidence in the composite.’ This is an extraordinarily revealing remark. The attitude it expresses seems ‘scientific’ at first blush, but in fact is just the reverse. Suppose (to vary Jensen’s example slightly) that the ‘atypical’ estimate was better than all the others - a fairly common situation in the natural sciences. The ‘objective’ estimate advocated by Jensen would still give it low weight. But it is standard procedure in the natural sciences to give more weight to good estimates
than
to bad ones.
Of course
subjective
judgment
is involved!
What
distinguishes
good from bad data analysis is precisely the quality of this judgment. Throwing ‘all the reported kinship correlations that [one] can find in [one’s] reprint files which are based on some kind of general intelligence test’ into a statistical mill is a good way to churn out objective results, but their scientific value will be minimal. As an example, consider the IQ. correlation between unrelated children reared together, a quantity that Jensen interprets (incorrectly) as a direct measure of the environmental contribution to I.Q. variance. The median correlation now quoted by Jensen is .25. Using appropriate weighting factors and statistical corrections, Jencks (1972, p, 292) arrives at an estimate of .38 for the same correlation. This failure to recognize and allow for qualitative differences between data from different sources is a characteristic feature of Jensen’s analysis, and its principal technical defect. By indiscriminately combining all the data in his files, he arrives at a neat table of kinship correlations showing a ‘pattern of correlation’ roughly similar to those of height and weight. The results are objective, but do they mean anything? The standard way of trying to answer this question is to examine their internal consistency. One section of my paper dealt briefly with this important point, which Jensen ignores both in his HER articles and in his Reply. The best discussion is that of Jencks (Jencks et al. 1972), who estimated the heritability h* separately for four different kinds of data. His findings may be summarized as follows: 0.29 < h2 < 0.76 0.45 < h2 < 0.60
from parent-child from comparisons
0.29 < h* < 0.50 from identical h < 0.25 from comparisons
data of identical
and fraternal
twins reared apart of siblings and unrelated
twins
children
Within each category, Jencks found inconsistencies among data from different sources, some of them systematic. The most pronounced systematic differences were between English and American data, the English data yielding consistently higher estimates of h* than the American. As the above tabulation shows, the results of estimates based on comparisons of siblings and unrelated children are inconsistent with all other estimates. Further, the estimates derived from studies of identical twins reared apart are substantially lower than those given by Burt, Jensen and other hereditarians. The reason is that Jencks’s estimates allow for expected environmental similarities between separated twins. Jencks summarized his conclusions about the probable value of hZ in these terms: ‘. . . we think the chances are about two out of three that the heritability of I.Q. scores, as we have defined the term, is between 0.35 and 0.55, and . . . the chances are about 19 out of 20 that [it] is between 0.25 and 0.65,’ Yet Jensen, using essentially the same data,
Jensen’s reply: The sounds of silence
463
concluded that h2 lies between 0.70 and 0.80, and in most of his discussions he uses the higher figure. Now, Jencks’s analysis indicates that the value 0.80 is about as likely to be correct as the value 0.10. Indeed, a heritability of 0.10 would at least be consistent with estimates based on comparisons between siblings and unrelated children, while Jensen’s estimates are inconsistent with all the data as analyzed by Jencks. How are we to interpret this discrepancy? A nonscientist might be tempted to suspect that Jensen, a hereditarian, had injected hereditarian bias into his data analysis and that Jencks might conceivably have allowed an anti-hereditarian bias to color his analysis. Fortunately, there appears to be a purely methodological explanation for the discrepancy. Where Jensen has unselectively combined data and uncritically applied theoretical formulas, Jencks has carefully analyzed and evaluated both the data and the causal relations (mathematical, biological and environmental) that are involved in establishing a link between I.Q. measurements and the notion of heritability. So far as I can tell, Jensen’s analysis is free from deliberate bias as it is free from other forms of human judgment. (Jensen’s belief in the existence of innate intellectual differences between races contrasts oddly with his belief that all data are created equal.) Jencks’s analysis, like all heritability analyses, presupposes that genetic and environmental factors contribute linearly, though not necessarily independently, to the observed phenotypic differences. Jencks himself does not question the applicability of standard heritability estimates to I.Q. test scores. However, as we have seen, his discussion reveals important internal inconsistencies in the data, as well as large systematic differences between heritability estimates based on different kinds of comparisons and different sources of data. How one interpets these inconsistencies, discrepancies and variations depends - and ought to depend - on one’s prior theoretical expectations (see my article for a fuller discussion of this point.) If there were good reasons for believing that I.Q. test scores represent genuine measurements of intelligence and that individual variations in intelligence have the mathematical structure required by heritability theory, then Jencks’s estimate of h2 could be interpreted in the conventional way, as an estimate of the fraction of I.Q. variance ‘explained by’ genetic variation. If, on the other hand, there are strong a priori reasons to doubt whether I.Q. test scores represent measurements and whether heritability theory would apply to them if they did, then we must begin by asking whether the results of Jencks’s analysis serve to allay these theoretical doubts. In my opinion they do not. On the contrary, it seems to me that they strongly confirm them. The internal inconsistencies and systematic discrepancies revealed by the analysis are just what one would expect to find if the kind of behavior sampled by I.Q. tests has a strong genetic component that cannot
464
David Layzer
be disentangled from environmental and experiential influences in the simple manner presupposed by heritability theory. But isn’t saying that cognitive differences have a strong genetic component the same as saying that I.Q. has a high heritability? Not at all. To say that cognitive development is strongly influenced by genetic factors implies that there may be substantial individual differences in the kinds of motivation and reinforcement that are effective in promoting different kinds of development, as well as differences in cognitive aptitudes and appetites. It does not, however, imply that environment plays a relatively minor role. In a Piagetian model of cognitive development, internal and external factors play complementary and equally essential roles, and both may be expected to contribute strongly to observed differences in cognitive performance, The model presupposed by heritability estimates implies, by contrast, that genetic and environmental factors contribute additively to the phenotypic variance, so that as the one grows in importance, the other must shrink. Thus Jensen, echoing Herrnstein, writes that ‘genetic factors . . . eventually would become [predominant] as we achieved more and more equality of environmental conditions, cultural and educational opportunities, and the like, thereby reducing the environmental variance.’ This argument implicitly equates equality of opportunity to enforced uniformity of personal experience. A society that strives to equalize opportunity can (and in my opinion should) also strive to increase the diversity of developmental possibilities. The more diversity of this kind a society affords, the greater will be the effective environmental differences between individuals. Misrepresentations
and misconceptions
Jensen’s arguments on the heritability of I.Q. rest on two basic misconceptions: (a) that I.Q. scores are scientific measurements, and (b) that intelligent behavior is determined by genetic and environmental factors in the excessively simple manner presupposed by conventional heritability theory. Jensen’s only rebuttal to my criticisms of these assumptions has been to point out that he is not alone in making them. The ad hominem defense of positions that are untenable on scientific grounds is matched by ad hominem attacks on those who venture to criticize his reasoning. Jensen’s argumentative technique is illustrated by his list of four alleged misrepresentations (pp. 428-430). (1) Jensen complains that I falsely attribute to him the view that I.Q. differences are 90 % genetic in origin. But a heritability of .8, the figure favored by Jensen, does imply that genetic differences account for 90 % of the observed variations as measured by their standard deviation. The same point is made by Jensen himself
Jensen’s reply: The sounds of silence
465
in his second HER article (Jensen 1969b, see especially fig. 3, p. 225). The rafio of genetic to environmental contributions to I.Q. differences is another matter. A figure of .8 for the heritability implies no definite value for this ratio; to calculate the ratio one must also know the genotype-environment correlation. If this correlation were zero, the ratio would be 2 to 1, the figure quoted by Jensen; if it were unity, the ratio would be 9 to 1; for a correlation of .2, the value estimated by Jencks, the ratio is about 3 to 1. The reason for this complicated mathematical behavior is that the genetic and environmental standard deviations, (TGand cE, add vectorially to give the phenotypic standard deviation op. That is, cp, (TGand eE are the three sides of a triangle (the G-E correlation is the negative cosine of the angle between uc and ua). A heritability of .8 implies that (TG= 0.9ap, but the ratio aJaE depends on the angle between bG and (TE. (2) Jensen’s second complaint has more substance. He did not, as I mistakenly said he did, explicitly interpret the difference 1 - .75 - .24 = .Ol as an estimate of the interaction contribution. This, however, was not the main point of my criticism. Jensen’s equation is bad science whether he interprets the residual of .Ol as interaction or as sampling error. The point is that the omission of uncertainty estimates for the figures .75 and .24 produces - perhaps inadvertently - a totally misleading impression of internal consistency because the uncertainties in question happen to be very large. Incidentally, since Jensen believes it would be ‘incredibly naive’ to estimate the interaction term by subtracting the explained variance from the total variance, as in the above equation, one wonders on what evidence he bases his confident assertion (p. 435) that this term ‘is either nonexistent or so insignificant as to be undetectable.’ I shall return to this point later. (3) The heritability theory applied by Jensen in his 1969 paper makes no allowance for such noncognitive genetic factors as sex and physical appearance which could appreciably influence I.Q. test scores. Jensen labels this criticism as a ‘misrepresentation’ on the grounds that in 1971 he published a ‘major research paper on sex differences and their interaction with racial differences’. (4) Finally, Jensen complains that I misrepresent his views when I contrast them with those of Waddington, who characterized the statistical techniques available for the analysis of heritability as ‘very weak and unhandy tools’ and who pointed out that experiments designed to clarify the detailed physiology and modes of inheritance of phenotypic characters cannot be sidestepped by refined statistical analyses. I am mildly surprised that Jensen does not perceive any substantial difference between Waddington’s approach and his own ‘operational’ approach, which proceeds from the premise that intelligence is what I.Q. tests measure. Jensen makes other scattered charges of misinterpretation, misrepresentation and misconception. In particular, he devotes an entire section to clearing up alleged
466
David Layzer
misconceptions about the much-quoted study by Skodak and Skeels (1949). Jensen concludes that the results of this study are ‘not at all consistent with a heritability of .80’. Yet Jencks (1972, pp. 28l-282), analyzing the same study more carefully, finds internal inconsistencies that make it impossible to fit the reported data by any plausibIe model based on conventional heritability theory. The importance of genotype-environment
interaction
I have argued that there are strong a priori reasons for questioning the simple additive model on which heritability estimates of I.Q. are based, and that results of careful statistical analyses like that of Jencks provide scant empirical support for the model. Jensen not only affirms the validity of the additive model, but adds that ‘existing models of heritability analysis take [genotype-environment interaction] into account and are capable of estimating the proportion of variance attributable to [it] . . . The interaction component [of I.Q.] is either nonexistent or so insignificant as to be undetectable in the existing data. If it were of substantial magnitude, it would show up with present methods of analysis . . .’ (p. 435). Although these claims are central to his argument, Jensen makes hardly any effort to explain or substantiate them, relying on simple repetition and a single reference to the literature. In this reference (Jinks and Fulker 1970) - submitted for publication after the appearance of Jensen’s HER article - we read that ‘numerous tests for [genotype-en~ronment] interaction have been described for use with controlled plant and animal breeding programs, but none, so far, have been proposed for use with human data.’ Jinks and Fulker do indeed propose and apply such a test (about which more later), but their conclusions fall far short of the sweeping assertions just quoted. But Jensen has not merely overstated his case. His remarks clearly demonstrate that he understands neither the mathematical nor the practical problems involved in the estimation of interaction effects. To explain these problems, I must use a little mathematics. Let X(x,y) denote a metric character that depends on a genetic variable x and an environmental variable y, and let R and 7 denote the mean values of x and y in a given population. We expand X in a Taylor series about these mean values: X(x,y) = X(%7) + (x - a) Xx (jt, 7) + (y - 7) Xy (x,7) + X*(x, y),where partial derivatives with respect to x and y are indicated by subscripts, and X* represents the second- and higher-order terms in the Taylor expansion. Now average this equation over the reference pop~ation~a~d subtract the resulting equation from the original one to obtain an equation of the form F=G+E+F*
Jensen’s reply: The sounds of silence
467
where F denotes the fluctuating part of X (i.e., the deviation of X from the population mean), G denotes the genetic contribution to F, E the environmental contribution, and F* is the part of F that depends nonlinearly on the genetic and environmental variations. Note that G depends linearly on the genetic variation (x - a) and E on the environmental variation (y - 7). Instead of interpreting the quantities in the last equation as numbers, we may interpret them as random variables. A random variable is specified by the set of its possible values and the corresponding probabilities (or frequencies). For example, the random variable F is specified by its possible values Fi (say) and the corresponding probabilities pi. Denoting averages by brackets < >, we can then write = Cf(Fi)pi where f denotes an arbitrary function, and the sum runs over all individuals in the given population. Next, consider two populations or subpopulations between whose members a one-to-one correspondence has been established. Let unprimed variables refer to the first population, primed variables to the second. We then have = + <EE’> + ( + ) + I(F,F’) (1) where the interaction term I(F,F’) is given by I(F,F’) = + + + + (2) If the second population coincides with the first and the one-to-one correspondence is a self-correspondence, equation (1) reduces to rVar F = + 2 + I(F,F) = Var G + Var E + 2 Cov(G, E) + I(F, F), the fundamental formula of variance analysis. Note, however, that the component I(F, F) is now given explicitly by equation (2). Next, let the two populations be composed of separated MZ twins, assume that the environments of co-twins are uncorrelated, i.e. <EE’> (1) reduces to = Cov(F,F’) = + + + 12(F,F’) (separated
(la) interaction and let us = 0. Then (lb) MZ twins)
Finally, let corresponding members of the two populations be unrelated foster children in the same home and assume that = = = 0, i.e., that the genotypes of the children are statistically unrelated to one another and to the home environment. Then (1) reduces to = Cov (F,F’) = <E2> + Is(F,F’) (lc) (unrelated foster children) In the same way, analogous formulas and other covariances. From suitable
may be written down for parent-child, sib-sib, information about assortative mating and gene
468
David Layzer
interaction, one can, in principle, relate the term = Cov(GG’) in each such equation to = Var G. But the kth equation contains an interaction term Ik(FF’) that cannot be related to the interaction terms in any of the other equations. This is clear from formula (2) for I and the fact that nothing is known about the way in which F* depends on the genetic and environmental variables. Thus the number of unknown interaction terms is always equal to the number of equations. Since the genetic and environmental variances, Var G and Var E, as well as the genotypeenvironment covariance Cov(G,E), are also unknown, it is obviously impossible to evaluate I(F, F), the interaction contribution to Var F. The only possible way of estimating the interaction contribution to Var F would be to make specific assumptions about the dependence of F* on G and E and about covariances like , , etc. But what could such assumptions be based on? One would need to have a fairly elaborate mathematical theory of interaction effects, and no such theory has yet been proposed. If we could assume that the interaction term F* was sufficiently small, we could represent it by the formula F* = a(G2 - ) + b(E2 - cE2>) + 2cGE, where a, b, c are undetermined constants. If, in addition we postulate that the joint probability distribution of G and E is normal, we can express the third- and fourthorder moments of G and E that occur in the Ik in terms of the second-order moments, Var G, Var E and Cov(GE). In principle, then, we could evaluate the unknown constants a, b, c [needed to calculate the interaction component I(F, F’)] if we had at least six independent relations like (la-c) - provided, of course that we also had adequate information about assortative mating and gene interaction. I think it would be extremely difficult to carry out such a program in practice. The data aren’t good enough and there are too many unknowns. But even this complicated and unpromising procedure for evaluating interaction effects breaks down unless F* is much smaller than the additive contributions G and E. When this condition is not met, the problem is mathematically indeterminate, and there is no possibility of estimating the interaction component of the phenotypic variance. So much for the theoretical situation. How do psychometricians cope with the problem of interaction in practice? So far as I can tell, they simply ignore it and hope for the best. As we have just seen, they have no practical alternative within the framework of conventional heritability analysis, The question then arises, whether the results of their analyses support the underlying assumption that interaction effects are small. Jensen asserts, emphatically and unequivocally, that they do, but he is mistaken. The test of a simplified mathematical model is the internal consistency of data analyses based on it. As we have seen, careful analyses of I.Q. test scores, like
Jensen’s
reply: The sounds
of silence
469
that of Jencks, reveal important internal inconsistencies. To what extent these inconsistencies result from interaction effects is not clear. All that can be said with confidence is that the simple theory does not fit the data at all adequately - certainly not well enough to cast doubt on the theoretically plausible proposition that genotypeenvironment interaction is strongly implicated in performance on mental tests. Statements to the contrary seem to be based on a combination of inadequate data analysis and wishful thinking. In principle, interaction effects are not difficult to observe. As before, let F(x, y) denote the deviation from the population mean of a metric character that depends on a genetic variable x and an environmental variable y, and let AXF = F(x + Ax,y) F(x,y) denote the change in F associated with a given genetic difference Ax. For the present argument x and Ax need not have quantitative significance; Ax may represent an unspecified genetic difference between two specified strains. Now consider how the phenotypic difference between the two strains varies with the environmental variable y (which could represent temperature, rainfall, etc.). If A,F is substantially independent of y within a given range yi < y 5 yz, we may say that interaction effects are small in the range. If A,F varies substantially, interaction effects are large. One reason why it has not so far been possible to investigate the effects of interaction on behavioral indices in this way is that we do not yet know what the relevant environmental variables are. Those environmental variables that can be adequately measured and controlled for (income, socioeconomic status, etc.) are not necessarily the most relevant to development, while variables that are thought to influence intellectual and emotional development strongly (e.g. mother-child interactions) cannot be adequately measured or controlled for. For this reason psychometricians are obliged to resort to less direct (and consequently less reliable) methods of studying interaction effects. The following crude argument is sometimes used. The correlation between separated twins reared in uncorrelated environments provides a measure of the purely genetic component of the variance; the correlation between unrelated foster children reared in the same home provides a measure of the purely environmental component. Since these two estimated correlations often add up to a figure greater than or equal to unity, the interaction contribution to the variance must be small. The flaw in this argument is simple. Both covariances (between identical twins and between unrelated foster children) are strongly influenced - though in different ways by interaction effects. Consider unrelated foster children raised in the same home. Cohabitation does not guarantee identity of attitudes, rewards, motivations, reinforcements, etc. Positive feedback often amplifies and fixes initial differences, whatever their origin. The following analogy may help to clarify this point. Suppose that a number of genetically identical seeds are sown in a forest. All have the same en-
470
David Layzer
vironment in the usual sense of the word. Yet some seeds will develop into large, healthy trees, others into small trees, and many will die out. So there will be a very large variance in the size of mature trees although both the environmental and genetic variances are nil. It is easy enough to understand how this comes about. Small accidental differences in initial conditions have been greatly amplified during development. The resulting differences are due entirely to interaction. Similarly, large differences (and hence low correlation) between foster children raised together may be presumed to be due largely to interaction. In the case of genetically identical individuals, the effects of positive feedback tend to amplify common d@rences between the twins and other children. Thus if the environments of split monozygotic (MZ) twins are not very dissimilar - as is usually the case - feedback tends to promote the convergence of developed traits, and thus tends to increase the intrapair correlation. Another test of interaction has been proposed by Jinks and Fulker (1970). Let F, F’ refer to split MZ twins reared in uncorrelated environments (<E,E’> = 0), and assume for simplicity that E and E’ (hence F and F’) have identical frequency distributions. Define F+ = l/2 (F+F’), F- = l/2 (F-F’). Jinks and Fulker argue that if interaction is present, F+ and F- will be correlated. They accordingly propose to measure the importance of interaction by Cov (F+, F-). But it follows from our assumptions that Cov(F+, F-) E = l/4 - l/4 = 0. That is, the correlation in question vanishes even if interaction is present. Jinks and Fulker may, however, have intended to use the positive quantity IF-1 instead of the actual difference F-. (Their worked example suggests this interpretation.) It is easy to show that Cov (IF-/,F+) vanishes in the absence of interaction but not, in general, otherwise. In particular, if F* CLGE, then Cov(/F-1, F+) = < /E-/B Var G. On the other hand, the same covariance vanishes for other kinds of interaction, e.g. F* = aEIGI, F* = crEG2, etc. Having a statistical test and being able to apply it are not, unfortunately, the same thing. The test just described requires a large sample of separated MZ twins satisfying certain statistical conditions presupposed by the test. For reasons discussed in my article and by Jencks (1972) these conditions are not met by existing twin studies. Again, interaction effects are most conspicuous for large environmental variations. Such variations (in factors thought by students of cognitive development to be relevant to performance on mental tests) are notably less frequent in the available twin studies than in the population at large. For these and other reasons, the available data on separated MZ twins contain little, if any, useful information on the magnitude of interaction effects. Jinks and Fulker applied their interaction test to I.Q. data from the Newman,
Jensen’s reply: The sounds of silence
471
Freeman and Holzinger (1937) study of 19 pairs of separated MZ twins and found no evidence of interaction. But in view of the small sample and the strong bias toward small environmental differences (emphasized by the authors of the original study), one would hardly have expected a statistically significant positive result even if very substantial interaction was present. So far as I know, no other application of the Jinks-Fulker test to I.Q. data has been reported in the literature. From an unpublished tabulation of Burt’s data on the 1.Q.s of separated MZ twins (53 pairs) Miss L. C(F-)z Schultz has calculated Var F- = -for three subsamples: Pairs with F+ < -15; N-l pairs with -15 5 F+ < 15; pairs with Ff > 15. The results are shown in the accompanying table.
Range of F+
<--I.5 -15
to +15
> +15
(Because
of the probably
strong
5.8
2.4
7
16.6
4.1
38
10.7
3.3
8
bias toward
small environmental
differences,
the
correlation of F+ and IF-1 would give undue weight to the middle range of F+.) In view of the small sample size, no strong conclusioncan be drawn. But there is a suggestion that smaller I.Q. differences are likely to develop between twins whose mean I.Q. lies in one of the tails of the distribution than between twins whose mean I.Q. lies within one standard devaiation of the mean. To sum up, neither the indirect test of internal consistency of heritability estimates nor the direct test proposed by Jinks and Fulker supports the extravagant claims quoted at the beginning of this section. The available data are consistent with the expectation that genotype-environment interaction is an important factor when large environmental differences - which are the ones we are chiefly concerned with - are involved. Jensen’s assertion that genotype-environment interaction ‘would easily show up’ - like an uninvited party guest as it were - ‘with the present methods of analysis’ has no basis in scientific experience, either in psychometrics or elsewhere. Unlookedfor effects do occasionally ‘show up,’ but not in data as noisy and internally inconsistent as the I.Q. data. Usually one finds what one is looking for. Thus, no significant contribution from genotype-environment correlation (the term 2 Cov(G,E)
472
David Layzer
in equation (la) above) ‘showed up’ in Jensen’s analysis of I.Q. variance. Yet this contribution is highly conspicuous and significant in Jencks’s (1972) reanalysis of essentially the same data.
Summary
and conclusions
I have argued that Jensen’s views on the nature and inheritance of intelligence and on racial differences in intelligence rest on fallacious arguments and misinterpretations of low-quality data whose limitations he fails to recognize. In his voluminous writings, Jensen rarely addresses himself to the central issues. When he does, as in his remarks on genotype-environment interaction, he relies on simple assertion and citation of authority.1 The substance and style of Jensen’s Reply should serve to remove any remaining doubts as to the credibility of his claims. The I.Q. controversy is not a conflict between scientific positions, scientifically supported, nor yet between ideologies or values, though these are of course involved. At its deepest level it is an argument about what constitutes scientific data and scientific forms of inference. To a nonscientist such questions may seem excessively academic. Yet they hold the key to the question: Are there any valid scientific grounds for taking seriously Jensen’s ideas about race and intelligence? The answer, in my opinion, is a resounding and unqualified No. 1. The reader will do well to examine these citations with care. For example, this is how Jinks and Fulker (1970, p. 342) summarize their findings on genotype-environment interaction: ‘The possibility of [genotype-environment interaction] still remains. However, the probable absence of substantial genotype-
environment interaction for Shields’s two intelligence tests, and for the Newman, Freeman, Holzinger (1937) data examined briefly in the final example.. . make it at least plausible to suggest this source of bias is not likely to be very important either.’
REFERENCES
Bronfenbrenner, vention
U. (In press) Is early interCornell University
effective?
Press. Burt, C. (1966) The genetic determination of differences in intelligence: A study of monozygotic twins reared together and apart. British Journal of Psychology, 57 (1 and 2), 137-153.
Jencks, C., Smith, M., Acland, H., Bane, M. J., Cohen, D., Gintis, H., Heyns, B. and Michelson, S. (1972) Inequality: A reassessment schooling
of the effect of family in America. New York
and
and
London, Basic Books, Inc. Jensen, A. R. (1969a) How much can we boost IQ and scholastic achievement?
Jensen’s reply: The sounds of silence
Harv. educ. Rev., 39 (l), 1-123. Jensen, A. R. (1969b) Reducing the heredityenvironment uncertainty, Harv. educ. Rev.,
39 (3), 449-483.
Jensen, A. R. (1972) The IQ controversy: A reply to Layzer. Cog., 1 (4), 427-452. Jinks, J. L. and Fulker, D. W. (1970) Comparison of the biometrical, genetical, MAVA, and classical approaches to the analysis of human behavior. Psychol. Bull., 73 (5), 311-349.
Kohlberg, L. and Mayer, R. (1972) Development as the aim of education. Harv. educ. Rev., 42 (4), 449-496.
473
Layzer, D. (1972) Science or superstition? (A physical scientist looks at the IQ controversy). Cog. 1 (2-3), 265-300. Newman, H. H., Freeman, F. N. and Holzinger, K. J. (1937) Twins: A study of heredity and environment. Chicago, The University of Chicago Press. Skeels, H. M. (1966) Adult status of children with contrasting early life experiences: A follow-up study. Child Devel. Mono., 31 (3), Serial No. 105. Skodak, M. and Skeels, H. M. (1949) A final follow-up study of 100 adopted children. 1. gen. Psychol., 75, 85.
