The Role of Cognition in Depression: A Critical Appraisal - American ...

Copyright 1983 by the American Psychological Association, Inc.

Psychological Bulletin 1983, Vol. 94, No. 3, 472-505

The Role of Cognition in Depression: A Critical Appraisal James C. Coyne

Ian H. Gotlib

University of California, Berkeley

University of Western Ontario London, Ontario, Canada

Research examining the role of cognition in depression is critically reviewed. Specifically, five broad areas of cognitive functioning are examined: expectations and evaluations of performance, perception of environmental information, recall of information, cognitive biases, and attributional processes. The review of the literature suggests that neither Beck's nor the learned helplessness model of depression has a strong empirical base. Depressed persons present themselves negatively on a variety of measures, but less consistently than either model suggests. Differences between depressed and nondepressed persons with respect to extralaboratory experiences and self-presentational strategies remain viable alternative explanations for those results that have been obtained. In addition, specificity to depression has not been demonstrated consistently for any measure of cognitive bias or distortion. Attention is given to conceptual and methodological difficulties in unambiguously establishing what people think, in demonstrating biased or distorted cognitive processes, and in testing hypotheses about the fundamental role of cognition in depression.

As measured in sheer number of articles appearing in the journals, cognitive approaches to the study of depression have achieved an ascendancy over alternative psychological conceptualizations of the disorder. A vast and growing literature seems to present a compelling case for the cognitive viewpoint. A large number of studies have been reported describing cognitive aspects of depression, with investigators arguing that depression-related cognitions are the cause (Shaw & Dobson, 1981), a symptom (Schreiber, 1978), or the consequence (Lewinsohn, 1974) of depression. There have also been demonstrations of the effectiveness of cognitive therapy for depression, with some indication that such an approach produces greater improvement, more cases of marked or complete remission, and fewer dropouts than either drug treatment (Rush, Beck, Kovacs, & Hollon, 1977) or behavior therapy (Shaw, 1977). Although cognitive conceptions of depression can be viewed as simply part of a general The authors would like to express their appreciation to Nicholas A. Kuiper, Paul H. Blaney, Michael R. MacDonald, and James M. Olson for their helpful comments on an earlier draft of this article. Requests for reprints should be sent to James C. Coyne, Department of Psychology, University of California, Berkeley, California 94720.

resurgence of cognition in psychology, it is also true that the plausibility of cognition explanations for the development and persistence of depression has been a major factor in spreading the "cognitive revolution" to the study of psychopathology. The verbalizations of depressed persons have been characterized as self-deprecating and unrealistically pessimistic. Furthermore, despite their obvious distress, depressed persons frequently fail to take initiative to remedy their situations, or do so only half-heartedly and typically without success. To many writers, such phenomena suggest that depressed persons suffer from distorted cognitive processes. It is a simple matter to construct questionnaires that capture the expressed self-complaints and pessimism associated with depression, and the statistically significant correlations of such scales with measures of depression such as the Beck Depression Inventory (BDI; Beck, Ward, Mendelson, Mock, & Erbaugh, 1961) provide a large part of the data invoked in support of cognitive models of depression. The ease with which such relationships can be demonstrated, as well as the general intuitive appeal of cognitive explanations of depression, has tended to discourage critical examination of the role of cognition in this disorder. Rival hypotheses are seldom

472

COGNITION, IN DEPRESSION

articulated, and weak correlational methodologies are frequently accepted as providing more stringent tests of cognitive hypotheses than is actually the case (Coyne, 1982). The purpose of the present article is to summarize and evaluate the accumulating research data concerning the role of cognition in depression. The article will begin with a brief overview of the two major theoretical positions on which the largest proportion of studies in this area are based—those of Beck (1967, 1976; Kovacs & Beck, 1978) and Seligman (1975; Abramson, Seligman, & Teasdale, 1978). A discussion of a number of general issues in the field concerning both the selection of depressed subjects and the choice of comparison control groups will then be presented. This discussion will be followed by a critical review of the research literature in five broad areas of cognitive functioning as they relate to depression: expectations and evaluations of performance, perception of environmental information, recall of information, cognitive biases, and attributional processes. Particular attention will be given to the adequacy of existing methodologies for testing cognitive hypotheses about depression. Finally, an overview and integration of this body of research will be presented, and suggestions concerning directions for future work will be advanced. Beck's Model of Depression Beck (1967, 1976) invokes three sets of cognitive concepts to explain psychological aspects of depression: the cognitive triad, schemata, and cognitive distortions or faulty information processing. The cognitive triad consists of thinking patterns that lead depressed persons to regard themselves, their current situation, and their future possibilities in negative terms. According to Beck, the triad manifests itself in the misperceptions and misinterpretations voiced by depressed persons. Depressives perceive situations negatively when more positive interpretations are equally plausible. Their interactions with the environment are misconstrued in terms of defeat and deprivation, and failure is expected when new tasks are undertaken. Depressed persons tailor facts to fit their negative conclusions, and in cases of more severe depression, thinking patterns may be-

473

come unresponsive to external input (Beck, Rush, Shaw, & Emery, 1979, pp. 11-13). The concept of schemata is utilized by Beck (1976) to explain why depressed persons persist in self-defeating and pain-inducing attitudes in the face of contradictory evidence. Cognitive schemas are stable, organized representations of past experience that provide the basis for screening, differentiating and coding environmental stimuli that impinge on a person. This concept can be used to explain why a particular individual may react differently to objectively similar or identical events, or may show the same type of response in apparently dissimilar situations (Kovacs & Beck, 1978). In depression, experiences are distorted through prepotent dysfunctional schemas. These schemata prevent the matching of more appropriate schemas to situations, and as they become more active, they can be evoked by an expanding range of stimuli. The operation of these idiosyncratic schemata is reflected in the systematic errors in the thinking of depressed persons. For example, depressed people overgeneralize from negative experiences, they selectively abstract negative details out of context, ignoring more positive features of their situations, and they negatively characterize themselves in absolutistic and dichotomous terms. Cognitive theorists assign a primacy to cognitive factors in depression, and treat affective and motivational phenomena as secondary derivatives. As Beck et al. (1979) state, "the patient's negative constructions of reality can be postulated to be the first link in the chain of symptoms (or 'phenomena')" (p. 19). Variations in depressive affect such as sadness, shame, guilt, or boredom are proposed to be the result of particular cognitive themes. Similarly, motivational deficits such as indecisiveness and abulia are hypothesized to stem from depressed persons' cognitive constructions of their situations. Beck (Beck et al., 1979) notes that changes in the social environment can precipitate the cognitive processes leading to depression, and also suggests that harmonious interpersonal relationships can neutralize depressives' selfderogatory tendencies. However, he has chosen to focus on depressed persons' constructions of their relations with the environment rather than on their actual interactions, and consis-

474

JAMES C. COYNE AND IAN H. GOTLIB

tently highlights the distortions in depressives' interpretations of events and their imperviousness to environmental feedback (e.g., Kovacs & Beck, 1978, p. 526). The Revised Learned-Helplessness Model of Depression The phrase "learned helplessness" was first invoked to explain the debilitated escapeavoidance responding shown by dogs exposed to inescapable shock in the laboratory (Overmier & Seligman, 1967). The concept was subsequently extended to account for the debilitated task performance of humans exposed to experimenter-induced failure, and was then offered as an explanatory model of depression. Inadequacies in the application of the learnedhelplessness model to humans led to a number of reformulations, each of which introduced attributional concepts as mediators between the perception of noncontingency and expections regarding future contingencies (cf. Abramson, Seligman, &Teasdale, 1978; Miller & Norman, 1979; Roth, 1979). These reformulations are better able to accommodate data from humans, but given the importance they attach to higher cognitive processes, they have rendered the parallels between human and infrahuman phenomena more tenuous. Both the original and the reformulated learned-helplessness models assert that depressed persons have learned that outcomes are uncontrollable, and that this learning results in the manifestation of cognitive, motivational, and emotional deficits. In the reformulated model, mere exposure to current uncontrollability is insufficient for these deficits to occur; persons must come to expect that future outcomes are uncontrollable in order for helplessness to be induced. "When a person finds that he is helpless, he asks why he is helpless. The causal attributions he makes determine the generality and chronicity of his helplessness deficits as well as his later selfesteem" (Abramson, Seligman, & Teasdale, 1978, p. 50). Thus, the theoretical sequence of the development of helplessness is Objective noncontingency —* Perception of present and past noncontingency —• Attribution for present or past noncontingency —• Expectation of future noncontingency —• Symptoms of helplessness. (Abramson, Seligman, & Teasdale, 1978, p. 52)

Persons who are prone to depression tend to attribute negative outcomes to internal, global, and stable factors. Seligman, Abramson, Semmel, and von Baeyer (1979) suggest further that depressed persons might also attribute good outcome to external, specific, and unstable causes. Although this was not specifically predicted by the Abramson, Seligman, and Teasdale (1978) reformulation of the model, it is consistent with the learned-helplessness position that depressed persons fail to perceive response-outcome associations. Attributions of negative events to internal factors lead to lowered self-esteem; attributions to global factors lead to a generalization of deficits across situations; and attributions to stable factors lead to a persistence of deficits over time. Abramson, Seligman, and Teasdale (1978) state that the "chronicity, generality, and intensity of depression follow inexorably, 'rationally' from the attributions made and the importance of the outcome" (p. 68). Abramson, Garber, and Seligman (1980), however, emphasize that it is the expectation of helplessness that is critical for depression, and point out that although formation of attributions is the most salient way of bringing this about, other sources of information such as verbal persuasion, modeling, and physiological changes may also affect expectations. The learned-helplessness model assumes that the behavioral problems of depressed persons involve failures to initiate responses, a pattern that also persists after exposure to successful outcomes. Alloy and Seligman (1979) acknowledge the difficulty in unambiguously establishing that the deficit underlying this response tendency is cognitive rather than motivational or emotional. The cognitive deficit is postulated to be an interference with the learning of the association between responses and outcomes. The motivational deficit is posited to be a retarded initiation of voluntary responses, and the emotional deficit is, of course, sad affect. Abramson et al. (1980) note that the cognitive and motivational deficits associated with helplessness can occur in the absence of dysphoria. "Only those cases in which the expectation of response-outcome independence is about the lack or loss of a highly desired outcome or about the occurrence of a highly aversive outcome are sufficient for the emotional component" (p. 28).

COGNITION IN DEPRESSION

Unlike sad affect, therefore, the cognitive and motivational deficits associated with helplessness are not postulated to depend on the importance of the outcome for their intensity. Comparison of the Models The original learned-helplessness model (Seligman, 1975) asserted that depressed persons perceive themselves as having no control over significant negative outcomes, and this perceived lack of control results in the motivational and emotional symptoms of depression. Beck's (1967, 1976) model of depression, on the other hand, asserted that depressed persons assume excessive responsibility for negative outcome (self-blame). The juxtaposition of these two formulations appeared to present a major paradox, if not an outright contradiction (Abramson & Sackeim, 1977; Blaney, 1977), and it was possible to specify experimental situations in which predictions from the two models could be contrasted (Rizley, 1978). However, with the introduction of an attributional component to the learned-helplessness model, it became more difficult to specify data that would conceivably support one model while strongly contradicting the other. Perhaps many of the apparent differences between the two models stem from their underlying paradigms or metaphors for cognition. With its emphasis on schemas and faulty thought patterns, Beck's model adopts an information-processing metaphor. Like other attributional models, the reformulated learned-helplessness model of depression is based on a Brunswik lens metaphor for cognition (see Taylor & Fiske, 1983, for a more complete discussion of these metaphors). Although the two models can be caricatured in order to produce contrasting predictions, they do not represent radically different conceptions of the depressed person; their differences are mainly matters of emphasis. Fiske and Linville (1980) have suggested that schematic and attributional analyses prove to be complementary in many situations. Schematic analyses such as Beck's, therefore, can be conceptualized as giving particular attention to the organization of prior knowledge and to how this organization shapes the processing of incoming information. In contrast, attributional models

475

such as the learned-helplessness model focus on how incoming information is explained and how such explanations determine subsequent cognitions, affect, and behavior. In the following review of the relevant research literature, it will be seen that differences between these two cognitive models of depression tend to disappear in attempts to account for the results of a given study. Empirical Studies Depressed Samples and Control Groups Before turning to specific studies of cognition in depression, we consider briefly a number of general issues surrounding the selection of "depressed" subjects and the choice of comparison control groups. Although problems in these areas tend to be downplayed or ignored by researchers, they do pose a major difficulty in any integration and interpretation of substantive findings. Subjects drawn from clinic and inpatient samples are now being utilized more frequently in investigations of depression. Nevertheless, the majority of existing studies have employed mildly depressed college students selected on the basis of their scores on the Beck Depression Inventory (BDI). The assertion is frequently made that mild depressive states in persons drawn from nonclinical populations represent the low end of a continuum of severity, with clinical depression at the opposite pole, quantitatively but not qualitatively different. Depue and Monroe (1978), however, summarized the results of a number of studies that suggest that although mild depressions in relatively normal subjects share the subjective mood and cognitive components of depression in clinical populations, they lack the overt behaviors, somaticized anxiety, and physical complaints of clinical depression. In addition, Golin and Hartz (1979) factor-analyzed a "trait" version of the BDI administered to college students and failed to find the coherent somatic disturbance factor previously identified in a psychiatric inpatient sample (Weckowitz, Muir, & Cropley, 1967). Not only are there differences between mildly and clinically depressed individuals with respect to their responses to the BDI, but there is also an equivocal relationship between

476


the BDI and ratings of depression based on clinical interviews. Although Bumberry, Oliver, and McClure (1978), for example, found a .77 correlation between BDI scores and a psychiatrist's rating of depression on a 4-point scale, the mean BDI score of students placed in the "moderately depressed" category was higher than that of those students rated as "severely depressed." In a second sample, a psychiatrist's ratings obtained 1-14 days after the BDI was completed yielded a correlation of only .30. Similar findings have been reported by Hammen (1980) and by Lewinsohn and Teri (1982). Furthermore, it is likely that many elevated scores on self-report questionnaires are of only limited duration. Hammen (1980), for example, found that when her sample of students scoring 15 or above on the BDI were tested 2-3 weeks later, their mean score had dropped from 18.37 to 10.27. In light of these data, it is important to note that researchers frequently allow considerable time to elapse between administration of the BDI for selection purposes and actual participation in experimental procedures; moreover, the length of this time lag is seldom reported. Discrepancies among similar studies utilizing college students, therefore, may be due to differences in this elapsed time (Sacco, 1981). Despite these concerns, the fact remains that mild depression in college students, in and of itself, is a legitimate topic for study. The study of mild depression can also be useful in attempts to test the hypothesis that a particular phenomenon occurs only in severe depression, and is therefore a consequence rather than a contributing factor in its development (cf. Strack & Coyne, 1983). Beyond this, there is little justification for broad generalizations beyond the populations studied (Doerfler, 1981). Nonetheless, for reasons that likely have more to do with the isolation of researchers from clinical populations than with the viability of the continuity hypothesis, the bulk of the literature examining cognition in depression currently involves the study of mildly depressed college students, and will probably continue to do so. The fact that investigations involving this population are likely to remain the pool from which hypotheses, methods, and measures will be drawn for less frequent studies of clinical populations raises a number of concerns. For

example, in some instances researchers may accept a null hypothesis of no depressed-nondepressed differences with respect to a particular phenomenon when comparisons involving a clinical sample would have produced significant results. Second, transient mood disturbance may be more susceptible to the demand characteristics of manipulations and measures (Buchwald, Strack, & Coyne, 1981), and differences arising due to these factors may mistakenly be given more substantive interpretations. This is particularly likely when the interest of the study is in demonstrating changes in affect or cognition. Third, factors associated with the perpetuation of severe mood disturbance may be discontinuous with those affecting changes of a more transient nature, and a nearly exclusive focus on mildly depressed college students may obscure these differences. All of these problems affect other psychological studies of depression, and cognitive researchers should not be singled out. However, an overreliance on college student samples poses a final problem, particular to the development and testing of cognitive models of depression. Common to cognitive conceptions of depression is the postulation of a cognitive bias or distortion as the fundamental disturbance in the disorder. A rival hypothesis is that the complaints of depressed persons are in some way congruent with or maintained by environmental conditions; that is, depressed people face depressing life circumstances (Brown & Harris, 1978; Coyne, 1976a, 1976b). Here there may be critical discontinuities between mildly depressed college students and clinically depressed individuals. For instance, most college students are unmarried, but a significant proportion of episodes of clinical depression are associated with marital disturbance (Briscoe & Smith, 1973; Coleman & Miller, 1975; Weissman & Paykel, 1974). It is possible that marital turmoil involves environmental events that corroborate the depressives' negative self-perceptions (e.g., hostile accusations by spouses), which are not experienced by mildly depressed college students. Self-derogation and blame would have to be interpreted in light of this. As researchers increasingly turn to clinic and inpatient populations, new issues concerning sampling and the appropriateness of


various control groups must be addressed. Depressed patients may differ from a sample of nondepressed patients on some dependent measure, but the implications of this difference for a cognitive theory of depression depend on assumptions about how each group would differ from a normal, nonclinical group. For example, DeMonbreun and Craighead (1977) found that depressed patients recalled receiving fewer reinforcements than did nondepressed patients; interpretation of this result, however, had to be tempered by the finding that the depressed patients did not differ significantly on this measure from a group of nondepressed non patient controls. Lewinsohn (1974) has argued cogently for the inclusion of a nondepressed patient control group in any study relying on differences between depress!ves and normals. In the absence of this group, researchers cannot rule out the "psychological deviation" hypothesis (depressed group = nondepressed .patient group + normal control group) as an alternative explanation of the differences attributed to depression. Unfortunately, as will be seen presently, few studies have included two control groups. In summary, problems in sampling and selection as well as the lack of appropriate control groups necessitate a qualification of any statements concerning the relationship between cognition and depression that are based on the current empirical literature. In the review that follows, it will be noted when a depressed sample comprised students selected on the basis of elevated BDI scores, when a patient population was sampled, and when at least two control groups were included. Performance Expectations and Evaluations Expectancy changes under skill and chance conditions. Until recently, there was considerable interest in depressed-nondepressed differences in expectancy changes following success and failure on skill and change tasks. Based on the original learned-helplessness model, it was hypothesized that because of their general belief in response-reinforcement independence, depressed subjects would respond to skill tasks as if outcomes were governed by chance, Thus, relative to the expectancies of nondepressed subjects, the expec-

477

tancies of depressed subjects in skill tasks should increase less after success and decrease less after failure. In chance tasks, on the other hand, the expectancy changes of depressed subjects would be similar to those of nondepressed subjects. Initial studies by Seligman and his colleagues (e.g., Abramson, Garber, Edwards, & Seligman, 1978; Klein & Seligman, 1976; Miller & Seligman, 1973, 1976; Miller, Seligman, & Kurlander, 1975) have supported some of these predictions, Other studies, however, have failed to support any of the stated predictions (e.g., Golin & Terrell, 1977; McNitt & Thornton, 1978; O'Leary, Donovan, Krueger, & Cysewski, 1978; Prkachin, Craig, Papageorgis, & Rieth, 1977; Sacco & Hokanson, 1978a; Smolen, 1978; Willis & Blaney, 1978). Overall, differences between depressed and nondepressed subjects with respect to changes in performance expectations have not been as strong or consistent as originally hypothesized. The revised learned-helplessness model took this conclusion into account: "In the absence of knowledge about individual attributions, the reformulated helplessness hypothesis cannot make clear-cut predictions about expectancy changes and helplessness" (Abramson, Seligman, & Teasdale, 1978, p. 63). The reformulation of the model also offered an explanation of how depressed persons might perceive skill tasks as different from chance tasks and still demonstrate reduced expectancy changes following success on a skill task. It was argued that although depressed subjects might perceive skill as relevant to the outside of the task, they also see themselves as lacking the necessary skill, as being personally rather than universally helpless. Consistent with this interpretation, Garber and Hollon (1980) found that although depressed and nondepressed students did not differ in their expectancy changes for how well they expected others to do, the depressed students showed smaller changes for their estimates of success for themselves. It must be noted that the hypothesis that depressed persons see themselves as personally helpless in task situations has been contradicted by repeated failures to find depressionassociated differences in ratings of control (e.g., Abramson, Garber, et al., 1978; Garber & Hollon, 1980; Rizley, 1978; Willis & Blaney,

478


1978), In addressing this issue, a number of writers have suggested that measures of expectancy change may be in some ways reactive to the experimental situation. Mischel (1958) has argued that the overt verbalization of expectancies increases subjects' commitment to the specific values of these expectancies, and therefore reduces change. Dweck and Gilliard (1975), in fact, have demonstrated that expectancies are altered when subjects verbalize them before each trial. Sacco and Hokanson (1978a) found that depressed-nondepressed differences depended on whether their responses were public or private. When expectancy statements were private, depressed, students actually exhibited greater changes in expectancy than did nondepressed students. Langer (1978) has questioned the ecological validity of differences in responses to skill and chance task instructions. Citing her own work on the "illusion of control," Langer suggests that when instructed to attend to chance elements in a situation, people are, in fact, capable of doing so. However, in routine and familiar situations, they do not make chance/skill discriminations. This interpretation fits the results obtained by Golin, Terrell, and Johnson (1977). In a Las Vegas style dice game, a significant Depression X Control interaction was obtained for confidence ratings concerning the outcome of a throw of the dice. Depressed students were significantly more confident when a croupier threw for them, whereas the reverse tendency was found for nondepressed students. Golin et al. observed that "the greater the illusion of control, the less sanguine depressed subjects were about their prospects for success" (p. 441). Golin, Terrell, Weitz, and Drost(1979) replicated these results with depressed and nondepressed psychiatric inpatients, and interpreted this replication as reflecting a general sense of personal incompetence among depressed persons. However, Rothbaum, Weisz, and Snyder (1982) have pointed to some possible complexities in the relationship between the concepts of chance or luck and control; they may not be simple antonyms. Luck can be seen as a disposition or ability to obtain certain outcomes, a relatively enduring feature of the person. People, whether depressed or nondepressed, may judge their possession of luck by their success in obtaining desired out-

comes or avoiding undesirable ones. Alternatively, luck or chance can be seen as an external force, potentially controllable and capable of being won over as an ally. Recently, there has been declining interest in depressed-nondepressed differences under skill and chance conditions. Even when hypothesized results were obtained, their interpretation remained ambiguous. One suggestion that can be garnered from this line of research will be substantiated as we continue our survey of literature concerning the role of cognition in depression: it may be more difficult to develop unambiguous tests of cognitive hypotheses than it first appears. Self-evaluation. Several studies have compared depressed and nondepressed subjects' evaluations of their performance using global rating scales. According to Beck's (1967, 1976) model of depression, depressed individuals perceive and interpret aspects of themselves and their environment to be more negative than is actually the case. Depressed subjects, then, should be expected, in the absence of explicit feedback, to underestimate the quality of their performance. In support of this prediction, a number of investigations have found that depressed psychiatric patients evaluate themselves more negatively than do nondepressed persons in the absence of differences in actual performance (e.g., Friedman, 1964; Lobitz & Post, 1979; Smolen, 1978). However, other investigations examining psychiatric populations have suggested that this pattern of lowered selfevaluation may not be specific to depression. Gotlib (1981, 1982), for example, found both depressed and nondepressed psychiatric inpatients to give themselves fewer reinforcements for their performance than did nondepressed nonpsychiatric controls, and this effect remained when group differences in actual performance were controlled.' 1 There is some question as to the equivalence of selfreinforcement and self-evaluation. Although a number of theorists have discussed self-reinforcement as an independent construct (e.g., Bandura, 1976; Rehm, 1977), empirical investigations have provided evidence of a strong relationship between self-reinforcement and self-evaluation (e.g., Bellack, 1975; Heaton & Duerfeldt, 1973). For the purposes of our discussion, therefore, self-reinforcement and self-evaluation will be considered as comparable constructs.


Studies using college students, too, do not always find depression-associated differences in evaluation. Wollert and Buchwald (1979), for example, found depressed females to evaluate their performance more negatively than 'did nondepressed females. However, a significant correlation between depression and actual performance makes it difficult to interpret these evaluation data unambiguously. Zarantonello, Johnson, and Petzel (1979) found depressed students to evaluate their performance more negatively than did nondepressed students only when the task was difficult and ego-involving; depressed-nondepressed evaluation differences were absent in the other experimental conditions. Finally, Ciminero and Steingarten (1978) found that following equivalent performances on a Digit Symbol test, depressed and nondepressed female students did not differ significantly from each other with respect to their self-evaluations. Although it did not involve laboratory task behavior, a study by Lewinsohn, Mischel, Chaplin, and Barton (1980) offered provocative data concerning depression-associated differences in self-evaluation of social skills. Subjects' self-ratings were compared with ratings obtained from judges who had observed them in group and dyadic interactions. Depressed outpatients rated themselves significantly more negatively than did nondepressed outpatients and nondepressed psychiatric controls. The depressives' self-ratings, however, were concordant with the ratings given them by the observers. In contrast, subjects in both nondepressed groups rated themselves significantly more positively than they were rated by the observers, a phenomenon conceptually similar to the "self-serving bias" observed in studies of attributional processes. Lewinsohn et al. (1980) conclude, and others have cited their study as indicating, that depressed persons may be realistic in their perceptions, whereas nondepressed persons bask in the "warm glow" of self-enhancing distortions. Recent evidence is challenging this conclusion. Strack and Goyne (1983), for example, failed to find significant discrepancies between the self-presentational strategies of either depressed or nondepressed students and the evaluations of them by persons with whom they interacted. Gotlib and Meltzer (Note 1) found that observer ratings of the social skill

479

of depressed college students engaged in dyadic interactions were significantly lower than both the students' self-ratings and ratings of them by their partners, which did not differ significantly from each other. Similarly, Kahn, Coyne, and Margolin (Note 2) found that, compared with controls, both depressed persons and their spouses significantly overestimated the negativity of the evaluations they received from each other. Finally, Arkowitz, Holliday, and Hunter (Note 3) found that depressed women underestimated their husbands' hostility toward them. Presumably, if these authors followed the logic of Lewinsohn et al. (1980), they would argue that it is depressed persons who have an illusory glow to their self-evaluations. However, as Kahn et al. point out, work done in the early 1950s showed that it is difficult to give substantive interpretations to discrepancies between self-perception and perception by others. Classic analyses by Cronbach (1955; Gage & Cronbach, 1955) indicated that apparently significant discrepancies are often reducible to statistical and methodological artifacts. The data for summary evaluations of self and of performance provide a crucial part of the data concerning the role of cognition in depression. Depressed patients and students tend to evaluate their performances and themselves more negatively than do nondepressed persons. However, there does not seem to be a firm basis for labeling this tendency of depressed persons as a distortion. Mischel (1973) has argued that self-evaluation may be more related to initial expectancies than to performances. It could be that the negative self-evaluations of depressed persons in specific task situations are based on expectancies that are congruent with and derived from their extralaboratory experiences. Although current methodologies provide for a demonstration of the tendency of depressed persons to make negative self-evaluations, they offer little clarification of the significance of this observation. Finally, person variables other than depression have been demonstrated to be associated with low self-evaluation. For example, in the absence of differences in observer ratings, nonassertive persons evaluate their social behavior less positively than do assertive persons (Alden & Cappi, 1981), infrequent daters less positively than frequent daters (Glasgow &

480


Arkowitz, 1975), and high socially anxious men less positively than low socially anxious men (Clark & Arkowitz, 1975). Further, in the absence of performance differences, high-testanxious females evaluate their anagram performance less positively than do low-test-anxious females (Holroyd, Westbrook, Wolf, & Badhorn, 1978). Perception of Information Several investigations have examined differences between depressed and nondepressed persons with respect to their perceptions of environmental stimuli. These studies have been conducted as tests both of Beck's and the learned-helplessness models of depression. Beck (1967, 1976) has postulated that the "cognitive triad" affects depressives' interactions with their environment, causing them to distort feedback in a negative manner. Thus, depressed individuals are posited both to selectively filter out positive information and to perceive negative or neutral information as being more negative than is actually the case. Studies assessing this aspect of Beck's model, consequently, have compared depressed and nondepressed subjects' perception of positive, neutral, and negative stimuli. The learned-helplessness model of depression also predicts inaccuracies in the perceptions of depressed persons, but for different reasons. This model maintains that depressed individuals are characterized by the belief that their responses are ineffective in bringing about a desired outcome. This belief is posited to interfere with contingency learning. Thus, according to the learned-helplessness model, depressed persons fail to accurately perceive response-outcome dependence when consequences actually are contingent on their responses. Investigations assessing distortions in the perception of information from the perspective of the learned-helplessness model, therefore, have examined judgments offered by depressed and nondepressed subjects concerning their perceived control over various outcomes in experimental tasks. We will examine in turn the studies addressing Beck's model of depression and those assessing predictions derived from the learned-helplessness model.

Immediate perception of feedback. Two studies have been reported that have .examined depression-associated differences in the immediate perception of specific instances of valenced feedback. DeMonbreun and Craighead (1977) presented depressed and nondepressed psychiatric outpatients and nondepressed nonpsychiatric controls with positive, neutral, and negative feedback on each trial of an ambiguous laboratory task. Five achromatic slides, ranging from light gray to black, were used to provide feedback, and subjects were required to report their perception of each feedback slide immediately following its presentation. Subjects were led to believe that there were only four feedback slides (100% and 75% acceptable, and 100% and 75% unacceptable), and thus were forced to report perceiving the middle-density slide as either positive or negative. Contrary to predictions derived from Beck's model, DeMonbreun and Craighead found no group differences with respect to immediate perception of either the positive or the neutral feedback. This finding was replicated by Craighead, Hickey, and DeMonbreun (1979), who presented depressed-anxious, nondepressed-anxious, and nondepressed-nonanxious university students with a task and feedback similar to that used by DeMonbreun and Craighead (1977). Again, no group differences were found with respect to immediate perception of feedback, and Craighead et al. concluded that, contrary to Beck's formulation, depressed persons do not demonstrate distorted perceptions of environmental feedback. The issue of depression-associated distortion of the environment was addressed in a different manner in a study conducted by Hoehn-Hyde, Schlottmann, and Rush (1982). Female depressed psychiatric patients, remitted-depressed subjects, and nondepressed nonpsychiatric controls were shown videotapes of positive, negative, and neutral social interactions. Half of the subjects in each group were instructed to imagine the interactions being directed toward themselves, and half toward another person. Subjects then rated the evaluative nature of the interactions. When these ratings were analyzed, Hoehn-Hyde et al. found neither the predicted significant main effect for diagnostic group nor a significant


Group X Self-Other interaction. All subjects were found to evaluate interactions in the self condition more positively, and the positive interactions were rated more positively than were the neutral or the negative interactions. Thus, depressed patients did not differ in this study from either remitted depressives or nondepressed controls with respect to their perception of the evaluative nature of environmental stimuli. In a similar study, Roth and Rehm (1980) trained depressed and nondepressed psychiatric inpatients to identify particular nonverbal behaviors as positive and others as negative, and asked the patients to count the occurrences of these behaviors in videotapes of their previous interactions with the experimenter. Depressed patients counted fewer positive behaviors than did raters observing the same tapes, but were accurate with respect to negative behaviors. Nondepressed patients, on the other hand, counted both fewer positive and fewer negative behaviors than did the raters. Although Roth and Rehm discuss their results as consistent with the hypothesis of a cognitive distortion associated with depression, it is important to note that all of the patients did poorly in the self-monitoring task; the nondepressed patients counted half as many negative behaviors and the depressed patients half as many positive behaviors as did the raters. In sum, studies of the immediate perception of environmental information are few in number, and they do not provide strong support for the hypothesis that depressed persons are characterized by biased or distorted initial perceptions. In fact, as will be seen in the following section, there is evidence to suggest that in some situations depressed individuals appear to be accurate in their perceptions, whereas nondepressed persons demonstrate a positive bias. Further research involving more complex stimuli and judgments is clearly required before a comprehensive understanding of the parameters surrounding depressive and nondepressive biases in perceptions of environmental stimuli is attained. Judgment of contingency. The learnedhelplessness model of depression postulates that the expectation of depressed individuals (that their responses do not control outcomes) biases them to underestimate the degree of

481

control they actually exert over an outcome in any particular situation. Alloy and Abrarnson (1979) examined this prediction in a series of four studies assessing depression-associated differences in judgments of contingency. Depressed and nondepressed students were presented with problems that varied in the actual degree of control that the subjects' responses had over outcomes. In each problem, subjects were required to estimate the degree of contingency between their responses and an observable outcome. Alloy and Abramson found that whereas depressed students detected rel-. atively accurately the degree of contingency between their responses and the outcomes in all problems, nondepressed students overestimated the degree of control they had over frequent or successful outcomes, and underestimated their control over failure outcomes. In a similar study, Abramson, Alloy, and Rosoff (1981) presented depressed and nondepressed students with a contingency learning problem in which the response-outcome contingency was set at 75%. Half of the subjects were required to develop their own hypotheses concerning the contingency, and half were provided with a small set of hypotheses, including the correct one. Results indicated that, relative to nondepressed students, depressed students in the "self-generated hypothesis" condition were less likely to perform the correct controlling response, and judged that they exerted less control over the outcome. The authors noted that although depressed students' judgments were an underestimate of the control they could have potentially exerted, they were actual reflections of the amount of control they actually exerted. Considered together, the results of both the Abramson et al. and the Alloy and Abramson (1979) studies provide no support for predictions derived from the either Beck's or the learned-helplessness model of depression. To the contrary, by identifying a type of situation in which depressed students are accurate and nondepressed students show a "cognitive distortion," these studies contradict both models. As a cautionary note, it is important to observe that these studies suggest some theoretically important depression-associated differences when subjects are explicitly instructed to attend to the degree of contingency in a labo-

482

JAMES C, COYNE AND IAN H. GOTLIB

ratory task situation. Questions still remain concerning the generalizability of these findings to situations in which such instruction or priming is not provided. Is "degree of contingency" a working concept of subjects in extralaboratory settings or even in other laboratory task situations? A sizable body of literature consistently indicates that people have difficulty using the concept of correlation, that they do not readily make use of notions of contingency versus noncontingency, and that expectations based on a priori theories attenuate or even refute the influence of data that are not concordant with these expectations (see Tversky & Kahneman, 1974, and Nisbett & Ross, 1980, for extensive reviews of this literature).

expected to recall a greater proportion of negative to positive experiences, and further, to recall the negative experiences more quickly than the positive. These two types of studies will be considered in turn. Recall of feedback. In a number of studies, depressed and nondepressed subjects were provided with predetermined feedback concerning their performance on a laboratory task, and their subsequent estimates of the frequency or percentage of positive and negative feedback they received were assessed. Wener and Rehm (1975), for example, provided depressed and nondepressed students with either 20% or 80% success feedback on a word association task, and found that depressed students estimated that they had been correct less often than did nondepressed students. Both groups of subjects, however, were generally inRecall of Information accurate in their estimates, with a considerable Various investigations have examined dif- number of subjects reporting receiving a 50% ferences between depressed and nondepressed rate of success. Using the same word associpersons with respect to their differential recall ation task, Kuiper (1978) gave depressed and of positive and negative events. These studies nondepressed students 20%, 55%, or 80% sucare conducted largely as tests of Beck's (1967, cess feedback. Although the depressed students 1976) cognitive model of depression. As out- generally tended to estimate fewer items corlined earlier, Beck has postulated that the de- rect than did the nondepressed students, this pressives' negative schema affects perception group difference was statistically significant and interpretation of environmental stimuli. only in the 55% success condition. Further, the depressive engages in a number Both Wener and Rehm (1975) and Kuiper of forms of cognitive distortion and conse- (1978), then, report group differences between quently is postulated to perceive and recall depressed and nondepressed subjects with reinformation from the environment to be more spect to the amount of positive and negative feedback recalled. It is important to note, negative than is actually the case. Typically, investigations of recall take one however, that these results do not adequately of two forms. In some studies, depressed and assess the strong predictions derived from nondepressed subjects participate in a labor- Beck's model. Again, Beck postulates that detory task during which they receive positive pressed persons underestimate the frequency and/or negative feedback. The subjects are of positive feedback and overestimate the then asked to recall the amount of positive amount of negative feedback received. A and negative feedback they received during proper test of these hypotheses, therefore, rethe task. Support for Beck's model would come quires an examination of the difference befrom the findings that, on recall, depressed tween the depressed person's recall of feedback subjects underestimate the amount of positive and the feedback itself. Differences between feedback and overestimate the quantity of depressed and nondepressed subjects in the negative feedback given them. Other investi- recall of feedback without explicit comparison gations require depressed and nondepressed to the actual quantity of feedback given do subjects to recall previously experienced ex- not address these predictions. tralaboratory positive and negative events. The A number of recent investigations have prodifferential speed of recall of pleasant and un- vided more adequate tests of Beck's predicpleasant experiences and/or the number of tions. Nelson and Craighead (1977), for exeach type of experience recalled is then as- ample, presented depressed and nondepressed sessed. In these studies, depressed subjects are students with a task on which they were either


rewarded or punished on either 30% or 70% of the trials. These investigators found that in the 70% reward condition, the depressed students recalled receiving a lower rate of reward than was actually the case. In addition, the depressed students who were punished at the 30% schedule recalled receiving a higher rate of punishment than did their nondepressed counterparts. This difference, however, was due to the nondepressed students' underestimating the amount of punishment they had received; the depressed students in this condition were found to be accurate in their recall. No group differences in recall were obtained in either the 30% or the 70% punishment conditions. In a similar study, DeMonbreun and Craighead (1977) assessed recall of feedback in two sets of trials in depressed and nondepressed psychiatric patients and nondepressed nonpatient controls. DeMonbreun and Craighead fotind that in the first set of trials (essentially a 50% correct condition), although the depressed patients made significantly lower estimates of their task performance than did the nondepressed patients, they did not differ in their estimates from the nonpatient controls. Furthermore, none of the three groups was found to differ significantly in their estimates from the feedback actually provided. For the second set of trials, subjects received either a high or low rate of positive feedback. In the high feedback condition, the depressed subjects underestimated their feedback, whereas subjects in the two nondepressed groups were relatively accurate in their recall. Finally, although no significant group differences were obtained in the low feedback condition, a significant main effect for feedback condition indicated that all subjects underestimated high rates of positive feedback and overestimated low rates. In their four judgment-of-contingency studies discussed in the previous section, Alloy and Abramsbn (1979) also obtained subjects' recall of the frequency of the reinforcement (onset of a green light) they had received. In only one of the studies did the difference between the depressed and nondepressed subjects on this measure approach statistical significance. Subjects in general were relatively accurate in recalling the frequency with which they were positively reinforced by the light alone, but when the onset of the light was

483

associated with winning money, they tended to overestimate its frequency. Gotlib (in press) has noted that the studies examining recall of feedback in depression have typically used impersonal laboratory tasks to generate the feedback, and has suggested that the inconsistent findings obtained in these studies may be a function of this task parameter. Gotlib gave depressed and nondepressed psychiatric patients and nondepressed nonpsychiatric controls evaluative interpersonal feedback ostensibly based on an observation of their prior dyadic interaction, and assessed subjects' memory of the feedback in a subsequent incidental recall paradigm. The depressed patients were found to demonstrate a negative bias, recalling the feedback to be significantly more negative than was actually the case. In contrast, the nondepressed psychiatric patients and the nondepressed nonpsychiatric controls were found to be relatively accurate in their recall of the feedback. It appears, then, that Beck's predictions concerning depressive distortion in the recall of information from the environment have received equivocal support. Moreover, it is important to note that in all of these investigations, feedback was not a function of the quality of the subjects' performance, but rather was predetermined by the experimenter. Although this procedure standardizes the amount of success or failure feedback provided, it may introduce a number of other problems. First, the feedback may be discrepant with the subjects' own observations of their task performance and their extralaboratory experience, and this discrepancy may influence their estimates. Experimenters have not typically assessed the credibility of their manipulations, and the general inaccuracy of both depressed and nondepressed subjects in these studies may reflect, in part, reactions to what is perceived as anomalous feedback. Second, as noted in the previous section of this article, differences between depressed and nondepressed subjects in the perception of contingency are an important component of the learned-helplessness model. Therefore, it is noteworthy that noncontingency between performance and outcome is also an integral part of a manipulation in studies designed to explore other depressed-nondepressed differences in cognitive functioning.

484


In all of the studies reviewed thus far, feedback was either predetermined by the experimenter or otherwise not tied directly to subjects' actual performance. Only a few studies have examined subjects' estimates when feedback was directly related to their actual performance. Buchwald (1977) rewarded depressed and nondepressed students for correct responses on a Thorndikian learning task. Buchwald found that both depressed and nondepressed students tended to underestimate the number of rewards they received. Part correlations between level of depression and recall of rewards indicated that subjects with higher levels of depression tended to underestimate the number of correct responses more, whereas less depressed subjects tended to underestimate less or to overestimate. However, the subjects' estimates of the number of rewarded responses were found to correlate more highly with actual performance than with mood. Gotlib (1981) examined subjects' estimates of how frequently they provided themselves with "rewards" and "punishments" during performance on a nonsense syllable recognition task. Depressed and nondepressed psychiatric inpatients and nondepressed nonpatient controls were given an opportunity to press a reward or a penalty button after each trial in order to indicate their evaluation of their performance, and were then required, in an incidental recall paradigm, to estimate the number of rewards and punishments they had given themselves. Gotlib found that whereas subjects in the two nondepressed groups were relatively accurate in their recall of both rewards and punishments, the depressed patients recalled administering fewer self-rewards and a greater number of self-punishments than was actually the case. Although studies of overall estimates of feedback and performance are cited as the key evidence that cognitive distortions are associated with depression, interpretation of these data should take into account the recurring rinding that both depressed and nondepressed subjects are generally inaccurate in their estimates of how often they have been correct or incorrect in a series of trials on a task. Explanations of the reports of depressed persons, therefore, cannot assume that nondepressed persons veridically report their performance or that depressed-nondepressed differences in themselves represent a cognitive

distortion on the part of the depressed persons. It is' likely that subjects do not routinely keep a running tally of their performance scores, and when queried, they must guess. In doing so, they may draw upon salient, extralaboratory experiences or upon some sense of how they generally perform. Any theoretical explanation of the estimates of depressed persons must start with assumptions about how nondepressed persons arrive at estimates of their performance. One may invoke the concept of schema in accounts of these processes, but complexities and inconsistencies in the data require more than a simple statement that depressed persons distort their experiences in terms of prepotent depressive schemata. Recall of positive and negative experiences. One of the earliest examples of research examining recall of positive and negative experiences in depression was reported by Beck and his colleagues (Beck, 1961; Beck & Hurvich, 1959; Beck & Ward, 1961), who examined the dreams, memories, and fantasies of depressed and nondepressed psychiatric patients. Beck found that depressed patients tended to report dreams in which they were the recipient of painful experiences such as being disappointed, rejected, or injured. Depression scores were found to be correlated with masochistic content both in reports of most-recent dreams and early memories, and in focused fantasies elicited by projective stimuli. Beck postulated that depressed persons had a "need to surfer," but recognized that the unpleasant material reported by depressed patients might simply be a reflection of their negative subjective state. Beck subsequently discarded his motivational hypothesis, and instead argued that this negative content reflected the action of idiosyncratic cognitive schemata that became prepotent during depression (Beck, 1967). More recent work has challenged this "concomitant hypothesis" by suggesting that patients who have experienced remittance of serious reactive depressions, who are no longer demonstrating overt depressive symptomatology, continue to report more "masochism" and hostility from the environment in their dreams than do nondepressed controls (Hauri, 1976). It appears, therefore, that negative dream content is not related in any simple manner to depressive symptomatology, and more work in this area is clearly required before a reasonable alternative relationship can be articulated.


Other investigations have required depressed and nondepressed persons to recall information about themselves and their environment. Lishman (1972) asked depressed and nondepressed psychiatric patients to rate 18 preselected topics with respect to "feeling tone," and assessed recall of the topics 2 weeks later. Depressed patients were found to recall a higher proportion of negative to positive topics than did nondepressed patients. Conceptually similar results have been reported by Lloyd and Lishman (1975) and by Teasdale and his colleagues (e.g., Clark & Teasdale, 1982; Teasdale&Fogarty, 1979; Teasdale & Taylor, 1981), both with depressed patients and with normal subjects who have undergone a mood-induction procedure, The results of these studies suggest that mood differentially affects the relative accessibility of negative and positive cognitions. Although there are difficulties surrounding the use of mood-induction techniques (cf. Buchwald, Strack, & Coyne, 1981; Polivy & Doyle, 1980), this line of research appears to offer a promising approach to the study of the relationship between cognition and affect in depressed individuals. A number of recent studies have used methods developed in experimental memory research to test Beck's hypotheses concerning depressives' tendency to negatively bias the processing of personal information. Davis (1979a, 1979b; Davis & Unruh, 1981) used an incidental recall task developed by Rogers, Kuiper, and Kirker (1977) with depressed outpatients and nondepressed normal controls. Subjects rated adjectives with respect to their structural, phonemic, semantic, or self-referent qualities. Subjects were not initially instructed that recall was to be tested, but immediately after the ratings were made, subjects were asked to recall as many of the adjectives as possible. Depressed-nondepressed differences in recall were obtained only for adjectives that had been rated "yes" for self-reference, with depressed patients recalling significantly fewer adjectives than did nondepressed subjects. Furthermore, multiple regression analyses performed on data from the depressed subjects revealed that the ratio of words recalled from the self-reference task to those recalled from the semantic task was significantly related to duration but not to severity of depression. On the basis of these and other findings with student samples, Davis argued that the

485

self-schema of depressed persons is initially weakened by the change in self-referents accompanying the onset of depression, but that over time, the self-schema regains its strength in information-processing as it reorganizes. Derry and Kuiper (1981) have challenged this formulation, noting that Davis collapsed across the positive, neutral, and negative adjectives used as stimuli in his studies. Given the assumed negative self-image of depressed persons, the appropriate test for schema-based processing should examine recall of depressive adjectives. Derry and Kuiper made the "content-specific" prediction that depressed persons should demonstrate superior recall for depressive adjectives, whereas nondepressed persons should show superior recall for nondepressed adjectives. Consistent with this prediction, Derry and Kuiper found that compared with nondepressed psychiatric patients and nondepressed nonpatient controls, depressed psychiatric patients were superior in their recall of depressive adjectives that they had endorsed as self-referent. Subjects in the two nondepressed groups demonstrated significantly higher recall for nondepressed selfreferent adjectives. Using a similar methodology, Roth and Rehm (1980) asked depressed and nondepressed male psychiatric inpatients to rate 20 positive and 20 negative adjectives with respect to how characteristic they were of themselves. After a 10-minute filler task, both recall and recognition memory for the words was assessed. In contrast to Derry and Kuiper's (1981) results, Roth and Rehm found no significant main effects or interactions for either analysis, and concluded that "negative memory distortion by depressed subjects was not evident" (p. 154). Kuiper and MacDonald (1982) examined depressed and nondepressed students' ratings of adjectives applying to themselves, the average person, or someone they knew well. The depressed students recalled equal numbers of positive and negative self-referent adjectives, whereas the nondepressed group recalled more positive than negative adjectives. In addition, for the negative adjectives the depressed students recalled more self- than other-referent adjectives, whereas the opposite pattern of recall was obtained with the nondepressed students. Although these results are consistent with Kuiper and MacDonald's hypotheses,

486


these authors also reported that both groups of subjects recalled a greater number of positive self- than other-referent adjectives, a finding that seems at odds with a depressive selfschema interpretation of the depressed subjects' recall. The work of Kuiper and Davis and their coworkers represents a creative approach to the study of the role of a negative self-schema in depression. However, Derry and Kuiper (1981) note important differences between their operationalization of schema in terms of precisely defined observations of task behavior, and Beck's use of schema in a broad sense to describe pervasive strategies for molding information into attitudes or assumptions. Derry and Kuiper are thus providing only a limited test of Beck's model of depression. Furthermore, given the indications that test-anxious, socially anxious, unassertive, and externally oriented persons have negative tendencies in their processing of information, more attention clearly needs to be devoted to the question of whether the patterns observed by Davis and Kuiper are specific to depression. Dysfunctional Cognitions: Measures of Cognitive Biases and Automatic Thoughts A number of investigations have been conducted assessing the relationship between depressed mood and various of the negative cognitions described by Beck (1967, 1976). In an early study, Weintraub, Segal, and Beck (1974) examined the correlations between a mood checklist and a story completion test administered to male students five times in a 2-month period. The responses to the story completion task were scored in terms of expectations of discomfort and failure, and negative perceptions both of the self and of interpersonal relations. Weintraub et al. found significant correlations between depressed mood and negative cognitions for four of the five testing sessions. Only 5 of the 20 time-lagged crosscorrelations between mood and cognition attained significance, however, and it appears that negative cognitions are a stronger concomitant of depressed mood than they are an antecedent or consequence. Krantz and Hammen (1979) developed a cognitive bias questionnaire (CBQ) designed to test Beck's hypotheses concerning negative cognitive processes in depression. The CBQ

requires subjects to imagine themselves in a number of specific problematic situations that could arise in college life. Subjects are asked to indicate, by responding to multiple-choice questions, what they would think, feel, and do in each situation. The four response options for each situation cross the two dimensions of depressed versus nondepressed and distorted versus nondistorted, yielding depressed-distorted, depressed-nondistorted, nondepressed-distorted, and nondepressed-nondistorted alternatives. Although an attempt was made to provide instances of such cognitive errors as arbitrary inference, selective abstraction, and overgeneralization, Krantz and Hammen reported that independent judges were unable to distinguish reliably among these categories. Results from a number of student and inpatient depressed and nondepressed samples suggest that a CBQ subscale constructed from depressed-distorted responses discriminates between these two groups of subjects. Hammen (1978), for example, compared depressed and nondepressed students on the CBQ who were either low or high in recent life stress. Depressed students gave more depressed-distorted responses, but an interaction effect revealed that depressed students with low life stress showed the greatest distortion. Blaney, Behar, and Head (1980) found that in two student samples, all of the CBQ subscales discriminated significantly between depressed and nondepressed groups. Correlations between the BDI and the depressed-distorted subscale were .26 and .44 in the two samples. Although depressed-distorted responses on the CBQ are significantly correlated with BDI scores, this type of response is relatively infrequent, and depressed-nondepressed differences are small in an absolute sense. For example, the CBQ allows for a possible score of 23. In Krantz and Hammen's (1979) first student sample, the mean score for depressed students was 2.15 and for nondepressed students, 1.56; in the second sample, group means were 3.02 and 1.71, respectively. Although responses may indeed represent cognitive distortion (and the interesting Hammen, 1978, results suggest this), cautions are in order. The hypothetical situation used as a sample in the Krantz and Hammen study involves the subject's losing the presidential election in some organization. The depressed-distorted response to the state-


487

ment "When you first heard you'd lost, you tained higher cognitive error scores than did immediately . . ." is "feel bad and imagine nondepressed subjects, but within the deI've lost by a landslide." It is conceivable that pressed groups the correlation between the an observer familiar with a given depressed BDI and the CEQ was only .39 (recall the person's life situation would judge this to be same BDI-distortion correlation reported by a realistic response. Finally, as with almost Krantz and Hammen, 1979). Lefebvre found every measure or finding in the current lit- that depressed subjects endorsed "personalerature, questions remain as to the specificity ization" less strongly than the other three cogof cognitive distortion to depression. The cor- nitive errors. Personalization was defined by relation of the CBQ depressed-distorted sub- Lefebvre as "taking personal responsibility for scale with repression-sensitization (r = .31), negative events" (p. 518). Given the imporfor example, approaches its correlation with tance accorded this concept by the learnedthe BDI (r = .39; Krantz & Hammen, 1979). helplessness model, it is noteworthy that perThe subscale was also correlated with the factor sonalization was the least strongly endorsed scores from the Profile of Mood States cognition in Lefebvre's study. Finally, Lewinsohn, Steinmetz, Larsen, and (McNair, Lorr, & Droppelman, 1971), and although the correlation between the subscale Franklin (1981) reported a longitudinal study and the depression factor (r = .33) was highest, of the relationship between cognition and all six measures of negative affect were sig- depression in a large (N = 998) community sample. Subjects first (Tl) completed a number nificantly correlated with distortion. Hollon and Kendall (1980) reported the de- of questionnaire measures of cognitions, invelopment of an automatic thoughts ques- cluding causal attributions, expectancies of tionnaire (ATQ) designed to measure the fre- positive and negative outcomes, irrational bequency of "automatic negative thoughts" as- liefs, perception of control, and self-esteem, sociated with depression. Subjects were asked and were interviewed 8 months later (T2). Perto indicate how frequently specific thoughts sons categorized as depressed at Tl differed occurred to them in the past week. The ATQ from nondepressed persons in the expected distinguished between depressed and nonde- direction on the measures of expectancies for pressed college students, and was highly cor- positive and negative outcomes, irrational berelated with the BDI and the Depression scale liefs, and self-esteem. Lewinsohn et al., howof the Minnesota Multiphasic Personality In- ever, cautioned that ventory (MMPI; see also Dobson & Breiter, in spite of the fact that these (negative cognitions were 1983). The ATQ was even more strongly cor- found to be correlated with being depressed, persons who related, however, with a trait anxiety measure, became depressed during the course of the study were not again raising the issue of whether such cog- characterized at Tl by many of the patterns of negative nitions are specific to depression. The question thinking of the type postulated by the cognitive theorists. wit: Prior to becoming depressed, these future dealso remains as to whether it is reasonable to To pressives did not subscribe to irrational beliefs, they did expect that subjects can reliably estimate the not have lower expectancies for positive outcomes or higher frequency of thoughts for a period of a week. expectancies for negative outcomes, they did not attribute Consistent with this concern, Beck (1976) success experiences to external causes and failure expenoted that depressed patients do not report riences to internal causes, nor did they perceive themselves as having less control over the events in their lives, (p. 218) automatic negative thoughts in their free associations, and suggested that specific training On the basis of these and other findings, the is required for them to do so. authors concluded that negative cognitions are Lefebvre (1981) developed the Cognitive neither antecedents nor sequelae of depression Error Questionnaire (CEQ), which assesses but suggested, rather, that such cognitions four types of cognitive errors identified by Beck make recovery from a depressive episode more et al, (1979): catastrophizing, overgeneraliza- difficult. tion, personalization, and selective abstraction. Lefebvre examined differential responses to Dysfunctional Cognitions: Measures of the CEQ by depressed and nondepressed low Attributional Processes back pain patients, depressed psychiatric inpatients, and nondepressed University of VerAlthough studies of attributional processes mont faculty and staff. Depressed subjects ob- have been cited in support of both Beck's and

488


the learned-helplessness models of depression, they are more theoretically bound to the latter. The learned-helplessness model postulates that depressed individuals attribute failure to internal, stable, and global factors, and success to external, unstable, and specific causes. A number of investigations have addressed these predictions, examining the attributions of both depressed and nondepressed subjects in a variety of situations. For purposes of discussion, these investigations can be divided into three major categories: those studies that have examined attributions following success and failure on laboratory tasks, those that have assessed attributions for hypothetical events, and those that have examined attributions for actual stressful life events. Studies in each of these three categories are presented in Table 1. Attributions for success and failure. A number of investigations have examined depressed-nondepressed differences in attributions for experimenter-controlled success and failure. Typically, subjects perform on a laboratory task for which success or failure is manipulated by the difficulty level of the task, false norms, or bogus outcomes. Following the task, subjects are asked to indicate on rating scales the perceived determinants of their performance. The earlier studies in this area (designed before the appearance of the reformulation in 1978) were based largely on the internal-external locus of control construct described by Rotter (1966), and on the four attributional causes of success and failure (effort, luck, ability, and task difficulty) outlined by Weiner and his colleagues (e.g., Weiner, Heckliausen, Meyer, & Cook, 1972). Although there are difficulties in translating Rotter's and Weiner's original constructs directly into the internal-external and stable-unstable dimensions outlined in the reformulation (cf. Miller & Seligman, 1982), the findings from these early studies do not appear to be remarkably discrepant from those of later investigations. All of the studies in this area, therefore, will be considered collectively. As can be seen from Table 1, a rather consistent finding of these investigations is that depressed subjects make more internal attributions for failure than do nondepressed subjects (e.g., Kuiper, 1978; Rizley, 1978; Zemore &Johansen, 1980;Zuroff, 1981). These studies

have typically used university students as subjects; the two investigations that examined attributions following success and failure in patient populations both failed to find the predicted group differences (Abramson, Garber, et al, 1978; Gotlib & Olson, in press). As is also evident from Table 1, group differences in internal attributions for success are either nonsignificant, or in one case, in the opposite direction of that predicted (Zuroff, 1981). When one examines the results of these studies for attributional dimensions other than internal-external, it is apparent that the remaining predictions of the learned-helplessness model have tended not to be supported. Failures to obtain significant depressed-nondepressed group differences have been reported for attributions to stability (Gotlib & Olson, in press; Kuiper, 1978; Rizley, 1978; Zuroff, 1981), for attributions to effort, luck, and task difficulty (Willis & Blaney, 1978; Zemore & Johansen, 1980), and for ratings of perceived control (Abramson, Seligman, & Teasdale, 1978; Garber & Hollon, 1980; Rizley, 1978, Experiment 1; Willis & Blaney, 1978, Experiment 3). Taken together, studies of attributions for success and failure support the hypothesis that depressed persons are more internal than are nondepressed persons in their attributions for failure; little support is found for any of the other predictions derived from the learned-helplessness model. Attributions for hypothetical events. Several studies have examined the attributions of depressed and nondepressed individuals for hypothetical good and bad events. As seen in Table 1, most of these studies have used some version of the Attributional Style Questionnaire (ASQ) developed by Seligman et al. (1979) to provide subjects with the hypothetical events. This scale provides descriptions of six situations with good and six situations with bad outcomes, for each of which subjects are asked to name a major cause and then to rate the cause on 7-point scales assessing internalexternal, stable-unstable, and global-specific attributions. Initial studies using the ASQ yielded promising results. Seligman et al. (1979), for example, reported that, as predicted by the learned-helplessness model, depressed students made more internal, stable, and global attributions for bad outcomes than did nonde-


pressed students; for good outcomes, the two groups of students differed only with respect to the number of attributions made to stable causes. Similar consistent findings in a patient population were reported by Raps, Peterson, Reinhard, Abramson, and Seligman (1982). In a study designed to examine the interaction of attributional style and negative events in predicting depression, Metalsky, Abramson, Seligman, Semmel, and Peterson (1982) found that in a sample of university students, internal and global attributions for negative events on the ASQ were significantly correlated with an increase in depressed mood following receipt of a low grade on a midterm examination. Interestingly, the largest (positive) correlation obtained in this study was that between stability attributions for negative events on the ASQ and increase in depressed mood following receipt of a high grade on the midterm examination. Several other investigations using the ASQ, however, have obtained noticeably weaker results. Blaney et.al. (1980), for example, state that Though most correlations are significant, most are unimpressive in absolute terms. Particularly disappointing is the relatively poor showing for the ... (ASQ) bad internality, stability, and total scores, each of which is appreciably lower than that reported by Seligman et al. (1979). (p. 679)

Similar sentiments have been expressed by Golin, Sweeney and Shaeffer (1981): Attribution subscale scores and the composite score for bad outcomes were significantly correlated with the BDI for both testing sessions; the correlations, however, were small in magnitude and less than those reported by Seligman et al. (1979). (p. 16)

Still other studies have failed to find the predicted depressed-nondepressed differences on any attributional dimension of responses to the ASQ. Manly, McMahon, Bradley and Davidson (1982) found no significant correlations between attributions and concurrent level of depression in a sample of pregnant women. Miller, Klee, and Norman (1982), too, failed to obtain ASQ attribution differences between depressed and nondepressed psychiatric inpatients. Finally, Lewinsohn et al. (1981) found that, none of the attribution scores from the Multi-dimensional Multi-attributional Causality Scale (Lefcourt, von Baeyer, Ware, &Cox, 1979) discriminated be-

489

tween depressed and nondepressed community volunteers. Two studies have been reported that have distinguished between two types of internal attributions: characterological and behavioral self-blame. Characterological self-blame refers to attributional blame directed at one's character or stable personality characteristics, whereas behavioral self-blame refers to an attributional style in which blame is directed at one's own behavior. Janoff-Bulman (1979) asked depressed and nondepressed female students to make attributions to each of four negative events. Surprisingly, depressed students were found to be more likely than nondepressed students to attribute negative outcomes to chance, the variable that discriminated most highly between the two groups of subjects. The depressed students did not exhibit a greater amount of either self-blame in general or behavioral self-blame than did the nondepressed students; the depressed students, however, did evidence more characterological self-blame. Peterson, Schwartz, and Seligman (1981) followed up Janoff-Bulman's study, examining the relationship between the BDI scores of female students and their responses to a modified ASQ questionnaire. In contrast to Janoff-Bulman's findings, Peterson et al. reported negative correlations between level of depression and behavioral self-blame attributions for negative events, but did find a significant positive correlation between BDI scores and characterological attributions for negative events. Several authors have pointed out that although correlations between depression and concurrent attributional scores are relevant to the claims of the learned-helplessness model, it is more critical to examine whether tendencies to make particular kinds of attributions constitute a source of vulnerability to subsequent depression. A number of studies have addressed this issue. Golin et al. (1981), for example, found that in a sample of university students stable and global attributions for negative events were related to depressed mood 1 month later. However, there was no support for the hypothesis that internal attributions for bad outcomes is a causal factor in depression, and even the statistically significant cross-lagged correlations for stability and globality attributions accounted for only 10% (Text continues on page 493)

Table 1 Studies of Attributions Study

Subject population

Target of attributions

Type of attributions

Results

Attributions for laboratory success and failure Abramson, Seligman, &Teasdale 1978

D nonschizophrenic psychiatric patients, D and ND schizophrenic patients, and ND nonpsychiatric controls

Success and failure on skill and chance tasks

Internal, control

No group differences

Kuiper, 1978

D and ND female students

Success and failure on a word association task

Internal, stable

Failure: D more internal than ND; no group differences for stability" Success: no group differences for internality" or stability"

Rizley, 1978 Experiment 1

D and ND students

Success and failure on a number-guessing task

Internal, stable, control

Failure: D more internal than ND; no group differences for stability" or control Success: no group differences"

Experiment 2

D and ND students

Success and failure on a social perception test

Internal, external, control, responsibility

Failure: D more control" over partner's poor performance than ND Success: no group differences

Willis & Blaney, 1978, Experiment 1

o

8 2

o

K O D and ND students

Performance on Sky ball task

Effort, luck, ability, task difficulty, control

D more control than ND; no other group differences

Experiment 2

Helplessness-induced and control Type A students

Performance on Sky ball task following helplessness induction


Helpless subjects attribute less to ability & control than control subjects; no other group differences

Experiment 3

D and ND students

Performance on an anagram task


No significant group differences

Garber & Hollon, 1980

D and ND students

Performance on skill and chance tasks

Control

No significant group differences

Zernore & Johansen, 1980

Students

Failure on a word association task

Effort, luck, ability, task difficulty

BDI positively correlated with ability; no • significant correlations for effort, luck, or task difficulty

P3

Table 1 (continued) Study

Subject population



Results

Zuroff, 1981

D, formerly D (FD), and ND female students

Success and failure on an "empathy" test

Internal, stable

Failure: D more internal than FD, NDa; no differences for stability' Success: D more internal than FD, ND°; no differences for stability

Gotlib & Olson, in press

D and ND psychiatric inpatients, ND rionpsychiatric controls

Self-selected success and failure on a syllable recognition task

Internal, stable

No significant group differences; all "success" subjects more internal and unstable than "failure" subjects

4 negative scenarios

Behavioral and characterological selfblame, chance

D more characterological blame than ND; D more chance than ND; no group differences in behavioral self-blame

Attributions for hypothetical events Janoff-Bulman, 1979

D and ND female students

Seligman, Abramson, Semmel, & von Baeyer, 1979

D and ND students

Attributional Style Questionnaire

Internal, stable, global

Bad outcomes: D more internal8-'' stable"'b and global"-" than ND Good outcomes: D less stable"'13; no differences for internality"-b or gloabh'ty^

Blaney, Behar, & Head 1980

D and ND students



Bad outcomes: D more stable"* and global6 than ND; no differences for mternality"-b Good outcomes: D less internal°-b and stable"* than ND; no difference for globality"-"

Golin, Sweeney, & Schaeffer, 1981

Students

Attributional Style Questionnaire; 2 test sessions, 1-month lag


Bad outcomes: BDI positively correlated with internal, stable, & global attributions (.10 < r< .25) Good outcomes: BDI negatively correlated with internal attributions; stability & globality ns. Time 1 stability & globality causally related to Time 2 BDI (10% & 3% of the variance)

Lewinsohn, Steinmetz, Larsen, & Franklin, 1981

Community residents

Affiliation items on the Multi-dimensional Multiattributional Causality Scale

Internal and stable for success and failure

No depression-associated differences in attributions

Peterson, Schwartz, & Seligman, 1981

Female students


Internal (behavioral & characterological), external

Bad outcomes: BDI negatively correlated with external & behavioral; positively with characterological

(table continued)

O

8z O

o tn rn

00 C/5

i

Table 1 (continued) Studv

Subject population


Results


Good outcomes: BDI positively correlated with external; no other significant correlations Characterological attributions did not predict subsequent depression Manly, McMahon, Bradley, & Davidson, 1982

Pregnant women

Attributionai Style Questionnaire


No significant concurrent correlations; attributional style did not predict subsequent postpartum depression

Metalsky, Abramson, Seligman, Sernmel, & Peterson, 1982

University students



Internal and global attributional styles for negative events on the ASQ predicted increased depression following receipt of low grades on an examination; no effects for stability

Miller, KJee, & Norman 1982

D and ND psychiatric inpatients

Attributionai Style Questionnaire; Single most stressful life event


No group differences on composite ASQ score; D higher composite score for life event than ND; Correlations of attributional consistency nonsignificant

Raps, Peterson, Reinhard, Abramson, & Seligman, 1982

D inpatients, ND schizophrenic inpatients (S), ND medical patients (N)



Bad outcomes: internality": D > N > S stability"-": D > N > S globality": D > N Good outcomes: internality"-b: D < N stability"-": D < N globalitya-b: ns.

w

W

n o o

m

z pc o r

Attributions for stressful life events

03

Gong-Guy & Hammen, 1980

D and ND psychology clinic patients

5 most stressful life events

Internal, stable, global, control

No group differences in attributions for all 5 events; D more internal" than ND for most stressful event

Barthe & Hammen, 1981

D and ND students

Success and failure on a midterm examination

Effort, Luck, Ability, Task Difficulty, Control

D attributed failure more and success less to ability than ND; no other significant group differences

Feather & Davenport, 1981

Male and female unemployed youth

Unemployment

Internal, External

D less internal than ND; D more external" than ND

Hammen & Cochran, 1981

D and ND students and ND high life stress control students


Internal, Stable, Global, Control

No group differences

Hammen, Krantz, & Cochran, 1981

D and ND students


Internal, Stable, Global, Control

D more globality "-b than ND D more control"-1" than ND


1* •3-B 8 SvcS 4J £P $3

'«!a ill oS| §

™

C3

s*- ^ Ipj?

ed

^ "R