Alloy & Abramson - University of Nottingham · Y. Abramson while both authors were at the University of Pennsylvania. National Institute of Mental Health Grant MH 19604 to M. E. P

Journal of Experimental Psychology: General1979, Vol. 108, No. 4, 441-485

Judgment of Contingency in Depressed and NondepressedStudents: Sadder but Wiser?

Lauren B. Alloy Lyn Y. AbramsonNorthwestern University State University of New York at Stony Brook

SUMMARY

How are humans' subjective judgments of contingencies related to objectivecontingencies? Work in social psychology and human contingency learning pre-dicts that the greater the frequency of desired outcomes, the greater people'sjudgments of contingency will be. Second, the learned helplessness theory ofdepression provides both a strong and a weak prediction concerning the linkagebetween subjective and objective contingencies. According to the strong predic-tion, depressed individuals should underestimate the degree of contingency betweentheir responses and outcomes relative to the objective degree of contingency. Ac-cording to the weak prediction, depressed individuals merely should judge thatthere is a smaller degree of contingency between their responses and outcomesthan nondepressed individuals should. In addition, the present investigation de-duced a new strong prediction from the helplessness theory: Nondepressed in-dividuals should overestimate the degree of contingency between their responsesand outcomes relative to the objective degree of contingency.

In the experiments, depressed and nondepressed students were presented withone of a series of problems varying in the actual degree of contingency. In eachproblem, subjects estimated the degree of contingency between their responses(pressing or not pressing a button) and an environmental outcome (onset of agreen light). Performance on a behavioral task and estimates of the conditionalprobability of green light onset associated with the two response alternativesprovided additional measures for assessing beliefs about contingencies.

Depressed students' judgments of contingency were surprisingly accurate in allfour experiments. Nondepressed students, on the other hand, overestimated thedegree of contingency between their responses and outcomes when noncontingentoutcomes were frequent and/or desired and underestimated the degree of con-tingency when contingent outcomes were undesired. Thus, predictions derived fromsocial psychology concerning the linkage between subjective and objective con-tingencies were confirmed for nondepressed students but not for depressedstudents. Further, the predictions of helplessness theory received, at best, minimalsupport.

The learned helplessness and self-serving motivational bias hypotheses areevaluated as explanations of the results. In addition, parallels are drawn betweenthe present results and phenomena in cognitive psychology, social psychology,and animal learning. Finally, implications for cognitive illusions in normal people,appetitive helplessness, judgment of contingency between stimuli, and learningtheory are discussed.

Copyright 1979 by the American Psychological Association, Inc. 0096-3445/79/0804-0441$00.75

441

442 LAUREN B. ALLOY AND LYN Y. ABRAMSON

Development of the Concept of Contingencyin Learning Theory

An important part of animals' andhumans' knowledge of the world is theknowledge of the relationship or contingencybetween events. The concept of contingencyprovides a cornerstone for a number ofhighly influential contemporary theories oflearning (e.g., Bindra, 1972; Bolles, 1972;Mackintosh, 1975; Maier & Seligman, 1976;Rescorla, 1967; Rescorla & Wagner, 1972).Each of these theories posits an organismthat is sensitive to relationships betweenstimuli in its environment and/or betweenits own responses and environmental out-comes. An important question emerges in ananalysis of an organism's commerce withenvironmental contingencies: Do organismsform subjective representations (i.e., beliefs,expectations, or cognitive representations)of contingencies that mirror objective, realworld contingencies? This article focuseson an empirical answer to this question.

Contemporary theorists distinguish be-tween contingency and temporal contiguityor pairing views of learning (Premack,1965; Prokasy, 1965; Rescorla, 1967). Forexample, Rescorla argued that contingency,rather than temporal contiguity, is the essen-tial condition for "true" Pavlovian condition-ing. The crucial distinction between the pair-ing and contingency views is that the former

The research reported in this article was sup-ported by National Science Foundation PredoctoralFellowship SM-76 22871 and National Institute ofMental Health Predoctoral Fellowship MH07284-01 to L. B. Alloy and a University ofPennsylvania Dissertation Year Fellowship to L.Y. Abramson while both authors were at theUniversity of Pennsylvania. National Institute ofMental Health Grant MH 19604 to M. E. P.Seligman also contributed to the research.

We wish to express our gratitude to Philip J.Bersh, Martin E. P. Seligman, Richard L. Solo-mon, and Julius Wishner for their criticisms andsuggestions at various stages of the research.

The order of authors is random.Requests for reprints should be sent to Lyn

Abramson, Department of Psychology, State Uni-versity of New York at Stony Brook, StonyBrook, New York 11794; or to Lauren Alloy,Department of Psychology, Northwestern Univer-sity, 18S9 Sheridan Road, Evanston, Illinois 60201.

centers on the relationship between the con-ditional stimulus (CS) and the unconditionalstimulus (UCS), whereas the latter centerson both this relationship and the relationshipbetween the absence of the CS (CS~) andthe UCS. A large body of evidence onanimal learning supports the assertion thatanimals are, in fact, sensitive to both posi-tive and negative contingencies (e.g., Res-corla, 1969c; Rescorla & Wagner, 1972;Wagner & Rescorla, 1972).

In addition, Rescorla (1967) has pointedout that in contrast to the contiguity view,the concept of contingency has an importantadvantage in that it allows one to distinguishbetween mere absence of excitatory condi-tioning and an active inhibitory process.According to the contingency view, inhib-itory conditioning occurs whenever there isa negative relationship between CS andUCS (i.e., the UCS occurs more often inthe absence of the CS than in its presence).This prediction has been strongly confirmed(Bull & Overmier, 1968; Dweck & Wagner,1970; Hammond, 1966, 1967, 1968; Ham-mond & Daniel, 1970; Kimble & Ost, 1961;Moskowitz & Lolordo, 1968; Reiss & Wag-ner, 1972; Rescorla, 1967, 1969a, 1969b,1969c, 1971; Rescorla & Lolordo, 1965;Siegel & Domjan, 1971; Weisman & Litner,1969, 1972). Finally, a contingency accountof conditioning suggests that the only appro-priate baseline against which to assess theassociative effects of both excitatory andinhibitory conditioning is a condition inwhich there is no contingency between CSand UCS. Indeed, Rescorla (1967) hasargued for just such a "truly randomcontrol" procedure.

From the standpoint of the contingencylearning framework, it is important to knowwhat the organism learns when there is noobjective contingency between stimulus andreinforcer. Does the organism merely learnnothing about the stimulus, or does it learnexplicitly that stimulus and reinforcer are,in fact, uncorrelated ? Mackintosh's (1973,1975) investigations of the phenomenon of"learned irrelevance" have led him to sup-port the latter position. Thus, Mackintosh(1973) has found that random presentations

JUDGMENT OF CONTINGENCY 443

of a given CS and UCS specifically retardthe subsequent formation of an associationbetween the two. To account for this phe-nomenon, Mackintosh (1973, 1975) pro-posed an attentional theory that posits thatorganisms actively learn to ignore stimuliuncorrelated with reinforcement.

Paralleling its development within thePavlovian tradition, the concept of contin-gency also has received increasing attentionin theoretical accounts of instrumental learn-ing (e.g., Bloomfield, 1972; Catania, 1971;Church, 1969; Gibbon, Berryman, &Thompson, 1974; Maier, Seligman, & Solo-mon, 1969; Premack, 1965; Seligman,Maier, & Solomon, 1971; Weiss, 1968). Forexample, Seligman et al. (1971) have arguedthat organisms are sensitive to conjointvariations in two response-outcome prob-abilities: the conditional probability of anoutcome given the occurrence of a response,P(O/R), and the conditional probability ofthat outcome given the nonoccurrence of thatresponse, P(O/R). According to Seligmanet al., any account of instrumental learningthat fails to incorporate both of these prob-abilities is incomplete.

Seligman and his colleagues (Maier &Seligman, 1976; Maier et al., 1969; Selig-man et al., 1971), in their investigation ofthe "learned helplessness" phenomenon, haveemphasized the theoretically interesting casein which the two response-outcome prob-abilities are equal. These investigators arguethat organisms that have been exposed toaversive events that terminate independentlyof any instrumental responses actively learnthat reinforcement is independent of theirbehavior and that these events are, in fact,uncontrollable. A great deal of evidence sup-ports the contention that organisms aresensitive to response-outcome independence(for a review of these studies see Maier &Seligman, 1976, and Seligman, 1975b).Furthermore, according to the learned help-lessness hypothesis, such learning results inmotivational and cognitive deficits as wellas in emotional disturbance. The motiva-tional component of learned helplessness maybe described as reduced incentive for initiat-ing voluntary responses, and the cognitive

deficit is reflected in difficulty in learningfuture response-outcome contingencies. Fi-nally, the learned helplessness hypothesisalso states that organisms that have learnedthat outcomes are noncontingently related toresponses become "depressed" (see Maier &Seligman, 1976, and Seligman, 1975b, for areview of the relevant evidence).

The debilitating consequences of exposureto situations in which responses and out-comes are unrelated have been observedacross a wide variety of experimental sit-uations and within a large number of species,including dogs (Overmier, 1968; Overmier& Seligman, 1967; Seligman & Groves,1970; Seligman & Maier, 1967), mice (e.g.,Braud, Wepman, & Russo, 1969), rats (e.g.,Baker, 1976; Goodkin, 1976; Maier, Albin,& Testa, 1973; Maier & Testa, 1975; Selig-man & Beagley, 1975; Seligman, Rosellini,& Kozak, 1975; Williams & Maier, 1977),fish (Padilla, Padilla, Ketterer, & Giacalone,1970), cats (e.g., Seward & Humphrey,1967), and humans (e.g., Fosco & Geer,1971; Glass & Singer, 1972; Hiroto, 1974;Hiroto & Seligman, 1975; Klein, Fencil-Morse, & Seligman, 1976; Klein & Selig-man, 1976; Miller & Seligman, 1975; Roth& Bootzin, 1974; Roth & Kubal, 1975;Thornton & Jacobs, 1971).

Of particular import to the present inves-tigation are the human helplessness studies,because they provide evidence about thesensitivity of humans to the objective pres-ence or absence of contingencies betweentheir responses and environmental outcomes.Hiroto's (1974) experiment is representa-tive of human helplessness studies. In atriadic design analogous to those employedin the animal helplessness studies, Hirotocompared three groups of college students:an "escapable" group that received experi-ence with noises contingent on button-pressing responses, an "inescapable" groupthat received experience with noises non-contingently related to button pressing, anda "no noise" group. All groups were sub-sequently tested on a human shuttlebox forescape/avoidance from noise. The resultswere similar to those obtained from otherspecies: Students who had received prior


exposure to noncontingent noise showedimpaired performance of the requisite escape/avoidance response in the shuttle box testcompared to students receiving prior ex-posure to contingent noise or no noise.These results are typical and provide evi-dence for the motivational and cognitivecomponents predicted by the learned help-lessness hypothesis.

The emotional component of learned help-lessness has come under investigation. Intwo studies of the affective component, non-depressed college students exposed to un-controllable noises became more depressedrelative to either a group that receivedcontrollable noises (Gatchel, Paulus, &Maples, 1975) or to a group that receivedno noises (Miller & Seligman, 1975). Notingthe similarities between the behaviors exhib-ited by animals and humans exposed tononcontingent aversive events in the labora-tory and the behavioral symptoms of na-turally occurring depression in humans,Seligman and his colleagues (Abramson,Seligman, & Teasdale, 1978; Miller, Rosel-lini, & Seligman, 1977; Seligman, 1975a,1975b; Seligman, Klein, & Miller, 1976)have proposed a learned helplessness theoryof depression. The helplessness theory claimsthat the expectation that important outcomesand responding are independent causes themajor motivational, cognitive, and emotionalsymptoms of human depression.

Subjective Reality of Contingency

This brief review of experimental evidencebearing on contingency learning does indeedsuggest that both animals and humans aresensitive to the presence and absence of cor-relations between stimuli and/or responsesand outcomes. More specifically, this evi-dence points to the role of objective contin-gencies as determinants of organisms'behavior in Pavlovian and instrumentallearning situations. Although the experi-mental evidence underscores the need fora concept of contingency in theoreticalaccounts of learning, the role of subjectiverepresentations of contingencies in suchlearning is still very much at issue. Res-

corla and Wagner (1972), for example, intheir model of Pavlovian conditioning, havedeveloped a molecular theoretical formula-tion that does not invoke a concept ofcontingency with subjective reality for theirexperimental subjects. Maier and Seligman(1976), on the other hand, argue that objec-tive contingencies are represented subjec-tively and that this representation directlyinfluences the organism's behavior.

Methods for Assessing Humans'Representations of Contingency

Although it is difficult to demonstrateconvincingly that animals have cognitiverepresentations of objective contingencies,methods developed in experiments on humanlearning may provide a more conclusive testof the subjective reality of concepts of con-tingency. Although the majority of humanhelplessness studies have examined perform-ance decrements within the transfer of train-ing paradigm, Seligman and his associates(Abramson, Garber, Edwards, & Seligman,1978; Klein & Seligman, 1976; Miller &Seligman, 1973, 1975, 1976) have employedwhat appears to be a more direct method forassessing humans' cognitive representationsof contingencies. These investigators havenoted the similarity between the helplessnessconcept that outcomes and responding areindependent and Rotter's (1966) conceptof external control.

Rotter and his associates (James, 1957;James & Rotter, 1958; Phares, 1957; Rotter,Liverant, & Crowne, 1961) used tasks inwhich success appeared to be determined byeither chance or skill. They demonstratedthat verbalized expectancies for future suc-cess are affected by outcomes of previoustrials. In particular, they found that out-comes of previous trials have a greatereffect on expectancies for future successwhen the person believes outcomes aredependent on responses (skill determined)than when he or she believes outcomes areindependent of responses (chance deter-mined).

Employing this logic, Miller and Seligman(1976) and Klein and Seligman (1976)


examined verbalized expectancies of successon skill and chance tasks for college studentsgiven prior exposure to contingent, noncon-tingent, or no noises. In both studies, stu-dents exposed to prior uncontrollable noisesshowed less expectancy change in an osten-sible skill task than students exposed toprior controllable noises or no noises, al-though the groups did not differ on a chancetask. From these results, Miller and Selig-man (1976) and Klein and Seligman(1976) inferred that helpless students hadacquired a generalized expectancy ofresponse-outcome independence.

However, a conceptual problem precludesacceptance of the chance-skill method asproviding an assessment of humans' cogni-tive representations of response-outcomecontingencies. Recent developments in attri-bution theory (Weiner et al., 1971; Weiner,Nierenberg, & Goldstein, 1976) suggest thatchanges in expectancy of success are a func-tion of the perceived likelihood that factorsthat produced prior successes will be presentagain in the future rather than a functionof subjects' perceptions of response-outcomecontingencies. Therefore, although learnedhelplessness theory postulates that organismsacquire cognitive representations of objectivecontingencies, no method within the help-lessness framework has yet proved adequateto assess the subjective reality of suchconcepts of contingency.

A more promising method for assessinghumans' capacities for cognitive representa-tion of contingencies was developed byJenkins and Ward (1965) independentlyof work on helplessness. They presented sub-jects with a series of contingency problemsin an instrumental learning situation. Foreach problem, subjects were given 60 trialson which a choice between two responses(Button 1 and Button 2) was followed byone of two possible outcomes (score or noscore). All subjects received some problemsin which responses and outcomes were con-tingently related and other problems inwhich responses and outcomes were noncon-tingently related. It is interesting that Jenk-ins and Ward used problems in which thedesired outcome (score) occurred frequently

but noncontingently as well as problems inwhich it occurred rarely but noncontingently.At the end of each problem, subjects wereasked to rate, on a 0 to 100 scale, the degreeof control (contingency) that their responsechoices had exerted over the outcomes. Theexperimenters argued that a contingencybetween response and outcome exists (i.e.,the outcome is controllable) when the prob-ability of that outcome given the occurrenceof one response differs from the probabilityof that outcome given the occurrence ofanother response. When, for all responses,there is no difference between these condi-tional probabilities, the outcome is said to beuncontrollable. Whereas Seligman et al.(1971) only distinguished between contin-gent and noncontingent cases, Jenkins andWard extended the concept of control toinclude the degree of control (or contin-gency) defined as the magnitude of the dif-ference between the two relevant conditionalprobabilities.

The results from Jenkins and Ward's(1965) studies indicate that humans' sub-jective representations of contingencies arenot isomorphic with objective contingencies.Regardless of the actual degree of contin-gency, subjects' ratings of degree of controlcorrelated highly only with the number ofsuccessful trials (the number of trials onwhich the score card appeared) and werecompletely unrelated to the actual degree ofcontrol.

Results similar to those reported by Jenk-ins and Ward have been obtained in studiesof the effects of noncontingent reinforcementon performance in learning tasks. Forexample, Bruner and Revusky (1961) foundthat human subjects in an instrumentallearning situation developed complex re-sponse patterns, which were irrelevant to theproduction of reinforcement, on nonfunc-tional telegraph keys. Similarly, Wright(1962) showed that subjects' response pat-terns were more orderly at high levels ofnoncontingent reward than at intermediatelevels. Finally, in a stimulus prediction sit-uation, Hake and Hyman (1953) found thatsubjects did not respond to a random seriesof binary digits as if it were random.


Factors Predicted to Affect the Judgment ofContingency

A general theme runs through these con-tingency learning studies: People often treatnoncontingent situations as if they were con-tingent. They act as though outcomes aredependent on responses when they are notand as though one event can be predictedfrom another when it cannot. Given thesebehavioral results, it is not unreasonable toexpect that peoples' representations of con-tingencies also may differ systematicallyfrom the objective, real world contingencies.From these studies, one might surmise thata variable likely to systematically affect therelationship between subjective contingenciesand objective contingencies is the frequencyof the desired outcome. People may be morelikely to believe that a contingency existsbetween -their responses and desired environ-mental outcomes when those outcomes occurwith relatively high frequency.

A second source of hypotheses concerningthe linkage of objective contingencies to sub-jective contingencies is the clinical andexperimental work on the psychopathologyof depression. As noted earlier, the learnedhelplessness model of depression states thatwhen people acquire the expectation thatimportant outcomes and responses are inde-pendent, they exhibit the major motivational,cognitive, and affective symptoms of depres-sion. Similar theories of depression, embody-ing the core concept of helplessness or hope-lessness, have been proposed by clinicians(e.g., Arieti, 1970; Beck, 1967; Bibring,1953; Lichtenberg, 1957; Melges & Bowlby,1969).

A deduction from the learned helplessnessmodel is that depressed individuals often willunderestimate the degree of contingencybetween their responses and environmentaloutcomes (e.g., Miller & Seligman, 1973;Seligman, 1975b). Seligman and his asso-ciates recently have tested his prediction byemploying the chance-skill method discussedearlier. Similar to nondepressed college stu-dents who received experience with uncon-trollable events, naive depressed studentsexhibited smaller expectancy changes follow-ing success (Klein & Seligman, 1976; Miller

& Seligman, 1973, 1975) and failure (Klein& Seligman, 1976; Miller & Seligman, 1975,1976) on a skill task than did nondepressedstudents who received experience with con-trollable events or naive nondepressed stu-dents. In addition, Abramson et al. (1978)reported that unipolar depressives alsoshowed small expectancy changes in a skilltask relative to hospitalized control subjectsand schizophrenics. Although the expectancychange differences between depressed andnondepressed individuals are robust andoften replicated findings, the interpretationof these findings with respect to depressives'cognitive distortions of contingencies is stillunclear. As noted earlier, expectancychanges on chance and skill tasks may notbe valid indices of people's beliefs aboutresponse-outcome contingencies.

Our summary of the work on human con-tingency learning and depression suggeststhat people's cognitive representations ofresponse-outcome relationships may notdirectly mirror environmental contingencies.Studies of people's behavior in noncontingentsituations provide a clue that one possiblesource of systematic error in transductionof environmental contingencies is the natureand frequency of the outcome itself. In addi-tion, the learned helplessness model of de-pression suggests that depressed and non-depressed individuals also differ syste-matically in their subjective representationsof response-outcome relationships. Both astrong and a weak prediction may be deducedfrom the helplessness model. According tothe strong prediction, depressed individualsunderestimate absolutely the degree ofobjective contingency between their re-sponses and outcomes. According to theweak prediction, depressed individualsmerely underestimate the degree of objectivecontingency relative to nondepressed individ-uals. Both views, therefore, predict a netdifference in subjective judgments of contin-gencies between depressed and nondepressedpeople. However, the strong view furtherspecifies that depressed individuals under-estimate the degree of contingency relativeto the objective contingency.

The present investigation tested these pre-


dictions by examining subjective judgmentsof contingency in depressed and nonde-pressed students using a modification of themethod developed by Jenkins and Ward(1965). A number of measures, includingbehavioral indices as well as verbalizedjudgments, were used to assess students'representations of contingency. Studentswere asked to judge the degree of controltheir responses exerted over outcomes ratherthan the degree of contingency betweenresponses and outcomes. Contingency is ageneral term and refers to the degree ofrelationship between any two events. In thePavlovian conditioning situation, the eventsof interest are both stimuli, and the relation-ship between them is best construed as oneof predictability. Alternatively, in the instru-mental learning situation, the two eventsconsist of the organism's responses and someoutcome or reinforcer. The relation betweensuch events is best construed as one of con-trollability—the response exerts either someor no degree of control over the outcome(see Seligman, 1975, for a discussion of thedistinction between predictability and con-trollability). Thus, control is best defined asthe dependence of an outcome on a response.A further reason for using the terminologyof control when interacting with the subjectsin the present experimental context is that itconveys the technical meaning of contingencyin everyday language (Jenkins & Ward,1965).

For the present series of experiments, itwas necessary to construct an index of theactual degree of control or dependency be-tween responses and outcomes. In statistics,the phi coefficient (<£) is commonly used toquantify the degree of relation between twoevents. However, in their study of judg-ments of contingency, Jenkins and Ward(1965) used the magnitude of the differencebetween the conditional probability of anoutcome given the occurrence of oneresponse versus the conditional probabilityof the outcome given the occurrence of an-other response as the index of degree ofcontrol or contingency. Mathematically, thisindex is a close approximation to <j> and, infact, is highly correlated with 4>. We chose

to use this difference-in-probability metric asour index of objective degree of control orcontingency because it is mathematically lesscumbersome than <f> and because it facilitatescomparison of prior studies on the judgmentof contingency with the present study. Anexample clarifies the difference-in-probabil-ity metric used in the present investigation.According to this metric, if the probabilityof obtaining an A on an exam is .90 whenyou study and .10 when you do not study,the degree of contingency between studyingand obtaining an A is .80. Alternatively, ifthe probability of obtaining an A is 1.0whether you study or not, then studying andobtaining an A are noncontingently related.

In Experiment 1, students were presentedwith one of a series of problems in whichthere was some degree of contingencybetween their responses and outcomes. Theproblems differed in the degree of contin-gency between responses and outcomes.Experiments 2, 3, and 4 examined students'abilities to detect contingency and noncon-tingency between responses and outcomesunder different conditions of outcome fre-quency and desirability.

Experiment 1

In this study, we examined depressed andnondepressed college students' abilities todetect the degree of contingency betweentheir responses and environment outcomes.Students were presented with one of threecontingency problems differing in degree ofcontingency or control. The contingencyproblems consisted of a series of trials onwhich the subject made one of two possibleresponses (pressing a button or not pressinga button) and received one of two possibleoutcomes (a green light or no green light).At the end of the series of trials, the subjectwas asked to judge the degree of contingencythat existed between button pressing andgreen light onset.

In Experiment 1, as well as in Experi-ments 2, 3, and 4, a modification of themethod of Jenkins and Ward (1965) wasused. Because Jenkins and Ward found thatsubjects' judgments of degree of controlcorrelated highly with the frequency of rein-


forcement, it was important to ensure thatfrequency of reinforcement was not con-founded with actual degree of contingency.The present experiment was designed sothat frequency of reinforcement and actualdegree of contingency were negatively corre-lated provided that the subject sampled thetwo response alternatives, pressing the but-ton and not pressing the button, fairly reg-ularly. This procedure ensured that if stu-dents relied on percentage or frequency ofreinforcement as an indicant of degree ofcontingency, they would give incorrectjudgments negatively correlated with actualdegree of contingency. Thus, the predictionthat subjective representations of contin-gency are more closely related to the fre-quency of the outcome than to objectivecontingency was examined.

Including both depressed and nonde-pressed students in Experiment 1 tested thelearned helplessness theory of depression.According to the strong helplessness predic-tion, depressed students should underesti-mate the degree of contingency on each ofthe three contingency problems. Accordingto the weak prediction, depressed studentsmerely should judge that they have lesscontrol than nondepressed students on eachof the three problems.

Making a judgment about the degree ofcontingency between one's responses and anenvironmental outcome is best conceived ofas an instance of drawing an inference fromraw data. If subjects do err in their judg-ments of contingency, it could be either be-cause they have not collected the appropriateraw data or because they have drawn incor-rect inferences from a set of appropriateraw data. For example, a subject may knowthe probability of green light onset associatedwith pressing and not pressing but be unableto organize these probabilities in the mannernecessary for making a correct judgment ofcontingency. Alternatively, a subject couldmake an incorrect judgment of contingencybecause he or she simply did not know theconditional probabilities of green light onset.Therefore, in Experiment 1, as well as inExperiments 2, 3, and 4, we examined sub-jects' judgments of the probability of green

light onset when they pressed and did notpress.

A further issue of interest is the subject'schoice behavior in a contingency learningsituation. Although we argued previouslythat subjects' instrumental behavior in acontingency learning situation is not a suffi-cient condition for inferring a cognitive rep-resentation of contingency, such behavioraldata may provide converging evidenceregarding such a representation. Conse-quently, after giving their series of judg-ments, subjects were presented with a be-havioral task in which they could obtainmoney by producing green light onset.

Method

Subjects. Ninety-six undergraduates, 48 malesand 48 females, from the Universiy of Pennsylvaniaserved as paid volunteers. The data of one addi-tional subject were discarded because she failed tofollow the experimental instructions. Subjects wereassigned to a depressed or nondepressed group onthe basis of their Beck Depression Inventory(BDI) scores (Beck, 1967).1 Subjects with BDI

1 Four criteria were used for the selection of adepression inventory. The inventory had to be self-administered, relatively short, capable of providingquanitative information for assessing depth ofdepression, and relatively well validated. AlthoughZung's (1965) Self-Rating Depression Scale andthe BDI both meet each of these selection criteria,the BDI was selected because its validity has beendemonstrated by more detailed and comprehensivestudies. Beck (1967) has reported reliability andvalidity studies involving nearly 1,000 subjects. Intwo studies, he found significant differences in meanBDI scores between groups of psychiatric inpatientsand outpatients rated as exhibiting none, mild,moderate, or severe depression. Metcalfe and Gold-man (1965) reported similar results in a cross-validation study. The correlations between BDIscores and clinically rated severity of depressionin these three studies were .65, .67, and .61, respec-tively. Recently, Bumberry, Oliver, and McClure(1978) found that the BDI is a valid, instrumentfor measuring depression in a college student pop-ulation. In addition, studies examining predictionsfrom the learned helplessness model of depressionhave found that the BDI is a very sensitive pre-dictor of behavior. These studies consistently havefound substantial correlations between BDI scoresand a variety of deficits. Miller and Seligman(1976) reported the range to be between .45 and.53. Correlations between BDI scores and noiseescape measures have been as high as .74 (Klein &Seligman, 1976).


Table 1Means and Standard Deviations of BDI and MA ACL Scores by Problem,Mood, and Sex for Experiment 1

Nondepressed Depressed

Problem

test

7S-SOBDIMAACL

75-25BDIMAACL

75- 0BDIMAACL

Males

M

3.113.1

3.516.0

4.010.9

SD

1.95.1

2.04.3

2.75.5

Females

M

3.413.8

4.814.6

4.811.4

SD

3.47.8

2.43.9

1.85.8

Males

M

11.817.9

12.917.6

12.117.1

SD

2.06.5

3.15.5

3.17.1

Females

M

13.421.0

16.114.6

20.017.8

SD

4.77.6

9.79.8

6.69.9

Note. BDI = Beck Depression Inventory; MAACL = Multiple Affect Adjective Check List.

scores of 9 or above were assigned to the depressedgroup, and subjects scoring 8 or lower wereassigned to the nondepressed group. The finalsample consisted of 48 depressed students (24 malesand 24 females) and 48 nondepressed students (24males and 24 females). In addition, subjects com-pleted the Multiple Affect Adjective Check List(MAACL) Today form (Zuckerman & Lubin,1965) as a further index of depression althoughthese scores were not used as a criterion for select-ing depressed and nondepressed students. The corre-lation between the BDI and the MAACL depres-sion scores was .51 (/><.001). Table 1 presentsthe mean BDI and MAACL scores for all experi-mental groups. Subjects were randomly assigned tothe three experimental conditions with the restrictionthat each condition contain equal numbers of malesand females and depressed and nondepressed stu-dents. Half or the subjects in each condition weretested by Experimenter 1 and half by Experimenter2.

Experimental design. The experiment was a 3(Problem Type) X 2 (Mood—depressed, non-depressed) X 2 (Sex—males, females) X 2 (Re-sponse—press, no press) factorial design. The threeproblems differed in degree of control. In Problem1( (75_50), subjects had 25% control; in Problem2 (75-25), they had 50% control; and in Problem3, (75-0), they had 75% control. The first numberof each problem name denotes the percentage oftrials on which the outcome of interest (green lightonset) occurred when the subject pressed thebutton. The second number denotes the percentageof trials on which green light onset occurred whenthe subject did not press the button. As definedabove, the degree of control (contingency) wasdetermined by the difference between these twonumbers." Of the 96 subjects, 32 (8 depressed males,8 depressed females, 8 nondepressed males, and 8

nondepressed females) were assigned to eachproblem.

Within each problem, subjects were counter-balanced for whether pressing or not pressing thebutton led to a higher percentage of green lightonsets. For example, in Problem 1 the green lightcame on 75% of the time when the button waspressed and 50% of the time when the button wasnot pressed (75-50) for half of the subjects. Forthe other half of the subjects in Problem 1, thegreen light came on 50% of the time when thebutton was pressed and 75% of the time when itwas not pressed (50-75). In addition to differing ondegree of contingency, the three problems werealso designed to differ on overall percentage ofreinforcement (overall frequency of green lightonsets). In particular, the percentage of reinforce-ment for the problems was negatively correlatedwith the degree of contingency provided that thesubjects sampled the two responses (pressing andnot pressing) fairly regularly. Table 2 summarizesthe experimental design and presents the numberof subjects in each cell of the design with the per-centage of reinforcement received by these subjects.As can be seen in Table 2, the percentag of rein-forcement did, in fact, correlate negatively withdegree of contingency (r = —.59, p < .001).

Dependent measures. The major dependent mea-sures were four judgment scales and performanceon a behavioral task. On the Judgment of Controlscale, subjects rated the degree of control theirresponses (pressing and not pressing) exerted over

2 Inspection of individual subjects' protocols forall four experiments revealed that the actual degreeof control experienced by each subject in theexperimental session deviated only slightly or notat all from the nominal degree of control.


the experimental outcome (green light onset). Onthe second judgment scale, Judgment of TotalReinforcement, subjects estimated the overall per-centage of trials on which green light onsetoccurred regardless of which response they made.The last two scales, Judgment of Reinforcement ifPress and Judgment of Reinforcement if Not Press,were designed to assess whether subjects knew theraw data necessary to compute the conditionalprobabilities that were necessary for making anaccurate judgment of control. On these two scales,subjects estimated the percentage of trials on whichthe green light came on when they pressed andwhen they did not press, respectively.

Because Ward and Jenkins (1965) found thatsubjects often used the invalid heuristics of per-centage of successes and confirming cases in arriv-ing at their verbalized judgments of control, thesemeasures were also used in the present experiment.In line with Ward and Jenkins (1965), percentageof successes was defined as the percentage of trialson which the green light appeared when the subjectpressed the button. Confirming cases was definedas the sum of the number of trials on which thesubject pressed the button and the green lightcame on plus the number of trials on which thesubject did not press the button and the green lightdid not come on.

A behavioral task was employed as a furthermeasure of subjects' representations of contin-gencies. In this task, subjects were given 10 trialsfrom the contingency problem they had justreceived and were paid to maximize green lightonset.

Several additional dependent measures wereincluded on a postexperimental questionnaire. First,subjects rated how certain they were of theaccuracy of their judgments of degree of control.In addition, two open-ended questions allowed sub-

jects to describe the evidence that convinced themthat they either had or did not have control as wellas the evidence that would have convinced them ofthe opposite conclusion. Finally, subjects wereasked whether they employed any complex hypoth-eses (e.g., sequential or time-dependent patterns)during the contingency learning problems, and ifso, what these hypotheses were.

Apparatus and materials. The experiment wasconducted in a two-room suite. Standard switchingrelay circuitry equipment for controlling stimuluspresentation and for recording subjects' responseswas housed in the observation room of the suite.Subjects were seated in the experimental room suchthat they could be observed through a one-waymirror by the experimenter in the observationroom.

The stimulus presentation consisted of a blackwooden stand-up platform (23 cm X 23 cm) onwhich a yellow and a green light were positioned5 cm from the top of the platform facing thesubject. The subject's response apparatus consistedof a 15.5 cm X 7.5 cm X 4 cm wooden box, alsopainted black, on which a spring-loaded button wasmounted in the center.

The contingencies between subjects' responses(pressing the button and not pressing) and experi-mental outcomes (green light onset or no greenlight) were programmed with two punched tapereaders and conventional switching circuitry. Theoutcomes were controlled by the subject's responsesthrough relays and a programming device. If thesubject pressed the button within 3 sec, TapeReader 1, which controlled the schedule of greenlight onsets associated with pressing, was activated.If the subject failed to press the button within 3sec, Tape Reader 2, which controlled the scheduleof green light onsets associated with not pressing,

Table 2Number of Subjects per Cell, Degree of Contingency, and Percent ofReinforcement for Experiment 1

Nondepressed

Males

Problem

75-50

75-25

75- 0

Response

75-5050-75

75-2525-75

75- 00-75

n

44

44

44

C

2525

5050

7575

R

62.265.5

54.048.0

49.537.0

n

44

44

44

Females

C

2525

5050

7575

R

63.564.2

57.547.2

41.246.7

n

44

44

44

Depressed

Males

C

2525

5050

7575

R

63.058.0

51.048.0

47.545.7

»

44

44

44

Females

C

2525

5050

7575

R

68.565.0

52.047.5

53.550.0

Note. Response refers to whether pressing or not pressing the button was associated with the higher percent-age of reinforcement, n = number of subjects per cell, C = percent degree of contingency, R = percent ofreinforcement.


was activated. (See the Experimental Designsection for the actual programmed contingencies.)

The BDI (Beck, 1967) and the MAACL(Zuckerman & Lubin, 1965) depression inventories,four judgment scales, and a postexperimental ques-tionnaire constituted the experimental materials.Each of the four judgment scales was marked offin units of five with extreme values of 0 and 100.For the Judgment of Control scale, the extremevalues were labeled No Control and CompleteControl and the 50% point was labeled Intermedi-ate Control. The other three scales (Judgment ofTotal Reinforcement, Judgment of Reinforcementif Press, and Judgment of Reinforcement if NotPress) were simply labeled as percentages.

Procedure. When the subject entered the room,he or she was seated at a desk and was admin-istered the BDI and MAACL. The experimenterwas not present while the subject filled out themood inventories. After completing the inventories,the subject was assigned to an experimental condi-tion and led to a table on which the apparatus forthe contingency learning problem was displayed.The actual procedure and instructions for each ofthe three contingency problem groups were iden-tical. Each of the contingency problems consistedof 40 3-sec trials on which the subject had theoption of either pressing a button or not pressinga button. Onset of a yellow light signaled the startof each trial. At the end of the 3-sec trial, a greenlight was either presented or not presented depen-dent on the subject's response and the contingencyproblem to which the subject had been assigned.The intertrial interval ranged from 10 to 25 secwith a mean of 14 sec. All subjects were giventhe following instructions:

Now, in this problem-solving experiment, it isyour task to learn what degree of control youhave over whether or not this green light comeson. Each time the yellow light comes on indi-cates the start of a new trial, the occasion to dosomething. For each trial, after the yellow lightcomes on, you have the option of either makinga button press response or not making a buttonpress response. A button press response consistsof pressing this button once and only once imme-diately after the yellow light comes on. Notmaking a button press response consists, ofcourse, of doing nothing when the yellow lightcomes on. If you do intend to press the buttonon a given trial, you must press within threeseconds after the yellow light comes on; other-wise the trial will be counted as a not presstrial. So, in this experiment there are only twopossibilities as to what you can do on each ofthe trials: either press the button-within threeseconds after the yellow light comes on, or else,just sit back and do nothing. Any questions sofar?

You may find that the green light will go on, onsome percentage of the trials on which you domake a button press response. You may also find

that the green light will go on, on some percent-age of the trials when you do not make a buttonpress response. Alternatively, you may find thatthe green light will not go on, on some percent-age of the trials on which you do make a buttonpress response. And, you may find that the greenlight will not go on, on some percentage of thetrials when you do not make a button pressresponse. So, there are four possibilities as towhat may happen on any given trial: 1) youpress and the green light does come on; 2) youpress and the green light does not come on; 3)you don't press and the green light does comeon; 4) you don't press and the green light doesnot come on. Since it is your job to learn howmuch control you have other whether the greenlight comes on, as well as whether the greenlight does not come on, it is to your advantageto press on some trials and not on others, so youknow what happens when you don't press as wellas when you do press. Moreover, the knowledgethat you gain from this problem will enable youto earn some money later on in the study. Anyquestions ?

When it was clear that the subject understoodthe outline of the task, he or she was then shownthe Judgment of Control scale and the concept ofcontrol was discussed briefly:

Forty trials will constitute the problem. Afterthe problem, you will be asked to indicate yourjudgment of control by putting an "X" someplaceon this scale: at 100 if you have complete con-trol over the onset of the green light, at 0 if youhave no control over the onset of the green light,and somewhere between these extremes if you havesome but not complete control over the onset ofthe green light. Complete control means that theonset of the green light on any given trial isdetermined by your choice of responses, eitherpressing or not pressing. In other words, whetheror not the green light goes on is totally determinedby whether you choose to press or to just sit backand not press. No control means that you havefound no way to make response choices so as toinfluence in any way the onset of the green light.In other words, the onset of green light hasnothing to do with what you do or don't do.Another way to look at having no control is thatwhether or not the green light comes on, on anygiven trial, is totally determined by factors such aschance or luck, rather than by your choice ofpressing or not pressing. Intermediate degrees ofcontrol means that your choice of responses, eitherpressing or not pressing, influences the onset of thegreen light even though it does not completelydetermine whether the green light goes on ornot. In other words, what you do or don't domatters to some extent but not totally. Anotherway to look at having intermediate control is thatone response, either pressing or not pressing,produces the green light onset more often than


does the other response. So, it may turn out thatyou will have no control, that is, your responseswill not affect the onset of the green light, or itmay turn out that you will have some degree ofcontrol, either complete or intermediate, that is,one response produces green light onset moreoften than does the other. Any questions beforewe begin?The experimenter then left the room and the

subject proceeded with the contingency learningproblem. At the end of the 40 trials, the experi-menter returned and reread the section of the in-structions discussing the concept of control. Thesubject then completed each of the four judgmentscales by placing an X on the scale correspondingto his or her estimate.

Following the judgment scales, the subject par-ticipated in the behavioral task. This task consistedof 10 trials from the contingency learning problemthat the subject had just completed. On each trialin which the green light came on, the subjectearned a quarter. The instructions for the be-havioral task were as follows:

Now, in this last part of the study, you have achance to earn money by demonstrating what youhave just learned in the first part of the study.You will now receive 10 trials from the problemyou just learned about. The relationship betweenyour responses and the onset of the green lightremains exactly the same. But this time yourobjective is to maximize the number of trialson which the green light goes on. On each trialon which the green light does go on, you willearn a quarter. Alternatively, on each trial onwhich the green light does not go on, you willnot earn any money. At the end of the 10 trialsyou will get to keep all of the money you haveearned. So, it is to your advantage to maximizethe number of trials on which the green lightgoes on. Any quesions?The experimenter was absent during the be-

havioral task but returned at the end of the taskand gave the subject the money he or she hadearned. The subject was then administered thepostexperimental questionnaire and was carefullydebriefed. After the debriefing, all subjects weregiven additional money regardless of how muchthey had actually earned so that their total equalled$2.50.

Results 8

Since the experiment was conducted bytwo different experimenters, it was neces-sary to determine whether there were anyexperimenter effects. Analysis of variance re-vealed no significant experimenter effects forany of the dependent measures. Therefore,this factor was not included in the subse-quent statistical analyses.

Judgment scales. No main effects orinteractions involving the factor of response

were obtained on any of the judgment scales.Subjects' judged control, judged reinforce-ment, and judged reinforcement if press andif not press scores were not affected bywhether the higher probability of green lightonset was associated with the response ofpressing or not pressing. For this reason,all statistical analyses on the judgmentscales employed a 3 (Problem Type) X 2(Mood) X 2 (Sex) design with subjects'scores collapsed over the response factor.

Subjects' scores on the Judgment of Con-trol scale indicated that they accuratelyjudged the degree of contingency betweentheir responses and outcomes in the threeproblems. A Problem X Mood X Sex anal-ysis of variance on the Judgment of Controlscores revealed a significant main effect forproblem type, F(2, 84) = 23.54, p < .001.Post hoc Scheffe tests indicated that subjectsbelieved they had significantly more controlin the 75-0 problem than in the 75-25problem and significantly more control in the75-0 problem than in the 75-50 problem(both pa < .001). Although subjects' ratingsof control did not differ significantly betweenthe 75-50 and the 75-25 problems, the differ-ence was in the appropriate direction. Some-what surprisingly, depressed and non-depressed subjects did not differ in theirratings of degree of control. The analysis ofvariance revealed no significant main effectsfor mood or sex and no significant inter-actions involving these factors. Table 3 pre-sents the mean judgment of control scoresfor each experimental group.

In addition to judging degree of control forthe contingency problem to which they wereassigned, subjects also rated how certainthey were of the accuracy of these judgments.A Problem X Mood X Sex analysis of vari-ance of the certainty ratings yielded a signifi-cant main effect for problem, F(2, 84) =4.83, p < .01. The more control subjects had,the more certain they were of the accuracyof their judged degree of control. A post hocScheffe test showed that subjects were sig-nificantly less certain in the 75-50 problemthan in the 75-0 problem (p < .05), with

8 All statistical tests in Experiments 1-4 aretwo-tailed.


Table 3Means and Standard Deviations of Judged Control Scores by Problem,Mood, and Sex for Experiment 1


Males Females Males Females

Problem M SD M SD M SD M SD

75-5075-2575- 0

32.638.182.8

27.628.617.3

38.154.265.6

20.016.416.5

36.256.571.1

21.316.316.1

33.837.671.9

29.231.021.5

the certainty in the 75-25 problem falling inbetween.

Although Jenkins and Ward (1965)found that subjects' judgments of controlwere unrelated to actual degree of controland highly correlated with the actual numberof successes or reinforcements, the oppositeresults were obtained in the present experi-ment. The product-moment correlation ofindividual judgments of control with theactual degree of control, based on all subjectsand all problems, was .56 (/><.001). Incontrast, the partial correlation between indi-vidual judgments of control and judgedpercentage of reinforcement (with actualpercentage of reinforcement held constant)was not significantly different from zero(r = — .12, ns).* Further, subjects judg-ments of control were also uncorrelated withthe heuristics of percentage of successes(r = —.13, ns) and confirming cases(r= -.01, ns).

To summarize, subjects were accurate injudging degree of control and did not appearto use invalid heuristics such as percentageof reinforcement, percentage of successes, orconfirming cases. As Figure 1 shows, judgedcontrol paralleled actual control for bothdepressed and nondepressed subjects. Itshould be noted, however, that subjects didshow a slight tendency to overestimatedegree of control in the 75-50 problem.

The fact that subjects accurately judgedthe degree of contingency between theirresponses and outcomes suggests that theyknew the relevant conditional probabilitiesand organized them appropriately. A mea-sure of the discrepancy between judged per-centage of reinforcement if press and the

actual percentage of reinforcement if presswas used to assess the accuracy of subjects'knowledge of the conditional probabilities.A similar discrepancy score was computedfor the percentage of reinforcement if notpress.

Figure 2 shows that both depressed andnondepressed subjects were quite accuratein judging the percentage of reinforcementwhen they pressed the button. The mean dis-crepancy scores (judged — actual) for thethree problems were as follows: —9.50 for75-50, +.54 for 75-25, and +2.22 for 75-0.A Problem X Mood X Sex analysis of vari-ance on these discrepancy scores yielded asignificant main effect for problem, F (2, 84)= 8.45, p < .001, and no other significantmain effects or interactions. Post hoc Scheffetests revealed that subjects were significantlymore accurate in judging the percentage ofreinforcement in the 75-25 problem thanin the 75-50 problem (p < .05) and moreaccurate in the 75-0 problem than in the 75-50 problem (p < .05).

Similarly, both depressed and nonde-pressed subjects were quite accurate in judg-ing the percentage of reinforcement whenthey did not press. The mean discrepancyscores were as follows: —13.88 for 75-50,-3.84 for 75-25, and +2.03 for 75-0. AProblem X Mood X Sex analysis of vari-ance of the discrepency scores for not press

4 In computing the correlation between judgedcontrol and judged percentage of reinforcement,actual percentage of reinforcement was partialedout, because the study was designed so that actualpercentage of reinforcement correlated negativelywith actual control.


yielded a significant main effect for problem,F(2, 84) = 9.02, p < .001, and no othersignificant main effects or interactions. Posthoc Scheffe tests revealed that subjects weresignificantly more accurate in judging the75-25 problem than the 75-50 problem(p < .05) and more accurate in judging the75-0 problem than the 75-50 problem(/X.05).

In all three problems, one of the tworesponses was associated with 75% rein-forcement. We wished to determine whethersubjects were equally accurate in discerningthis conditional probability in the threeproblems. For those subjects who had press-ing associated with 75% reinforcement, themean discrepancy scores (judged reinforce-ment if press — actual reinforcement ifpress) were -14.06 for 75-50, +1.00 for75-25, and -.69 for 75-0. A Problem XMood X Sex analysis of variance on thesediscrepency scores revealed a significantmain effect for problem, F(2, 36) = 6.74,p < .003, and no other significant maineffects or interactions. Post hoc Scheffe testsshowed that subjects were more accurate indetecting the 75% reinforcement rate in the75-25 problem than in the 75-50 problem

100

80

60

40

20

75-50 75-25

Problem75-0

Figure 1. Judged control for depressed (D) andnondepressed (ND) students as a function ofproblem type in Experiment 1. (Actual degree ofcontrol [ACON] and actual percentage of rein-forcement [ARF] are also shown as a function ofproblem type.)

-6

-26

• Press

O Not Press

76-60 76-26

Problem76-0

Figure 2. Discrepancy between judged and actualpercentage of reinforcement for pressing and notpressing as a function of problem type in Experi-ment 1.

(p < .05) and more accurate in the 75-0problem than in the 75-50 problem (p <.05).

For those subjects who had not pressingassociated with 75% reinforcement, themean discrepancy scores (judged reinforce-ment if not press — actual reinforcement ifnot press) were -20.50 for 50-75, -9.06for 25-75, and +1.44 for 0-75. A ProblemX Mood X Sex analysis of variance of thesediscrepancy scores yielded a significant maineffect for problem, F(2, 36) = 6.70, p <.003, but no other significant main effectsor interactions. A post hoc Scheffe testshowed that subjects were significantly moreaccurate in the 0-75 problem than in the 50-75 problem (p < .01). In summary, althoughsubjects were in general quite accurate injudging the conditional probabilities asso-ciated with pressing and not pressing, it isclear that as the two conditional probabilitiesbecame more similar (i.e., as the degree ofcontingency decreased), the degree ofaccuracy decreased.

The final judgment scale that subjectscompleted measured their judgments of totalpercentage of reinforcement irrespective oftheir responses. Subjects were also highlyaccurate in making this judgment. The dis-crepancy between judged total reinforcement


and actual total reinforcement was verysmall for all three problems: —.22 for 75-50,+4.16 for 75-25, and +6.28 for 75-0. AProblem X Mood X Sex analysis of vari-ance of these discrepancy scores revealedno significant effects. Further, the product-moment correlation between judged totalreinforcement and actual total reinforcementwas .73 (p < .001).

Behavioral task. The results of the be-havioral task provided converging evidencethat subjects were sensitive to the degree ofcontingency between their responses andoutcomes. On the behavioral task, subjectsmost frequently performed the response thathad been associated with the higher prob-ability of green light onsets in the contin-gency problems. For example, if not pressinghad been associated with the higher percent-age of green light onsets in the contingencylearning problem, subjects tended to notpress more often than press on the behavioraltask.

Further, Figure 3 shows that subjects'tendency to perform the response that max-imized green light onsets in the behavioraltask varied directly with the degree of con-trol in the three problems. That is, subjectswere most likely to perform the responsethat maximized green light onsets in the 75-0 problem and least likely to perform thisresponse in the 75-50 problem. A ProblemX Mood X Sex X Response analysis of vari-ance on the number of correct responses inthe behavioral task yielded a significant maineffect for problem, F(2, 72) = 13.39, p <.001, and no other significant main effectsor interactions. Post hoc Scheffe testsshowed that subjects maximized onset of thegreen light more frequently in the 75-0 andthe 75-25 problems than in the 75-50problem (pa < .01).

Postexperimental questionnaire. Sub-jects' responses on the postexperimentalquestionnaire suggest that they relied on therelative efficacy of one response versus an-other as the crucial evidence for determina-tion of degree of control. A typical subjectassigned to the 25-75 problem, when askedwhat evidence convinced her that she hadsome degree of control, replied, "Because thegreen light went on more often when I sat

• Press

o Not Press

75-50 75-25

Problem

75-0

Figure 3. Number of correct responses in the be-havioral task as a function of problem type inExperiment 1 for subjects who had pressing asso-ciated with the higher percentage of reinforcement(press) and for subjects who had not pressingassociated with the higher percentage of reinforce-ment (not press).

back, and usually when I pressed the buttonit did not." The majority of subjects re-ported that they did not try any complexpatterns of responding such as alternationor double alternation of pressing and notpressing during the contingency problem.

Discussion

Taken together, the results of Experiment1 suggest that at least under some circum-stances, when a contingency between re-sponses and outcomes exists, subjectiverepresentations of contingencies mirror ob-jective contingencies across a wide range ofthe response-outcome contingency space (c.f.Seilgman et al., 1971; see also Ward &Jenkins, 1965, for another demonstration ofisomorphism between subjective and objec-tive contingencies). When presented with acontingency learning problem, students werenot only able to perform the response thatmaximized reinforcement, but more impor-tantly, they were able to judge accurately thedegree of contingency between their re-sponses and outcome. This result obtainedregardless of whether the response associatedwith the higher frequency of the outcomewas pressing or not pressing the button. Theresults also indicate that students knew theconditional probabilities relevant to a deter-mination of degree of control. Therefore, notonly were students sensitive to the relative


efficacy of one response versus the other,but they appeared to use this difference asthe basis for their judgments of control.

It should be noted, however, that only afew students actually subtracted one prob-ability from the other in judging degree ofcontrol. The majority of students did notsubtract formally but appeared to rely ona more "holistic" or "gestalt" impression ofthe difference between probabilities. Addi-tional support for the conclusion that stu-dents relied on the relative efficacy of oneresponse versus the other as providing thecritical evidence for degree of control wastheir verbalized logic on the postexperimen-tal questionnaire.

That students were able to judge accu-rately the degree of contingency betweentheir responses and outcomes and did not relyon an invalid heuristic such as percentage ofreinforcement does not agree with theresults reported by Jenkins and Ward(1965). The most probable explanation forthe difference between the two studies in-volves the operationalization of the degree ofcontingency itself. In Jenkins and Ward'sstudy, subjects were presented with tworesponse buttons, RI and Rg. Jenkins andWard determined the "actual degree of con-trol" in each of their contingency problemsby comparing the probability of the outcomewhen RI is made to the probability of theoutcome when RS is made. In their analysisof the actual degree of control, Jenkins andWard did not take into account the prob-ability of the outcome when no response(neither RI nor R2) was made. The fact thattrials were self-initiated in the Jenkins andWard experiment increases the likelihoodthat subjects did, indeed, consider the prob-ability of the outcome when no response(neither RI nor Rg) occurred. If the sub-jects in the experiment were "computing"this third conditional probability in additionto the two conditional probabilities asso-ciated with RI and R2, then degree ofcontrol would correlate with percentage ofsuccesses. It is possible that subjects inJenkins and Ward's experiment arrivedat their judgments of control by comparingthe probability of the outcomes' occurrence

when they did nothing with some average ofthe probability of the outcomes' occurrencewhen they pressed RI and R2. Given that theprobability of the outcomes' occurrence whenneither RI nor R2 was made was zero, thisdifference would, in fact, be very close to theactual percentage of successes the subjectreceived. If the preceding analysis is correct,subjects in both Jenkins and Ward's experi-ment and the present experiment used therelative efficacy of responses as the basis forjudgments of contingency.

Not only did the two experiments differin the operationalization of the concept ofcontingency but they also differed in thenumber of contingency problems presentedto each subject. Whereas subjects in Jenk-ins and Ward's study were presented withseveral contingency problems, subjects in thepresent experiment received only one prob-lem. Work in progress in our laboratory(Alloy & Abramson, Note 1) suggests thatexperience with one type of contingencyproblem systematically affects judgments ofcontrol on other types of contingency prob-lems. The above procedural differences mayaccount for the discrepancy between theresults of the present experiment and thoseof Jenkins and Ward.

Experiment 1 provided no evidence thatdepressed students distort response-outcomerelations in the manner predicted by thelearned helplessness theory of depression.Depressed and nondepressed students didnot differ in their judgments of contingency,and both groups were accurate. This resultis potentially important because it is basedon a method for assessing cognitive repre-sentations of contingency in depressed andnondepressed people that is more direct thanearlier methods (e.g., Klein & Seligman,1976; Miller & Seligman, 1973, 1975). Be-cause Experiments 2, 3, and 4 are also rele-vant to whether depressed and nondepressedindividuals differ systematically in their rep-resentations of contingencies, further con-sideration of the failure to confirm thelearned helplessness predictions in Experi-ment 1 is postponed until the General Dis-cussion. In Experiment 2, we examineddepressed and nondepressed students' judg-


merits of contingency in two problems inwhich responses and outcome were noncon-tingently related.

Experiment 2

Although both depressed and nonde-pressed students in Experiment 1 were quiteaccurate in judging the degree of controlwhen faced with problems in which therewas some degree of contingency, it waspossible that they would not be as accuratein detecting a lack of contingency betweentheir responses and outcomes. Indeed, anumber of investigators (Chapman & Chap-man, 1967; Langer, 1975; Smedslund, 1963;Starr & Katkin, 1969; Wortman, 1975)have found that people often believe a rela-tionship is contingent when it is noncontin-gent. In Experiment 2, students were pre-sented with one of two problems in whichresponses and outcomes were noncontin-gently related but differed in the overallfrequency of green light onset. If people usethe invalid heuristic of percentage of rein-forcement in the noncontingent case, thenthey will believe that they have more controlin the problem in which green light onsetoccurs frequently than in the problem inwhich green light onset occurs infrequently.

Affective state was not predictive of judg-ments of control in Experiment 1; however,the learned helplessness model of depressionagain predicts that judgments of control willvary systematically with affect in Experi-ment 2. Although several proponents of thelearned helplessness model (e.g., Klein &Seligman, 1976; Miller & Seligman, 1973)have argued that depressed and nondepressedpeople should not differ in their perceptionof noncontingent response-outcome relation-ships and that both groups will be accurate,we believe the model does not support thisprediction. On the contrary, the learnedhelplessness model of depression makes botha strong and a weak prediction concerningpeople's representations of contingency inthe noncontingent case. According to thestrong prediction, nondepressed peopleshould tend to overestimate the degree ofcontingency when responses and outcomes

are noncontingently related, whereas de-pressed people will be accurate. Accordingto the weak prediction, there simply will bea net difference between depressed and non-depressed people's judgments of noncontin-gency. The weak prediction does not requirethat depressed people accurately perceivenoncontingency, but merely that nonde-pressed people believe they have morecontrol than depressed people believe theyhave.

Helplessness theory explains difficulty indiscerning objective contingency by de-pressed people by proactive interferencefrom an expectation of uncontrollability.Similarly, because the model regards non-depressed people as having generalizedexpectations of control, these expectationsshould proactively interfere with the detec-tion of noncontingencies. Just as depressives'generalized expectation of response-outcomeindependence interferes with their ability toperceive that outcomes are now dependenton responses, nondepressives' generalizedexpectation that outcomes are dependent onresponses should interfere with their abilityto perceive that outcomes are independent ofresponses. Such reasoning has been used toexplain the immunization phenomenon inhelplessness (e.g., Maier & Seligman, 1976),Therefore, according to both the weak andstrong predictions of learned helplessnesstheory, nondepressed students should tendto show an "illusion of control" (Langer,1975) in Experiment 2.

Method

Subjects. Sixty-four undergraduates, 32 malesand 32 females, from the University of Pennsyl-vania served as paid volunteers. Using the sameBDI criterion as in Experiment 1, subjects wereassigned to a depressed or nondepressed group. Asin Experiment 1, subjects completed the MAACLas a further index of depression. The correlationbetween the BDI and MAACL scores in Experi-ment 2 was .44 (/><.OOS). The final sample con-sisted of 32 depressed students (16 males and 16females) and 32 nondepressed students (16 malesand 16 females). Table 4 presents the mean BDIand MAACL scores for all experimental groups.Subjects were randomly assigned to two experi-mental conditions, with the restriction that eachcondition contain equal numbers of males andfemales and depressed and nondepressed students.





and test

25-25BDIMAACL

75-75BDIMAACL

M

2.614.0

4.18.5

SD

2.65.3

3.04.2

M

5.012.5

5.514.4

SD

1.83.8

2.85.2

M

10.615.6

11.013.9

SD

1.85.2

2.42.4

M

15.019.8

15.121.0

SD

6.67.0

6.15.3


Half of the subjects in each condition were testedby Experimenter 1 and half by Experimenter 2.

Experimental design. The experiment was a 2(Problem Type) X 2 (Mood—depressed, non-depressed) X 2 (Sex—male, female) factorialdesign. The two problems differed in the percent-age of reinforcement but did not differ in degreeof control. In Problem 1 (25-25), subjects had nocontrol and were reinforced (green light onset)on 25% of the trials. In Problem 2 (75-75), sub-jects also had no control but were reinforced on75% of the trials. Of the 64 subjects, 32 (8 de-pressed males, 8 depressed females, 8 nondepressedmales, and 8 nondepressed females) were assignedto each problem.

Dependent measures. The dependent measureswere identical to those employed in Experiment 1.

Apparatus and procedure. The apparatus andprocedure were also identical to those employed inExperiment 1.

Results

Because the experiment was conducted bytwo different experimenters, it was necessaryto determine whether there were any experi-menter effects. Analysis of variance revealedno significant experimenter effects for anyof the dependent measures. Therefore, thisfactor was not used in the subsequent statis-tical analyses.

Judgment scales. Subjects' scores on theJudgment of Control scale revealed that non-depressed subjects showed an illusion ofcontrol in the 75-75 problem, but not in the25-25 problem, whereas depressed subjectswere relatively accurate in both problems. AProblem X Mood X Sex analysis of varianceof the judgment of control scores yielded sig-

nificant main effects for problem, F(\, 56)= 7.18, p < .01, and mood, F(l, 56) = 4.16,p < .05; a significant Problem X Moodinteraction, F(l, 56) = 5.54, p < .03; anda significant Problem X Mood X Sex inter-action, F(l, 56) = 4.40, p < .04.

Interpretation of the significant maineffects and double interaction is difficultwithout analysis of the significant tripleineraction. To analyze the triple interaction,Problem X Mood analyses of variance wereperformed separately for both males andfemales. The Problem X Mood analysis ofvariance conducted for females revealed asignificant main effect for problem, F(l, 28)= 6.90, p < .02, and a significant Problem XMood interaction, F(l, 28) = 9.94, p<.004. A simple main effects test (Winer,1962) revealed that nondepressed femalesjudged that they had more control in the 75-75 problem than in the 25-25 problem,F(l, 28) = 16.71, p < .001, whereas thejudgments of depressed females did not dif-fer between the two problems. In addition,nondepressed females' judgments of controlwere significantly greater then depressedfemales' judgments of control in the 75-75problem, F(l, 28) = 12.70, p < .005, butnot in the 25-25 problem. Although theProblem X Mood interaction was not sig-nificant for the males, their pattern of resultswas similar to that of the females. Table 5presents the mean judged control scores forall of the experimental groups.

To summarize, nondepressed subjects




Males

Problem

25-2575-75

M

20.030.3

SD

26.712.8

Females

M

7.551.4

SD

14.529.8

Males

M

13.821.2

SD

12.728.5

Females

M

17.113.1

SD

23.713.6

overestimated the degree of control in the75-75 noncontingency problem, but thoughtthey had little control in the 25-25 noncon-tingency problem. In contrast, the depressedsubjects made relatively accurate judgmentsabout both problems. The presence of thetriple interaction indicated that the effect ofoverestimating the degree of control in the75-75 problem was more pronounced fornondepressed females than for nondepressedmales. The triple interaction is presentedgraphically in Figure 4.

In addition to judging degree of controlfor one of the two problems, subjects alsorated how certain they were of the accuracyof their judgment of control ratings. AProblem X Mood X Sex analysis of variancerevealed no significant effects for thesecertainty ratings.

The finding that nondepressed subjectsdid not accurately judge the degree of con-tingency between their responses and greenlight onset suggests either that they did notknow the relevant conditional probabilitiesassociated with pressing and not pressing orthat they knew the relevant probabilities butorganized them inappropriately. Analysis ofthe scores measuring discrepancy betweenjudged percentage of reinforcement andactual percentage of reinforcement associatedwih pressing provides evidence that bothdepressed and nondepressed subjects wererelatively accurate in judging the conditionalprobabilities. The mean discrepancy scores(judged — actual) for percentage of rein-forcement associated with pressing in thetwo problems were +1.62 for 25-25 and-6.12 for 75-75. A Problem X Mood X Sexanalysis of variance of these discrepancy

scores yielded a significant problem effect,F(l, 56) = 4.32, p < ,05, and no other sig-nificant main effects or interactions. Themean discrepancy scores (judged — actual)

60

50

40

30

20

10

0

50

40

30

20

10

0

50

40 -

30 •

20 -

10 -

0

Males

Females

25-25 75-75

ProblemFigure 4. Judged control for depressed (D) andnondepressed (ND) students as a function ofproblem type in Experiment 2. (The Problem XMood X Sex triple interaction is also presented.)


-15

• PressO Not Press

25-25 75-75

Problem


for percentage of reinforcement associatedwith not pressing in the two problems were-5.12 for 25-25 and -11.44 for 75-75. AProblem X Mood X Sex analysis of vari-ance yielded no significant effects. Figure 5shows the discrepancy scores for subjects'judgments of the percentage of reinforce-ment associated with pressing and notpressing.

It appears, then, that all subjects did havethe appropriate data with which to makean accurate judgment of control. However,unlike the depressed subjects, the non-depressed subjects' actual judgments ofcontrol in the 75-75 problem were inaccu-rate. This suggests that the nondepressedsubjects did not rely on the differential effec-tiveness of responses in organizing theprobability data but rather may have relied

on some invalid organizational rule. As inExperiment 1, a correlational analysis of thejudgment of control data was performed inorder to determine whether subjects were, infact, using the invalid heuristics of percent-age of reinforcement, percentage of successes,or confirming cases. Because the analysis ofvariance showed that depressed and non-depressed subjects differed systematically in

their judgments of control, separate correla-tional analyses were performed for depressedand nondepressed subjects. Nondepressedsubjects' judgments of control correlatedsignificantly with the invalid heuristics ofjudged prcentage of reinforcement (r = .52,p < .005), percentage of successes (r = .49,p < .005), and confirming cases (r = .48,p < .005). Depressed subjects' judgments ofcontrol did not correlate significantly withjudged percentage of reinforcement (r =—.02), percentage of successes (r = —.01),or confirming cases (r = .15).

The final judgment scale to be reported issubjects' judgments of total reinforcementirrespective of their responses. The discrep-ancy between judged total reinforcement andactual total reinforcement was very small forboth problems: -1.16 for 25-25, and -3.33for 75-75. A Problem X Mood X Sexanalysis of variance of these discrepancyscores yielded no significant effects or inter-actions. In addition, the correlations betweenjudged total reinforcement and actual totalreinforcement, based on all subjects, wassignificant (r - .94, p < .0001). Thus, sub-jects' judgments of total percentage of rein-forcement were highly accurate.

Behavioral task. Although there is no"correct response" for maximizing greenlight onset in the behavioral task when out-comes are noncontingently related to re-sponses, it is interesting to determinewhether any systematic differences emergein these data. A Problem X Mood X Sexanalysis of variance of the number of pressesin the behavioral task yielded no significantmain effects or interactions. Neither moodnor problem type affected subjects' perform-ance in the behavioral task. In general, themajority of subjects pressed the button anintermediate number of times, although afew subjects pressed rarely and others mostof the time.

Postexperimental questionnaire. UnlikeExperiment 1, a number of subjects reportedthat they tried complex patterns of respond-ing while engaged in the problems of Experi-ment 2. A chi-square analysis revealed thatthe nondepressed subjects reported tryingcomplex hypotheses more often than did the


depressed subjects, x"(l) = 3.14, p < .08,Seventeen of the nondepressed subjects and10 of the depressed subjects reported tryingcomplex hypotheses.

Subjects also answered open-ended ques-tions about the kind of evidence that con-vinced them that they either did or did nothave control. In the 25-25 problem, bothdepressed and nondepressed subjects tendedto use appropriate logic in determining theirjudgments of control. For example, a de-pressed female in the 25-25 problem said,"I believe I had no control because I trieddifferent sequences of pressing and non-pressing which didn't make any difference. Iwould have believed I had some control ifthe green light went on or off in accord withmy actions—at least to some degree."

It is interesting that nondepressed sub-jects often used incorrect logic in arrivingat their judgments of control in the 75-75problem. The following examples illustratethe kinds of errors in reasoning that char-acterized the judgments of the nondepressedsubjects. Some of the nondepressed subjectswho showed an illusion of control seemed touse "intuition" rather than logic in judgingthat they had control: "I believe I had somedegree of control. There was no objectiveevidence. I merely worked on intuition,whim, and so forth." Other nondepressedsubjects' logic indicated that they were usingthe invalid heuristic of percentage of rein-forcement : "I believed I had control becausethe green light came on a lot. I would havebelieved I had no control if the green lightnever came on." Finally, another subset ofnondepressed subjects believed that verycomplex patterns of responses were respon-sible for the high frequency of green lightonset: "Counting in intervals of five andjudging how the light reacted at a certaintime within that interval led me to believeI had control." Thus, the illusion of controlshown by nondepressed subjects in the 75-75 problem was often associated with in-appropriate reasoning.

Discussion

Although the results of Experiment 1strongly support the idea that humans' rep-

resentations of contingency mirror objectivecontingencies, the findings of Experiment 2suggest that when responses and outcomesare noncontingently related, subjective repre-sentations of contingencies are not alwaysisomorphic with objective contingencies.Both affective state and frequency of rein-forcement interacted to produce systematicerrors in judgment of noncontingency. Non-depressed individuals showed an illusion ofcontrol in a noncontingent, high-densityreinforcement situation, but not in a noncon-tingent, low-density reinforcement situation,Depressed individuals, on the other hand,were not affected by the density of reinforce-ment in situations in which outcomes werenoncontingently related to responses. Theyaccurately detected lack of relation betweenresponses and outcomes.

Both in the contingency problems of Ex-periment 1 and in the noncontingent, low-density reinforcement problem of Experi-ment 2, all subjects (including the non-depressives) tended to use appropriate logicin arriving at their judgments of conrol. Inthe noncontingent, high-density reinforcementproblem, however, the logic of nondepressedstudents' judgments seemed to break down.Although nondepressed students' judgmentsof control correlated with actual degree ofcontrol in Experiment 1, in Experiment 2they correlated with invalid heuristics suchas how frequently the outcome occurredoverall, how frequently the outcome occurredwhen they pressed the button, and how fre-quently the outcome occurred when theypressed plus how frequently it did not occurwhen they did not press.

Nondepressed students' use of invalidheuristics in arriving at their judgments ofcontrol was corroborated by their answersto open-ended questions that probed theirreasoning. Some nondepressed students ap-peared to abandon all rational strategies infavor of intuitive strategies in arriving atjudgments of control. Moreover, students'judgments of the conditional probabilitiesclearly indicate that it is generally the organ-ization of the conditional probability data,rather than the perception of the data them-selves, that is responsible for failure todetect noncontingency. Accordingly, Ward


and Jenkins (1965) found that when theconditional probability data were organizedfor the subjects in tabular form, rather thanpresented on a trial-by-trial basis, sub-jects' judgments of contingency improvedmarkedly.

The results of Experiment 2 partiallyconfirm the strong prediction of the helpless-ness model of depression. Although de-pressed students were relatively accurate indetecting a lack of contingency between theirresponses and outcomes, nondepressed stu-dents showed an illusion of control whenthe outcome occurred frequently, but notwhen the outcome occurred infrequently. Inthe 25-25 problem, 50% of both depressedand nondepressed students actually said thatthey had zero control. In the 75-75 problem,50% of depressed students, but only 6%of nondepressed students, said that they hadzero control.

What is the important psychologicaldimension that distinguishes between the 25-25 and 75-75 problems? Of course, bydefinition, the two problems differed on fre-quency of the outcome's occurrence. Stu-dents may, however, have construed highfrequency of reinforcement as a "good" out-come and low frequency of reinforcement asa "bad" outcome. Informal conversationswith the subjects during the debriefingrevealed that they did, in fact, view "good-ness" of outcome as being determined by thefrequency of green light onset. Thus, valenceof the outcome, rather than frequency perse, may have been the crucial psychologicalvariable underlying nondepressed students'errors in the judgment of noncontingency.

It is interesting that the illusion of controlwas much more pronounced in nondepressedfemales than in nondepressed males. Thissex difference is puzzling from the standpointof learned helplessness because the modelmakes no differential predictions concerningsex. Langer's (1975) study on the illusionof control provides a clue for interpretationof the sex effect in Experiment 2. Langernoted that males, unlike females, did notshow illusions of control in those studies inwhich demand for rationality was very high.This explanation pertains to the presentstudy because rational judgments of contin-

gency were strongly emphasized in theexperimental instructions. This reasoningleads to the prediction that nondepressedmales would show a bigger illusion of controlrelative to depressed people if there wereless focus on rationality in the experimentalinstructions.

The considerations of this discussionprompted us to perform a third experimentin which we further examined the illusionof control. First, obtaining an illusion orcognitive distortion in a "normal" group isa surprising finding and warrants replicationin a different context. Second, valence of theoutcome needs to be specifically manipulatedwhile holding frequency of reinforcementconstant because outcome valence may becrucial in accounting for differences injudged contingency between depressed andnondepressed people. Finally, a situation inwhich less demand for rationality is placedon the subject may be required in order toenhance the illusion of control.

Experiment 3

Experiment 3 further examined the illu-sion of control in a situation that was lesshighly structured and more like the "realworld." In Experiment 2, students' majortask was to determine the degree of controltheir responses exerted over green lightonset. In everyday life, however, detectingenvironmental contingencies occurs in thelarger context of learning how to obtaindesired outcomes and avoid aversive out-comes. Thus, Experiment 3 examined judg-ment of contingency in a situation in whichoutcomes were response independent andstudents were focused on obtaining monetaryrewards. Students were informed from theoutset that they would be asked to judgehow much control they had at the end of thetask.

Experiment 3 also investigated the impor-tance of the valence of the outcome as a psy-chological determinant of the illusion ofcontrol. Depressed and nondepressed stu-dents were assigned to one of two problemssimilar to those of Experiment 2, exceptthat green light onset was associated witheither gain of loss of money. At the same





and test

LoseBDIMAACL

WinBDIMAACL

M

4.49.1

3.28.5

SD

1.84.5

1.83.2

M

2.88.0

3.19.0

SD

2.43.5

2.93.3

M

14.220.0

14.518.9

SD

3.54.5

4.63.6

M

18.523.0

13.022.1

SD

8.97.0

5.43.2


time, frequency of green light onset was heldconstant across the two problems. Changesin mood were examined to determinewhether this outcome valence manipulaionwas effective for both depressed and non-depressed students.

In Experiments 1 and 2, the Beck Depresrsion Inventory, which measures enduringdepression, was used to select depressed andnondepressed students. For most students,scores on the BDI correlated highly withinitial depression scores on the MultipleAffect Adjective Check List, a measure oftransient mood. A few subjects, however,scored high on one inventory and low on theother. To insure separation of depressed andnondepressed groups, the initial depressionscore of the MAACL was also used as amood selection criterion in Experiment 3.

Method

Subjects. Sixty-four undergraduates, 32 malesand 32 females, from the University of Pennsyl-vania served as paid volunteers. The data of onesubject were discarded because of an error inexperimental procedure. Subjects were assigned todepressed or nondepressed groups on the basis ofboth BDI and MAACL scores. The BDI criterionwas identical to that of Experiments 1 and 2. Inaddition, subjects had to score 14 or above, or 13or below on the MAACL to qualify as depressedor nondepressed, respectively. Subjects who did notmeet both criteria did not qualify for the experi-ment. Given that subjects were selected on the basisof both BDI and MAACL scores, it is not sur-prising that the correlation between scores washigher than in Experiments 1 and 2 (r = .84, P <

.001). Table 6 presents the mean BDI andMAACL scores for all experimental groups. Thefinal sample of subjects consisted of 32 depressedstudents (16 males and 16 females) and 32 non-depressed students (16 males and 16 females). Sub-jects were randomly assigned to the two experi-mental conditions, with the restriction that eachcondition contain equal numbers of males andfemales and depressed and nondepressed students.Half of the students in each condition were testedby Experimenter 1 and half by Experimenter 2.

Experimental design. The experiment was a 2(Problem Type) X 2 (Mood) X 2 (Sex) factorialdesign. The two problems differed in the valenceof the outcome, but did not differ in either degreeof control or frequency of the outcome. In Problem1 (lose), subjects had no conrol and were rein-forced (green light onset) on 50% of the trials.Subjects in this problem started out with $5. Eachtime the green light did not come on, subjectslost a quarter; when the green light did come on,no money was lost. In Problem 2 (win), subjectshad no control and were also reinforced (greenlight onset) on 50% of the trials. Subjects in thisproblem began with no money and gained a quartereach time the green light came on. On trials onwhich the green light did not come on, subjects didnot gain any money. Thus, all subjects in the losecondition lost $5, and all subjects in the win condi-tion won $5. Of the 64 students, 32 (8 depressedmales, 8 depressed females, 8 nondepressed males,and 8 nondepressed females) were assigned to eachcondition,

Dependent measures. With the exception of thebehavioral task, all dependent measures employedin Experiments 1 and 2 were also employed inExperiment 3. In addition, the MAACL wasadministered both before and after the problems todetermine whether subjects' moods were affectedby the outcome valence manipulation. The MAACLyielded three affect dependent measures: an anxietychange score, a depression change score, and a


hostility change score. An additional question wasadded to the postexperimental questionnaire toprovide another measure of subjects' beliefs aboutcontrol: "To what extent do you think that onsetof the green light was due to factors other thanyour own responding?"

Apparatus. The apparatus for Experiment 3 wasidentical to that of Experiments 1 and 2, with oneaddition. A Kodak 6SO H Carousel slide projectorwith a remote control panel presented slides depict-ing the amount of money subjects had at any pointin time. The slides were projected onto a 75 cm X60 cm screen positioned on the wall approximately90 cm from the subject. The remote control panelwas operated from the control room of the two-room suite.

Procedure. When the subject entered the roomhe or she was seated at a desk and was admin-istered the BDI and MAACL according to theprocedure of Experiment 1. Upon completion ofthe mood inventories, the subject was assigned tothe appropriate group and led to a table on whichthe apparatus for the contingency learning problemwas displayed. The procedure for the win and loseproblems was identical to that of the problems inExperiment 1, with the addition that a new slidedepicting the subject's current earnings was pre-sented on each trial after the green light had eitherappeared or not appeared. The instructions for thewin and lose problems were similar to those ofExperiment 1. The major difference in the instruc-tions was that in Experiment 1 subjects were toldthat their major task was to learn the degree ofcontrol their responses had over the onset of thegreen light whereas in Experiment 3 their osten-sive major task was to learn how to turn on thegreen light. This difference was reflected in achange in two sentences of the original instructions.The first sentence was changed from "Now, in thisproblem-solving experiment, it is your task to learnwhat degree of control you have over whether ornot this green light comes on" to "Now, in thisproblem-solving experiment, it is your task to learnhow to turn on this green light." The sixth sentenceof the second paragraph was changed from "Sinceit is your job to learn how much control, you haveover whether the green light comes on, as wellas whether the green light does not come on, itis to your advantage . . ." to "Since it is yourjob to learn how to turn on the green light, it isto your advantage. . . ."

An additional paragraph concerning the gain orloss of money aspect of the experiment was insertedbetween paragraphs 2 and 3 of the original instruc-tions, before the judgment of control instructions.The paragraph for the win problem was:

Moreover, how often the green light comes on inthis problem will determine how much moneyyou earn in the experiment. On each trial onwhich the green light does go on, you will earna quarter. Alternatively, on each trial on whichthe green light does not go on, you will not

earn any money. At the end of the problem, youwill get to keep all of the money you haveearned up to a maximum of $5. So, in general,the more successful you are in producing thegreen light, the more money you will take awaywith you at the end of the experiment. Duringthe problem, you will be able to see how muchmoney you currently own by watching the greenscreen. The slide projector to your left will keeptrack of your accumulation, if any, of quarters.

The paragraph for the lose problem was:

Moreover, how often the green light comes onin this problem will determine how much moneyyou earn in the experiment. I will start you offwith $5. On each trial on which the green lightdoes not go on, you will lose a quarter. Altern-atively, on each trial on which the green lightdoes go on, you will not lose any money. At theend of the problem, you will get to keep all ofthe money you have left. Of course, the maximumamount of money you can lose is $5. I will notask for any money out of your pocket. In generalthough, the more successful you are in producingthe green light, the more money you will takeaway with you at the end of the experiment.During the problem, you will be able to see howmuch money you currently own by watching thegreen screen. The slide projector to your leftwill keep track of your loss, if any, of quarters.

After the judgment of control instructions wereread, the experimenter left the room and the sub-ject proceeded with the contingency learningproblem. At the end of the 40 trials, the experi-menter returned and paid the subject $5 orremoved $5, depending on the problem to whichthe subject had been assigned. The control instruc-tions were then reread, and the subject completedeach of the four judgment scales. Following thejudgment scales, the subject completed the MAACLfor a second time, followed by the postexperimentalquestionnaire. The experimenter left the roomwhile the subject completed these questionnaires.Finally, the subject was carefully debriefed andgiven $5 for participating if he or she had beenin the lose condition. Subjects in the win condi-tion were allowed to keep the $5 they had alreadyreceived.

Results

As in Experiments 1 and 2, analysis ofvariance showed no experimenter effects onany of the dependent measures, and thisfactor was dropped in subsequent statisticalanalyses.

Judgment scales. A Problem X Mood XSex analysis of variance of the judgment ofcontrol scores yielded significant main effectsfor problem, F(l, 56) = 31.97, p < .001;




Males

Problem

LoseWin

M

6.949.4

SD

11.929.3

Females

M

21.264.4

SD

8.89.8

Males

M

21.217.2

SD

20.316.5

Females

M

11.436.0

SD

13.128.1

mood, F(l, 56) = 8.88, p < .004; and sex,F(l, 56) = 4.14, p < .05; and a significantProblem X Mood interaction, F(l, 56) =11.96, p< .001. The significant main effectfor sex indicated that females thought thatthey had more control than males. Tointerpret the significant main effects forproblem and mood, it is necessary to analyzethe Problem X Mood interaction. To analyzethe interaction, a simple main effects test(Winer, 1962) was used. Nondepressedsubjects judged that they had more controlin the win problem than in the lose problem,F(l, 56) =41.53, p< .001, whereas de-pressed subject' judgments did not differbetween the two problems. In addition,nondepressed subjects judged that they hadmore control than did depressed subjects inthe win problem, F(l, 56) = 20.73, p<.001, but not in the lose problem. Table 7presents the means and standard deviationsfor the judgment of control ratings for allexperimental groups. In addition, Figure 6graphically portrays the Problem X Moodinteraction.

A Problem X Mood X Sex analysis ofvariance of the certainty ratings also revealeda significant Problem X Mood interaction,F(l, 56) = 5.87, p < .02. Depressed sub-jects were more certain of their judgmentsof control in the win problem than in theloss problem. Nondepressed subjects, on theother hand, were more certain of their judg-ments in the lose problem than in the winproblem.

Figure 7 shows that subjects again wererelatively accurate in judging the conditionalprobabilities of green light onset associatedwith pressing and not pressing. The mean

scores of discrepancy between judged per-centage of reinforcement and actual percent-age of reinforcement associated with pressingwere —1.38 for the win problem and —12.97for the lose problem. A Problem X Mood XSex analysis of variance of these discrepancyscores revealed a significant problem maineffect, F(l, 56) =9.50, p < .003, and noother significant main effects or interactions.The mean discrepancy scores (judged —actual) for percentage of reinforcement asso-ciated with not pressing were —6.50 for thewin problem and —8.69 for the lose prob-lem. No significant effects were obtained ona Problem X Mood X Sex analysis of vari-ance of these scores.

In Experiment 2, nondepressed subjects

8"8

100

80

60

40

20

• ND

O D

Lose Win

Problem

Figure 6. Judged control for depressed (D) andnondepressed (ND) students as a function of prob-lem type in Experiment 3.


-20

• Press

O Not Press

Lose Win

Problem


appeared to use the invalid heuristics of per-centage of reinforcement, percentage ofsuccesses, and confirming cases whereasdepressed subjects did not. These threeheuristics are intercorrelated in that theyall rely on the common element of the fre-quency of green light onset. If frequency ofgreen light onset were the psychologicaldimension underlying nondepressed subjects'judgments of control in Experiment 3, thenjudgments of control again should have beencorrelated with these heuristics for non-depressed subjects. No such significant cor-relations were obtained, however, for eitherdepressed or nondepressed subjects. If thenondepressed subjects had only been usingthese heuristics, they would have over-estimated the degree of control in both thewin and lose problems, since both problemshad a high frequency of reinforcement(50%). The obtained result of a Problem XMood interaction on the judged controlscores points to the psychological importanceof the valence of the outcome for the non-depressed subjects over and above the fre-quency of the outcome.

Although subjects' judgments of total

reinforcement irrespective of their responseswere highly accurate in Experiments 1 and2, the outcome valence manipulation of thepresent experiment affected these judgmentsadversely. A Problem X Mood X Sex anal-ysis of variance of the scores measuring thediscrepancy between judgment of total rein-forcement and actual total reinforcementand actual total reinforcement yielded a sig-nificant main effect for problem, F(l, 56) =10.07, p < .002. In general, subjects accu-rately estimated the frequency of green lightonset in the win problem (discrepancy score= —.62), but underestimated green lightfrequency in the lose problem (discrepancyscore = —8.66). However, a marginally sig-nificant Problem X Mood interaction, F(l,56) = 3.76, p < .06, shows that this effectwas more pronounced in nondepressed thanin depressed subjects.

Affect change scores. All subjects experi-enced mood changes as a function of outcomevalence, suggesting that the outcome valencemanipulation was effective. A Problem XMood X Sex analysis of variance was per-formed on each of the three affect changescores (anxiety, depression, and hostility).For the anxiety change score, significantmain effects for problem, F(l, 56) = 6.53,p < .02, and mood, F(l, 56) = 11.80, p <.001, were obtained, with no other significantmain effects or interactions. The analysis ofvariance for change in depression alsoyielded significant main effects for problem,F(l, 56) = 10.60, p < .002, and mood, F(l,56) = 16.14, p < .001. Finally, the analysisof variance for the hostility change scoresrevealed a significant main effect for prob-lem, F(l, 56) = 4.62, p < .04, and a mar-ginally significant main effect for mood, F(l,56) = 3.88, p < .06. As can be seen in Fig-ure 8, all subjects became more dysphoric inthe lose problem than in the win problem.In addition, the main effect for mood isprobably best interpreted as representing aconstraint placed on subjects' affect changescores by their initial MAACL scores. Thatis, depressed subjects showed larger changesin the euphoric direction in the win problem,and nondepressed subjects showed larger


(0

6

3

2

1

0

-1

-2

5

4

3

2

1

0

8 -i

Anxiety

Depression

Hostility

Lose WinProblem

Figure 8. Anxiety, depression, and hostility changescores for depressed (D) and nondepressed (ND)students as a function of problem type in Experi-ment 3. (Positive values indicate affect changes inthe dysphoric direction, and negative values indi-cate affect changes in the euphoric direction.)

changes in the dysphoric direction in thelost problem.5

Postexperimental questionnaire. In gen-eral, the analysis of the questionnaire datacorroborated the findings already reportedfor judgments of control. A Problem XMood X Sex analysis of variance on thequestion assessing the degree to which sub-jects believed that factors other than theirown responding produced green light onset

revealed a significant problem effect, F(\,56) = 13.14, p < .025, and a significantProblem X Mood interaction, F(l, 56) =4.70, p < .04. A simple main effects testshowed that nondepressed subjects believedthat factors other than their own respondingwere significantly more important in the loseproblem than in the win problem, F(l, 56)= 10.12, p < .005, whereas there was nodifference between the problems for de-pressed subjects. In addition, depressed sub-jects believed that factors other than theirown responding were significantly moreimportant in the win problem than did thenondepressed subjects, F(l, 56) =4.18, p< .05. There was no difference betweendepressed and nondepressed subjects in thelose problem. Thus, subjects' answers on the"Factors" question were consistent withtheir answers on the Judgment of Controlscale.

It is interesting that in the open-endedquestions assessing students' logic in arriv-ing at their judgments of control, non-depressed subjects often reported that win-ning the $5 was sufficient evidence forconcluding that they had control. Depressedsubjects, on the other hand, did not use thistype of reasoning. Finally, unlike Experi-ment 2, there was no difference between thenumber of depressed and nondepressed sub-jects who reported trying complex hypoth-eses. Only 9 of the total 64 subjects reportednot trying complex patterns of responding.

Discussion

Experiment 3 provided strong corrobo-rative evidence for the hypothesis developed

5 Although one might think that analysis ofcovariance should be used to equate for initialMAACL scores, we believe this statistical techniqueis inappropriate because the depressed and non-depressed groups were selected parially on thebasis of their initial MAACL depression score. Inaddition, since the MAACL initial anxiety andhostility scores tended to correlate highly with theinitial depression score, it was deemed inappro-priate to use the analysis of covariance on thesechange dependent measures as well. By this samelogic, analysis of covariance was not used inExperiment 4.


in Experiment 2: Depressed people accu-rately detect noncontingency between theirresponses and outcomes whereas non-depressed people show illusions of control.It is interesting that nondepressed peopleshowed more pronounced illusions of controlin situations in which they were attemptingto obtain desired outcomes than in situationsin which they were judging the degree ofcontrol over more neutral outcomes. In addi-tion, recall that in Experiment 2, non-depressed males showed only a slight tend-ency to overestimate the degree of contin-gency in the noncontingent, high-densityreinforcement situation whereas in Experi-ment 3 their illusion of control was consider-ably larger and, in fact, quite pronounced.This finding is in line with Langer's (1975)hypothesis that males are more susceptibleto illusions of control in situations in whichthere is less emphasis on rationality.

The results of Experiment 3 demonstratedthat overestimations of control in non-depressed people are a function of thevalence of the outcome of interest. Non-depressed students believe that they have ahigh degree of control in situations in whichthey obtain desired outcomes noncontin-gently, but believe that they have very littlecontrol in situations in which they losedesired outcomes noncontingently.

The importance of the role of valence,rather than mere frequency, of the outcomein determining judgments of contingency inExperiment 3 is supported by the actualpattern of judged control scores for non-depressed students. If frequency of thedesired outcome were the sole determinantof nondepressed students' judgments ofcontrol in situations of response-outcomeindependence, then they should have ratedthat they had about 50% control (i.e., showan illusion) in the lose problem as well asin the win problem. The obtained result wasthat they only showed an illusion in the winproblem. Lack of correlation between invalidheuristics involving frequency of the desiredoutcome and judged control scores for non-depressed students provides further supportfor the role of outcome valence as a deter-minant of judgments of control.

Moreover, the finding that nondepressedstudents overestimated the degree of controlonly in the win problem and not in the loseproblem provides partial, but not complete,support for the learned helplessness modelof depression. The model makes no differ-ential predictions concerning outcome val-ence. Everyday observation suggests that inthe real world when bad outcomes happento an individual he or she often has nocontrol over them, and when good outcomeshappen the individual often does havecontrol. This observation is most likely dueto the fact that when individuals actuallyhave control, they do not allow bad outcomesto occur. Perhaps, then, past experiencewith having control over good outcomes butnot bad outcomes acts in conjunction withthe generalized expectancy of control pre-dicted for nondepressives by the helplessnesstheory to enhance the illusion of control overgood outcomes and to attenuate the illusionof control over bad outcomes (see Bern,1972, for a related discussion of attributionalstyle).

Perhaps one could explain the failure ofdepressed students to show an illusion ofcontrol in the win problem similar to non-depressed students by arguing that winning$5 was not as "good" an outcome fordepressed students as for nondepressed stu-dents. Indeed, Costello (1972) views rein-forcer ineffectiveness as the sine qua non ofdepression. Such an explanation is discon-firmed, however, by the results of the affectchange scores. On all three of the affect mea-sures, depressed students showed at leastas great an enhancement of mood in the wincondition as the nondepressed students, ifnot greater. Thus, the outcome valencemanipulation was equally as effective fordepressed students as for nondepressedstudents.

Finally, in addition to judgments of con-trol, judgments of reinforcement or outcomefrequency also were adversely affected bythe outcome valence manipulation of Experi-ment 3. All students believed that the desiredoutcome of green light onset was more fre-quent in the win problem than in the loseproblem; however, this effect was more


pronounced in the nondepressed than in thedepressed students. Thus, in hedonicallycharged situations in which outcomes arenoncontingently related to responses, non-depressed individuals show two kinds oferrors: distortions in judgments of contin-gency and distortions in judgments of out-come frequency.

Experiment 4

The finding in Experiments 2 and 3 thatnondepressed people overestimated thedegree of control between their responsesand desirable outcomes whereas depressedpeople were relatively accurate, supports thenew prediction of the helplessness model.This finding provides a unique kind of sup-port for helplessness theory, since previoustests of the theory only have looked atdepressive deficits in learning contingentrelationships. Failure to observe underesti-mation of degree of control by depressedstudents in Experiment 1 while at the sametime observing overestimation of degree ofcontrol by nondepressed students is puzzlingfrom the standpoint of helplessness theory.Because helplessness theory depends soheavily on absolute or relative depressivedeficits in contingency learning, it is impor-tant to determine whether under some condi-tions depressive do, in fact, underestimatethe degree of contingency between theirresponses and outcomes. Perhaps depressedstudents would underestimate the degree ofcontrol that their responses exerted overcontingent outcomes if these outcomes werehedonically charged. The fact that the non-depressive illusion of control was morerobust in Experiment 3 than in Experiment2 supports this line of reasoning. Thus, inExperiment 4 students received one of twoproblems, both of which had a 50% degreeof contingency between responses and greenlight onset, but which differed in outcomevalence.

Method

Subjects. Sixty-four undergraduates, 32 malesand 32 females, from the University of Pennsyl-vania served as paid volunteers. Procedures for

subject selection were identical to those in Experi-ment 3. As in Experiment 3, the correlation be-tween the BDI and the MAACL was quite highdue to the selection procedure (r = .78, /><.001).Table 8 presents the mean BDI and MAACLscores for all experimental groups. The final sampleof subjects consisted of 32 depressed students (16males and 16 females) and 32 nondepressed students(16 males and 16 females). Subjects were randomlyassigned to the two experimental conditions, withthe restriction that each condition contain equalnumbers of males and females and depressed andnondepressed students. One experimenter tested allsubjects."

Experimental design. The experiment was a 2(Problem Type) X 2 (Mood) X 2 (Sex) x 2 (Re-sponse) factorial design. The two problems differedin the valence of the outcome, but did not differ indegree of control. In Problem 1 (lose), subjectshad 50% control (75-25 or 25-75). Subjects in thelose problem started out with $4. Each time thegreen light did not come on subjects lost 30 cents.On trials in which the green light did come on, nomoney was lost. In Problem 2 (win), subjects alsohad 50% control (75-25 or 25-75). Subjects in thewin problem began with no money, and each timethe green light came on they gained a dime. Ontrials in which the green light did not come onsubjects did not gain any money. Thus, all subjectsin the lose condition lost between $3.50 and $4,and all subjects in the win condition won between$3.50 and $4. Within each of the problems, subjectswere counterbalanced for whether pressing or notpressing led to the higher percentage of green lightonset (i.e., 75-25 or 25-75).

Dependent measures. All dependent measureswere identical to those in Experiment 3.

Apparatus and procedure. The apparatus andprocedure were identical to those in Experiment 3,with the exception of specific instructions concern-ing the amount of money that could be won or lost.

Results

Judgment scales. A Problem X Mood XSex X Response analysis of variance of thejudgment of control scores yielded significantmain effects for problem, F(l, 48) = 13.45,p< .001; and response, F(l, 48) = 11.02,p < .002; and significant Problem X Mood,F(l, 48) = 6.90, p< .02; and Mood X Re-sponse, F(l, 48) =3.87, p < .055, inter-actions. To interpret the obtained maineffects, it is necessary to analyze the inter-actions involving these factors. To analyzethe Problem X Mood interaction, a simple

6 We would like to thank Linda Kanefield forconducting Experiment 4.


Table 8Means and Standard Deviations of BDI and MAACL Scores by Problem,Mood, Sex, and Response for Experiment 4


Problem

Lose

Win

Responseand test

75-25BDIMAACL

25-75BDIMAACL

75-25BDIMAACL

25-75BDIMAACL

Males

M

4.59.8

2.09.0

1.59.5

4.210.5

SD

1.74.5

2.40.7

1.72.7

2.52.1

Females

M

5.88.5

3.29.2

3.06.5

2.811.2

SD

2.63.5

2.62.8

1.42.7

1.71.1

Males

M

11.521.0

12.018.5

12.520.0

11.515.5

SD

3.14.4

2.93.4

2.12.8

2.12.1

Females

M

11.219.2

13.225.2

13.221.2

12.025.0

SD

1.03.4

4.61.3

3.95.1

2.73.3

Note. Response refers to whether pressing or not pressing the button was associated with the higher percent-age of reinforcement. BDI = Beck Depression Inventory, MAACL = Multiple Affect Adjective Check List.

main effects test was used (Winer, 1962).Nondepressed subjects judged that they hadless control in the lose problem than in thewin problem, F(l, 48) = 19.60, p < .001,

100

80

TJ

S)TJ

20

• NO

O D

ACON

Lose Win

ProblemFigure 9. Judged control for depressed (D) andnondepressed (ND) students as a function ofproblem type in Experiment 4. (Actual degree ofcontrol [ACON] is also shown as a function ofproblem type.)

whereas depressed subjects' judgments didnot differ between the two problems. Inaddition, nondepressed subjects judged thatthey had less control in the lose problemthan did depressed subjects, F(l, 48) =8.46, p < .01, but not in the win problem.Figure 9 portrays the Problem X Moodinteraction.

A simple main effects test on the Mood XResponse interaction showed that non-depressed subjects judged that they had lesscontrol when not pressing was associatedwith 75% reinforcement (25-75) than whenpressing was associated with 75% reinforce-ment (75-25), F(l, 48) = 14.40, p < .001.The judgments of depressed subjects werenot affected by whether pressing or notpressing was associated with 75% reinforce-ment. In addition, nondepressed subjectsjudged that they had significantly less con-trol when not pressing was associated with75% reinforcement than did depressed sub-jects, F(l, 48) = 5.78, p < .05, but notwhen pressing was associated with 75%reinforcement. Table 9 presents the meansand standard deviations for the judgment ofcontrol ratings for all experimental groups.

A Problem X Mood X Sex X Responseanalysis of variance on the certainty ratings


Table 9Means and Standard Deviations of Judged Control Scores by Problem,Mood, Sex, and Response for Experiment 4

Nondepressed

Problem Response

Depressed


M SD M SD M SD M SD

Lose

Win

7S-2S25-75

75-2525-75

53.86.2

75.041.2

36.47.5

12.029.5

42.56.2

67.066.2

16.69.5

15.532.0

54.544.2

64.563.5

15.413.4

12.622.5

52.549.5

56.040.0

36.923.3

18.625.5

Note. Response refers to whether pressing or not pressing the button was associated with the higher per-centage of reinforcement.

revealed a significant main effect for sex,F(l, 48) = 4.08, p < .05. Females were lesscertain of their judgments of control thanmales.

Depressed subjects were more accuratethan nondepressed subjects in judging theconditional probability of green light onsetassociated with pressing the button. A Prob-lem X Mood X Sex X Response analysis ofvariance of the mean discrepancy (judged —actual) scores for percentage of reinforce-ment associated with pressing yielded a sig-nificant Mood X Response interaction, F(l,48) = 6.06, p < .02. For the nondepressives,the mean discrepancy scores associated withpressing were +4.12 for 75-25 and —5.75for 25-75. For depressives, the mean dis-crepancy scores were —.50 for 75-25 and+ 1.44 for 25-75. A simple main effects teston these scores showed that nondepressives'discrepancy scores differed between 75-25and 25-75, F(l, 48) = 8.46, p < .01, where-as depressives' scores did not differ. Withinthe 75-25 condition, the discrepancy scoresof depressed and nondepressed subjects didnot differ, whereas they did differ in the 25-75 condition, F(l, 48) = 4.48, p < .05. Themean discrepancy (judged-actual) scoresfor percentage of reinforcement associatedwith not pressing were —.38 for the winproblem and +.44 for the lose problem. Nosignificant main effects or interactions wereobtained on a Problem X Mood X Sex XResponse analysis of variance of the meandiscrepancy scores for percentage of rein-

forcement associated with not pressing.Figure 10 portrays these discrepancy scores.

By the same logic as in Experiment 3, useof the invalid heuristics of percentage ofreinforcement, percentage of successes, andconfirming cases cannot be invoked toexplain nondepressives' different judgmentsof control in the win and lose problems.

i-10

-15

-20

-25

• ND - Press

O D - Press

A ND - Not Press

A D - Not Press

26-75 75-25

Response

Figure 10. Discrepancy between judged and actualpercentage of reinforcement for pressing and notpressing and for depressed (D) and nondepressed(ND) students as a function of response in Experi-

ment 4. (Response refers to whether pressing [75-25] or not pressing [25-75] was associated withthe higher percentage of reinforcement.)


3

2

1

0

-1

-2

4

3

2

1

0

-1

-2

-3

-4

3

2

1

0

-1

Anxiety

Depression

Hostility

Lose WinProblem

Figure. 11. Anxiety, depression, and hostility changescores for depressed (D) and nondepressed (ND)students as a function of problem type in Experi-ment 4. (Positive values indicate affect changes inthe dysphoric direction, and negative values indi-cate affect changes in the euphoric direction.)

These three heuristics are intercorrelated inthat they all rely on the common element ofthe frequency of green light onset. Sincethe frequency of green light onset did notdiffer in the win and lose problems, use ofany of these heuristics as the basis for judg-ments of control would lead to similar ratingsof control in the two problems. Interestingly,nondepressed subjects' judgments of controldid correlate with percentage of successes(r = .47, p < .01) and confirming cases (r= .45, p < .01). Depressed subjects' judg-

ments of control did not correlate signifi-cantly with these heuristics. The finding thatjudged control ratings correlated with per-centage of successes and confirming cases fornondepressed but not depressed subjects isconsistent with and may explain the MoodX Response interaction for judgments ofcontrol.

A Problem X Mood X Sex X Responseanalysis of variance of the scores measuringthe discrepancy between judgment of totalreinforcement and actual total reinforcementyielded a significant main effect for problem,F(l, 47) = 4.44, p < .05. In general, sub-jects accurately estimated the frequency ofgreen light onset in the lose problem (dis-crepancy score = — .09) but overestimatedgreen light frequency in the win problem(discrepancy score = +6.12).

Subjects experienced mood changes as afunction of outcome valence. A Problem XMood X Sex X Response analysis of vari-ance was performed on each of the threeaffect change scores (anxiety, depression,and hostility). For the anxiety change score,significant main effects for problem, F(l,48) = 9.85, p < .003, and mood, F(l, 48)= 7.67, p < .008, were obtained, with noother significant main effects or interactions.The analysis of variance for change in de-pression also yielded significant main effectsfor problem, F(l, 48) = 7.87, p < .007, andmood, F(l, 48) = 23.26, p < .001.

The analysis of variance for change inhostility revealed no significant effects. Fig-ure 11 shows that subjects tended to becomemore dysphoric in the lose problem than inthe win problem. As in Experiment 3, themain effect for mood probably represents aconstraint placed on subjects' affect changescores by their initial MAACL scores.

Post experimental questionnaire. Sub-jects' responses on the postexperimentalquestionnaire provided converging evidencefor nondepressives' errors in judgments ofcontrol. A Problem X Mood X Sex X Re-sponse analysis of variance of the questionassessing the degree to which subjects be-lieved that factors other than their ownresponding produced green light onsetrevealed a significant Problem X Mood


interaction, F(l, 48) = 4.68, p < .04, and asignificant Mood X Response interaction,F(l, 48) = 4.68, p < .04. The simple maineffects test analyzing the Problem X Moodinteraction showed that nondepressed sub-jects believed that factors other than theirown responding were significantly more im-portant in the lose problem than in the winproblem, F(l, 48) = 7.91, p < .01, whereasthere was no such difference for depressedsubjects. In addition, nondepressed subjectsbelieved that factors other than their ownresponding were significantly more impor-tant in the lose problem than did thedepressed subjects, F(l, 48) = 4.84, p <.OS.There was no difference between depressedand nondepressd subjects in the win problem.

The simple main effects test analyzing theMood X Response interaction showed thatnondepressed subjects believed that factorsother than their own responding were signifi-cantly more important when not pressing wasassociated with 75% reinforcement (25-75)than when pressing was associated with75% reinforcement (75-25), F(l, 48) =7.91, p < .01. Depressed subjects did notrate factors other than their own respondingas differentially important in the two condi-tions. In addition, nondepressed subjectsbelieved that factors other than their ownresponding were significantly more impor-tant than did depressives when not pressingwas associated with 75% reinforcement (25-75), F(l, 48) =4.84, p < .05. Depressedand nondepressed subjects did not differwhen pressing was associated with 75%reinforcement (75-25). Thus, subjects'answers on the "factors" question were con-sistent with their answers on the judgmentof Control scale.

It is interesting that in the open-endedquestions assessing subjects' logic in arrivingat their judgments of control, nondepressivesaccurately reported that one response wasmore effective than the other in producingthe green light in the win problem, but ofteninaccurately reported that the two responseswere equally ineffective in the lose problem.Depressed subjects, on the other hand, accu-rately reported that pressing and not press-ing were differentially effective regardless

of whether they were winning or losingmoney.

Discussion

Contrary to the prediction of the learnedhelplessness hypothesis, depressed studentsdid not underestimate the degree of contin-gency between their responses and outcomesin a hedonically charged situation. Outcomevalence was not without an effect in Experi-ment 4, however. The nondepressed studentsunderestimated the degree of control theirresponses exerted over outcomes. Nonde-pressives erroneously believed that theirresponses had very little control over thegreen light when they lost money, but notwhen they won money, even though theactual degree of control was 50% in eachcase. Thus, as in Experiment 3, outcomevalence influenced the judgments of contin-gency of nondepressed, but not depressed,students.

In addition, nondepressed students' judg-ments of control also were affected bywhether the active (pressing) or the passive(not pressing) response was associated withthe higher frequency of reinforcement. Non-depressives underestimated the degree ofcontingency when the passive response led tothe higher frequency of reinforcement. De-pressed students, on the other hand, accu-rately judged the degree of contingencybetween their responses and outcomes re-gardless of whether the active or passiveresponse was more effective.

Nondepressives underestimated the degreeof control when not pressing was the moreeffective of the two responses. This findingis consistent with and might be deduced fromthe additional fact that nondepressives'judgments of control correlated with theinvalid heuristics of percentage of successesand confirming cases. Recall that percentageof successes is defined as the percentage oftrials on which the green light appears whenthe subject presses the button, and confirm-ing cases is defined as the sum of the numberof trials on which the subject presses thebutton and the green light comes on plusthe number of trials on which the subjectdoes not press the button and the green light


does not come on. Use of either of theseheuristics would lead to a lower rating ofcontrol in the 25-75 condition (passive) thanin the 75-25 condition (active). The effectof outcome valence (lose vs. win) cannot beexplained by use of these heuristics, sincefor half of the students in both the lose andwin problems, the active response was moreeffective, whereas for the other half, thepassive response was more effective.

General Discussion

The results of Experiments 1 through 4suggest that contrary to previous assertionsin the psychological literature (Jenkins &Ward, 1965; Smedslund, 1963), people dohave a concept of contingency, that in theresponse-outcome case entails the relativeefficacy of one response versus another indetermining outcomes. However, in manyinstances, people do not use this concept indetermining the degree of contingency be-tween their responses and outcomes. Thus,people's subjective representations of contin-gencies often fail to mirror objective contin-gencies.

The most convincing demonstration ofappropriate use of relative efficacy in deter-mining degree of contingency is provided bythe judged control data of Experiment 1.College students were able to quantify accu-rately the degree of relation between theirresponses and an environmental outcomeacross a wide range of the contingency space.The vast majority of past studies concludingthat people do not rely on relative efficacyin judging contingency have centered onpeople's behavior and beliefs only in sit-uations in which responses and outcomes arenoncontingently related (Bruner & Revusky,1961; Langer, 1975; Wortman, 1975;Wright, 1962). Although the noncontingentcase is certainly interesting (see below), itis only one type of contingency and thereforeis not appropriate as the sole base fromwhich to draw broad inferences aboutpeople's general concept of contingency.Also, students' verbalized logic in Experi-ment 1 suggests that they relied on differ-ential effectiveness of responses in judgingthe degree of contingency. It is not surpris-

ing that students also ably performed theresponse that maximized green light onsetwhen green light onset was associated withmonetary reward in the behavioral task ofExperiment 1.

Although people's subjective represen-tations of contingencies mirrored objectivecontingencies in Experiment 1, such isomor-phism broke down in Experiments 2, 3, and4. These errors in judgments of contingencywere not random. Instead, they were sys-tematically related to both emotional stateand characteristics of the outcome. In brief,depressed students accurately judged thedegree of contingency regardless of its actualmagnitude. Nondepressed students over-estimated degree of contingency when non-contingent outcomes were frequent and/ordesired and underestimated degree of contin-gency when contingent outcomes were un-desired. Moreover, nondepressed studentswere relatively accurate in judging theconditional probabilities of outcomes givenresponses. This suggests that the locus oftheir judgment errors is in the organisationof the incoming response-outcome data andnot in the perception of the data themselves.

An interesting asymmetry exists in non-depressed students' susceptibility to errorsin judging contingency. Nondepressiveswere more prone to erroneous judgmentswhen responses and outcomes were noncon-tingently related than when responsesand outcomes were contingently related.Although manipulations of reinforcementfrequency affected judgments of controladversely in the noncontingent case, suchmanipulations were not sufficient to produceerrors in the contingent case. These resultsmay imply that noncontingency is a psy-chologically more difficult relationship toperceive or understand than contingency.

Two pieces of evidence from Experiment1 suggest that students do find lack of con-tingency more difficult to perceive thanpresence of contingency. First, students wereleast able to maximize reinforcements in thebehavioral task for the problem with theleast degree of contingency. Second, studentsalso felt least certain about the accuracyof their judgments of control in the problem


with the least degree of contingency. More-over, Piagetand Inhelder (1975) have arguedthat understanding independent relationscomes at a developmentally later stage thanunderstanding dependent relations. Theseauthors contend that the notion of chance isabsent in the young child and necessarilydepends for its development on intuitions ofcausality or nonchance. Of course, this dis-cussion need not imply that people will nevererr in judging degree of control in situationsin which a contingency does exist (seeExperiment 4), but rather that people aremore susceptible to errors in judgingnoncontingency.

Parallels in Cognitive Psychology,Social Psychology, and Animal Learning

Similar to the present finding that subjec-tive and objective contingencies are notalways isomorphic, Kahneman and Tversky(Kahneman & Tversky, 1972, 1973; Tver-sky & Kahneman, 1971, 1974) have foundthat subjective and objective probabilitiesalso do not always coincide. These cognitivepsychologists have found that people oftenfail to use the mathematical rules of chancein making predictions. Instead, they rely ona limited number of heuristics such as repre-sentativeness and availability. People's useof representativeness leads them to makepredictions based on assessments of similar-ity between evidence and a predicted out-come. The heuristic of availability leads themto predict on the basis of the ease with whichrelevant instances of the event to be pre-dicted come to mind. Although use of theseheuristics sometimes yields reasonable judg-ments, it often leads to severe and systematicerrors. Thus, use of invalid heuristics is notlimited to detection of contingencies betweenresponses and outcomes but extends to otherinstances of subjective judgment as well.

Review of work in social psychology sug-gests that in addition to frequency andvalence of outcome, several other psycho-logical dimensions are relevant to people'scognitive representations of contingencies.For example, Langer (1975) has shownthat when elements characteristic of skillsituations (e.g., practice) are introduced into

situations in which outcomes and responsesare noncontingently related, people oftenbehave as if they have control. The factorsof foreknowledge of the goal and personalinvolvement in the task also have beenshown to induce illusions of control (Wort-man, 1975). In addition, if successes occurearly in a task as opposed to late in a taskor randomly interspersed throughout thetask, people often act as if they have morecontrol (Langer & Roth, 1975). Finally,Chapman and Chapman (1967) have shownthat the higher the associative connectionbetween two events, the more likely it is thatthese events will be seen as correlated whenin fact they are uncorrelated. In common withthe present investigation, these studies foundthat when faced with objective lack of con-tingency, people may use a number ofinvalid heuristics that lead to erroneousjudgments of control. An extrapolation fromthe results of the present study is thatdepressed individuals would be less likelyto succumb to these various types of illusionof control. Indeed, Golin, Terrell, and John-son (1977) found that depressed subjectsdid not show an illusion of control (Langer,1975) in a chance task in which elementscharacteristic of a skill task had beenintroduced.

The majority of these social psychologystudies also employed a control group thatwas not subjected to the various manipula-tions shown to induce the illusion of control.Subjects in these control groups were accu-rate in assessing noncontingency. Therefore,these studies corroborate the results ofExperiment 1 that people do have a conceptof control that entails differential effective-ness of responses, but in the context of cer-tain powerful psychological factors (e.g.,hedonically relevant outcomes, personalinvolvement, and practice) people do not usedifferential effectiveness of responses as theindex of contingency and instead resort tomore primitive heuristics or strategies.(Abramson & Alloy, in press, recently havesuggested explanations about the kinds ofpsychological factors that induce distortionsin subjective judgments of contingency.)

A number of parallels also appear to exist


between the present results and work onanimal contingency learning. For example,Rescorla (1968) has reported that thegreater the degree of contingency betweenCS and UCS in a Pavlovian fear condition-ing paradigm, the greater the degree of fearshown by rats in a conditioned emotionalresponse (CER) test. These results indicatethat animals are behaviorally sensitive todegrees of contingency. The counterpart ofthis rinding in the present investigation isrepresented by the ability of students inExperiment 1 to quantify accurately thedegree of contingency between theirresponses and outcomes.

Similarly, those variables that determineillusions of control in humans appear to havecounterparts producing analogous effects inanimals. An animal analogue of the problemsin which outcomes were noncontingent in thepresent investigation is the truly randomcontrol (TRC) procedure where CSs andUCS are presented randomly with respect toeach other (Rescorla, 1967). To assess thedegree of conditioning accruing to the CS, aCER test typically is employed (Annau &Kamin, 1961; Estes & Skinner, 1941). Thegeneral finding is that at asymptote, animalsexposed to the TRC procedure show noCER suppression, indicating the absence ofboth excitatory and inhibitory conditioning(Rescorla, 1967, 1972). Preasymptotically,however, animals exposed to the TRC pro-cedure often do show excitatory conditioning(Ayres, Benedict, & Witcher, 1975; Bene-dict & Ayres, 1972; Keller, Ayres, & Ma-honey, 1977; Kremer, 1974; Kremer &Kamin, 1971; Quinsey, 1971; Rescorla,1968, 1972).

It is interesting that a number of the vari-ables that determine the magnitude andduration of this initial excitatory condition-ing in the TRC procedure have parallels inthe variables shown to produce overesti-mations of contingency by nondepressed stu-dents in the present study. For example,Rescorla (1972) has shown that the greaterthe overall probability of the UCS in aprocedure where CSs and UCSs are un-correlated, the greater the magnitude ofpreasymptotic conditioning. This finding is

analogous to the results of Experiment 2,in which nondepressed students over-estimated the degree of contingency betweenresponses and outcomes when the outcomeof interest occurred with high probability,but not when it occurred with low prob-ability. Second, Quinsey (1971) demon-strated that the greater the magnitude orintensity of the UCS in a TRC procedure,the greater the initial excitatory conditioningthat is obtained. One can view UCS intensityas one of the many variables that wouldcontribute to the valence of the outcome.In Experiment 3, of course, nondepressedstudents showed illusions of control whengood outcomes occurred but not when badoutcomes occurred.

Another variable that has been demon-strated to affect preasymptotic conditioningin the TRC procedure is the number of ini-tial CS-UCS pairings: the greater the num-ber of initial pairings, the greater the excita-tory conditioning (Benedict & Ayres, 1972).Analogously, Langer and Roth (1975)found that people are most likely to succumbto an illusion of control in an objectivelychance situation when they receive a largenumber of initial successes. Finally, theeffect of valence on subjective judgmentsof contingency has a parallel in the work on"superstitious" conditioning (Skinner, 1948)in operant situations in which reinforcementis noncontingent. The vast majority ofdemonstrations of superstitious behavior inanimals have been in appetitive rather thanaversive paradigms (for a review of thesuperstition literature, see Herrnstein, 1966,and Staddon & Simelhag, 1971).

These striking parallels between animalsand humans in contingency learning situa-tions suggest that there may be certainfundamental processes underlying contin-gency learning across species. If there aresuch basic processes common to animals andhumans, then other variables shown to affectthe magnitude and duration of preasymptoticconditioning in the TRC procedure inanimals may be predicted similarly to affectillusions of control in humans. For example,the finding that the longer an animal isexposed to the TRC procedure, the less


excitatory conditioning it shows (Keller etal., 1977; Rescorla, 1972) suggests that ifnondepressed students in the present studywere given greater numbers of trials to learnabout noncontingency, the magnitude of theiroverestimations of degree of control mightalso diminish. Similarly, the other variablesshown to affect preasymptotic conditioningin animals such as the salience of the CS(Kremer & Kamin, 1971), proportion of thesession occupied by the CS (Kremer &Kamin, 1971), intertrial interval length(Kremer & Kamin, 1971; Quinsey, 1971),and number of unpaired UCSs (Keller etal., 1977; Rescorla, 1968) should also havecounterparts in instrumental learning situ-ations that affect the magnitude of theillusion of control.

It is important to remember that depressedstudents did not show illusions of controlwhen the variables of outcome frequencyor valence were manipulated in the presentinvestigation. Accordingly, we would predictthat depressed people also would be lessaffected by the other aforementioned vari-ables and further speculate that "depressed"animals—animals that have been exposed touncontrollable outcomes (Seligman, 1975b)—should also be less influenced by thesevariables in the TRC procedure. Alloy andEhrman (Note 2) have found support forthis prediction. Rats that receive uncon-trollable shocks show less excitatory condi-tioning in a TRC procedure in which a toneand shock are noncontingently related thanrats that receive controllable or no shocks.

Theoretical Alternatives

We have demonstrated that both de-pressed and nondepressed people have aconcept of contingency that entails relativeefficacy of responses, but nondepressives failto apply this concept under many circum-stances. The following section of the articleevaluates the learned helplessness theory asan account of the judgment of contingencyfindings and presents several alternativeexplanations.

The following two predictions werederived from the helplessness model ofdepression. First, depressed individuals will

underestimate the degree of control theirresponses exert over outcomes relative toeither the objective degree of control or tonondepressed individuals. Second, non-depressed people will overestimate thedegree of response-outcome relation relativeto either the actual degree of relation or todepressed people. Clearly, the overall patternof results in Experiments 1 through 4 doesnot support these predictions. Althoughdepressed students were surprisingly accu-rate in judging degree of contingency, non-depressed students showed both illusions ofcontrol and illusions of no control dependingon the particular experimental situation.

The present investigation's failure to con-firm predictions made by the learned help-lessness theory of depression is particularlyimportant because this study provides themost direct assessment of people's percep-tions of response-outcome contingencies todate. A number of theorists (Buchwald,Coyne, & Cole, 1978; Costello, 1978; Hues-mann, 1978), have argued cogently thatstudies on learned helplessness and depres-sion have failed to provide convincingevidence for the specific "associative deficit"postulated by helplessness theory: Depres-sives fail to perceive the relation betweentheir responses and outcomes. In the main,previous studies on learned helplessness anddepression have demonstrated that similarto nondepressed students made helpless,depressed students show impaired perform-ance on a number of laboratory tasks relativeto nondepressed students. Insofar as thesestudies show that depressed studentsperform similarly to helpless students, theyprovide some support for the learned help-lessness theory of depression. The majorproblem with these studies, however, isthat they fail to isolate the cognitive com-ponent from the motivational component ofhelplessness (Costello, 1978). The resultsof the present investigation provide no evi-dence for the hypothesized associative deficitin depressives; if anything, they suggestthat nondepressives have difficulties inassessing response-outcome relationships.(See Alloy & Seligman, in press, for acomprehensive discussion of the associative


deficit in learned helplessness and depres-sion.)

How can the present study's failure toobtain evidence for the associative deficit indepression be reconciled with the often repli-cated finding (e.g., Klein & Seligman, 1976;Miller & Seligman, 1975) that depressivesdo poorly on instrumental learning tasks?The learned helplessness hypothesis, ofcourse, postulates that depressives' poorperformance on instrumental tasks is, inpart, mediated by the associative deficit. Oneway to explain these two seemingly disparatefindings is to propose a revised helplessnesstheory of depression. The revised hypothesismaintains that depressives are characterizedby a generalized expectation of no control,but postulates that this expectation onlyinterferes with initiation of responses (the"motivational deficit" of helplessness) andnot with the perception of response-outcomerelationships (the "associative deficit" ofhelplessness).

In other words, the revised helplessnesshypothesis suggests that depressed individ-uals often perform poorly on instrumentaltasks because they fail to generate theresponse that increases the probability of thesuccessful outcome, not because they havetrouble discerning the effect their responsesexert on these outcomes. The observation inExperiment 2 that depressed students wereless likely to generate complex hypothesesthan were nondepressed students providessome support for this revised helplessnessmodel. Additional support for this hypoth-esis comes from projective testing in theclinic. Depressives generate fewer and lesscomplex responses than nondepressives onthe Rorschach test (Beck, 1945). Further-more, experiments on animal helplessnesshave shown that rats exposed to uncon-trollable shocks only show deficits on testsfor helplessness if the response necessary tocontrol reinforcement in the test task is adifficult one (Maier et al., 1973; Seligman &Beagley, 1975).

This revised hypothesis makes a testableprediction: Depressives should underesti-mate the degree of control a response exertsover an outcome in situations in which the

controlling response is complex (becausedepressives will not generate the appropriateresponse), but not in equally complex situ-ations in which the controlling responses aregenerated for them. Abramson, Alloy, andRosoff (Note 3) recently have confirmedthis prediction. A major disadvantage ofsuch a revised hypothesis is that it fails toexplain the robust inaccuracies of nondepres-sives in judging contingencies.7

An alternative, though not necessarily in-compatible, framework for examining thejudgment of contingency results 'is a moti-vational account derived from social psy-chology. The finding that nondepressed stu-dents perceive a contingency between theirresponses and desirable outcomes but notbetween their responses and undesirableoutcomes is reminiscent of a general phenom-enon in social psychology (Cohen, 1964;Feather, 1969; Fitch, 1970; Johnson, Fei-genbaum, & Weiby, 1964; Langer & Roth,1975; Medow & Zander, 1965; Streufert &Streufert, 1969; Weiner et al., 1971; Wort-man, Costanzo, & Witt, 1973). These in-vestigators have examined the causal attribu-tions of individuals who received false feed-back that they had either succeeded or failedat a task. Subjects in these experiments gen-erally have attributed causality to them-selves when they succeeded and to factorsin the environment or situation when theyfailed.

In explaining this general finding, sometheorists (e.g., Fitch, 1970; Wortman et al.,1973) have relied on the notion that peopleare motivated to maintain or enhance self-esteem. The logic here is that taking creditfor good outcomes maintains or enhancesself-esteem and viewing bad outcomes ascaused by factors outside the self is notdamaging to self-esteem (see Bradley, 1978;Miller & Ross, 1975; and Miller, 1978, fora review of evidence bearing on self-servingattributions). Similar reasoning may beapplicable to the judgment of contingency

T A further indication of the robustness of thenondepressive "illusion of control" is the perform-ance of R.L.S., a learning ^ theorist who was apilot subject and showed a larger illusion than anyother subject in the 75-75 problem.


results for nondepressed students in Experi-ments 2, 3, and 4. That is, nondepressed stu-dents may have shown illusions of controlwhen good outcomes occurred in order topreserve or increase self-esteem but mayhave shown illusions of no control whenbad outcomes occurred in order to preventloss of self-esteem. It should be noted thatlack of control can result either from internalfactors such as personal incompetence orexternal factors such as the harshness of theenvironment (see Abramson et al., 1978, fora comprehensive discussion of the relationbetween perceived control and attributions).For such a motivational account to gothrough, nondepressives who believe theyhave no control over bad outcomes mustattribute this lack of control to externalrather than to internal factors.

If one assumes that depressed individualsare not motivated to maintain or enhanceself-esteem, then their accuracy in detectingcontingencies regardless of outcome valencefollows. Indeed, the depressive often is char-acterized by low self-esteem (Beck, 1967,1976; Bibring, 1953; Freud, 1917/1957),and moreover, Bibring (1953) has arguedthat depressives are not motivated to regainself-esteem precisely because the mechanismfor self-deception in these individuals hasbroken down. Freud (1917/1957) appearsto espouse a similar view:

When in his [the depressive's] heightened self-criticism he describes himself as petty, egoistic,dishonest, lacking in independence, one whose soleaim has been to hide the weakness of his ownnature, it may be, so far as we know, that he hascome pretty near to understanding himself; weonly wonder why a man has to be ill before hecan be accessible to a truth of this kind. (p. 246)

Thus, the present findings are compatiblewith a view that postulates differential moti-vation for maintaining self-esteem in de-pressed and nondepressed individuals. Adeduction from this motivational hypothesisis that the attributional style observed in thesocial psychological experiments would becharacteristic of nondepressed, but notdepressed, people. In line with this predic-tion, a number of investigators have foundthat nondepressives show the self-enhancingattributional style, but depressives do not

(Klein, Fencil-Morse, & Seligman, 1976;Kuiper, 1978; Rizley, 1978; Seligman,Abramson, Semmel, & von Baeyer, 1979).Further work is needed to directly test theadequacy of a motivational account of differ-ences between depressed and nondepressedpeople in judging contingencies.

General Implications

A consistent and intriguing theme in ourresults is the tendency of nondepressed stu-dents to succumb to various "cognitiveillusions." While cognitive accounts ofdepression emphasize depressives' errors ininterpreting their environment and behavior(e.g., Beck, 1967, 1976), our findingsdemonstrate the existence of cognitive errorsmade by nondepressed people.

Interestingly, a number of recent studiesdocument nondepressive illusions in areasother than the judgment of contingency. Forexample, in a study of the selective recallof positive and negative feedback, Nelsonand Craighead (1977) found that depressedcollege students accurately recalled the fre-quency of negative feedback on a laboratorytask, whereas nondepressed students under-estimated the frequency of negative feedback(see also DeMonbreun & Craighead, 1977).Similarly, Rozensky, Rehm, Pry, and Roth(1977) reported that nondepressed controlpatients rewarded themselves to a greaterdegree than their objective performanceswould warrant. Although depressed patientsalso tended to over-reward themselves, theywere more accurate in self-reward than thenondepressives. Finally, Lewinsohn, Mischel,Chaplin, and Barton (in press) found thatdepressed patients accurately assessed theirsocial competence whereas nondepressedpsychiatric and normal control subjects per-ceived themselves more positively than otherpeople saw them. Interestingly, even afterthey had received therapy, the depressedpatients were more realistic in assessingtheir competencies than were the controlsubjects.

Taken together, these studies suggest thatat times depressed people are "sadder butwiser" than nondepressed people. Non-depressed people succumb to cognitive illu-


sions that enable them to see both themselvesand their environment with a rosy glow. Acrucial question is whether depression itselfleads people to be "realistic" or whether real-istic people are more vulnerable to depres-sion than other people.

An obvious question emerges concerningthe generality of the present results toclinical populations of depressives. Furtherstudies are necessary to determine whichsubtypes of depressed inpatients (unipolar-bipolar, endogenous-reactive), if any, showsimilar accuracy in judging relations be-tween their responses and outcomes relativeto individuals with no psychopathology orother forms of psychopathology. A secondclinical question concerns the relation be-tween the illusion of control and mania.Perhaps individuals in manic states wouldshow greater overestimations of noncontin-gent response-outcome relations than non-depressed individuals (see Langer, 1975, fora similar discussion).

It is worth pointing out that the illusionof control shown by nondepressed individualswhen good outcomes occur noncontingentlyhas important implications for the experi-mental paradigm of "appetitive helplessness"(Maier & Seligman, 1976). Appetitive help-lessness refers to a procedure in which anorganism is exposed to uncontrollable posi-tive events and then is tested for deficits ininstrumental responding. This paradigm isof particular theoretical interest for thelearned helplessness model because the bulkof experimental evidence testing the theorycomes from uncontrollable aversive para-digms whereas the theory states that beliefin uncontrollability in either appetitive oraversive situations will lead to helplessness.The results of the present investigation sug-gest that empirically, appetitive helplessnessmay be difficult to obtain, since nondepressedsubjects in these paradigms may believeerroneously that they have control when infact they do not. Therefore, to insure a fairtest of helplessness theory, investigators ofappetitive helplessness need to determinewhether their subjects are, in fact, conclud-ing that they have no control over objectivelyuncontrollable positive outcomes.

In the experiments in this article, studentsjudged the degree of contingency betweenresponses and outcomes. Previous studies ofbehavior in stimulus prediction situationshave demonstrated that people often believethat noncontingently related stimulus eventsalso are correlated (e.g., Chapman & Chap-man, 1967; Hake & Hyman, 1953; Starr &Katkin, 1969). Future research is necessaryto determine whether depressed individuals'relative accuracy in judging the degree ofcontrol that their responses exert over out-comes is accompanied by similar accuracy injudging degree of contingency between twostimulus events in a prediction situation.

The starting point of the present investi-gation was the concept of contingencydeveloped within the learning tradition; anappropriate ending point is the implicationsof this investigation for learning theory.Many contemporary learning theorists haveproposed views of learning that require anorganism capable of cognitively representingrelations between stimuli and/or responsesand outcomes (Bandura, 1977; Bolles, 1972;Estes, 1972; Mackintosh, 1973; Maier &Seligman, 1976). In these views, changes inbehavior are mediated by these cognitiverepresentations. The fact that humans' sub-jective representations of contingencies donot always mirror objective contingenciessuggests that such learning theories areincomplete until they can specify the rulesfor relating the two.

Reference Notes

1. Alloy, L. B., & Abramson, L. Y. The judgmentof contingency: An experimental test of thelearned helplessness theory of depression. Manu-script in preparation, University of Pennsyl-vania, 1979.

2. Alloy, L. B., & Ehrman, R. N. Instrumental toPavlovian transfer: Learning about response-reinforcer contingencies affects subsequent learn-ing about stimulus-reinforcer contingencies.Manuscript submitted for publication, 1979.

3. Abramson, L. Y., Alloy, L. B., & Rosoff, R.Depression and the role of complex hypothesesin the judgment of response-outcome contin-gencies. Manuscript submitted for publication,1979.


References

Abramson, L. Y., & Alloy, L. B, Judgment of con-tingency: Errors and their implications. In A.Bautn & J. Singer (Eds.), Advances in environ-mental psychology (Vol. 2). Hillsdale, N.J.:Erlbaum, in press.

Abramson, L. Y., Garber, J., Edwards, N. B., &Seligman, M. E. P. Expectancy changes in de-pression and schizophrenia. Journal of AbnormalPsychology, 1978, 87, 102-109.

Abramson, L. Y., Seligman, M. E. P., & Teasdale,J. Learned helplessness in humans: Critique andreformulation. Journal of Abnormal Psychology,1978, 87, 49-74.

Alloy, L. B., & Seligman, M. E. P. On the cogni-tive component of learned helplessness and de-pression. In G. H. Bower (Ed.), The psychologyof learning and motivation (Vol. 13). NewYork: Academic Press, in press.

Annau, Z., & Kamin, L. J. The conditioned emo-tional response as a function of intensity of theUS. Journal of Comparative and PhysiologicalPsychology, 1961, 54, 428-432.

Arieti, S. Cognition and feeling. In M. B. Arnold(Ed.), Feelings and emotions. New York:Academic Press, 1970.

Ayres, J. J., Benedict, J. O., & Witcher, E. S.Systematic manipulation of individual events in atruly random control in rats. Journal of Compar-ative and Physiological Psychology, 1975, 88, 97-103.

Baker, A. G. Learned irrelevance and learned help-lessness : Rats learn that stimuli, reinforcers, andresponses are uncorrelated. Journal of Experi-mental Psychology: Animal Behavior Processes,1976, 2, 130-141.

Bandura, A. Self-efficacy: Toward a unifyingtheory of behavioral change. Psychological Re-view, 1977, 84, 191-215.

Beck, A. T. Depression: Clinical, experimental,and theoretical aspects. New York: Harper &Row, 1967.

Beck, A. T. Cognitive therapy and emotional dis-orders. New York: International UniversitiesPress, 1976.

Beck, S. J. Rorschach's test II: A variety of per-sonality pictures. New Yorb: Grune & Stratton,1945.

Bern, D. J. Self-perception theory. In L. Berkowitz(Ed.), Advances in experimental social psychol-ogy (Vol. 6). New York: Academic Press, 1972.

Benedict, J. O., & Ayres, J. J. Factors affectingconditioning in the random control procedure inthe rat. Journal of Comparative and PhysiologicalPsychology, 1972, 78, 323-330.

Bibring, E. The mechanism of depression. In P.Greenacre (Ed.), Affective disorders. NewYork: International Universities Press, 1953.

Bindra, D. A unified account of classical condition-ing and operant training. In A. H. Black & W.F. Prokasy (Eds.), Classical conditioning II:

Current theory and research. New York: Apple-ton-Century-Crofts, 1972.

Bloomfield, T. M. Reinforcement schedules: Con-tingency or contiguity? In R. M. Gilbert & J. R.Millenson (Eds.), Reinforcement: Behaviouranalysis. New York: Academic Press, 1972.

Bolles, D. C. Reinforcement, expectancy, and learn-ing. Psychological Review, 1972, 79, 394-409.

Bradley, G. W. Self-serving biases in the attribu-tion process: A reexamination of the fact orfiction question. Journal of Personality and SocialPsychology, 1978, 36, 56-71.

Braud, W., Wepmann, B., & Russo, D. Task andspecies generality of the "helplessness" phenom-enon. Psychonomic Science, 1969,16, 154-155.

Bruner, A., & Revusky, S. H. Collateral behaviorin humans. Journal of the Experimental Analysisof Behavior, 1961, 4, 349-350.

Buchwald, A. M., Coyne, J, C., & Cole, C. S. Acritical evaluation of the learned helplessnessmodel of depression. Journal of Abnormal Psy-chology, 1978, 87, 180-193.

Bull, J. A., Ill, & Overmier, J. B. Additive andsubtractive properties of excitation and inhibi-tion. Journal of Comparative and PhysiologicalPsychology, 1968, 66, 511-514.

Bumberry, W., Oliver, J. M., & McClure, J. N.Validation of the Beck Depression Inventory ina university population using psychiatric estimateas the criterion. Journal of Consulting andClinical Psychology, 1978, 46, 150-155.

Catania, A. C. Elicitation, reinforcement, and stim-ulus control. In R. Glaser (Ed.), The nature ofreinforcement. New York: Academic Press, 1971.

Chapman, L. J., & Champan, J. P. Genesis ofpopular but erroneous psychodiagnostic cate-gories. Journal of Abnormal Psychology, 1967,72, 193-204.

Church, R. M. Response suppression. In B. A.Campbell & R. M. Church (Eds.), Punishmentand aversive behavior. New York: Appleton-Century-Crofts, 1969.

Cohen, J. Behavior in uncertainty. London: Unwin,1964.

Costello, C. G. Depression: Loss of reinforcers orloss of reinforcer effectiveness? BehaviorTherapy, 1972, 3, 240-247.

Costello, C. G. A critical review of Seligman'slaboratory experiments on learned helplessnessand depression in humans. Journal of AbnormalPsychology, 1978, 87, 21-31.

DeMonbreun, B. G., & Craighead, W. E. Distortionof perception and recall of positive and neutralfeedback in depression. Cognitive Therapy andResearch, 1977, 1, 311-329.

Dweck, C. S., & Wagner, A. R. Situational cuesand correlation between conditioned stimulus andunconditioned stimulus as determinants of theconditioned emotional response. PsychonomicScience, 1970, 18, 145-147.

Estes, W. K. Reinforcement in human behavior.American Scientist, 1972, 60, 723-729.

Estes, W. K., & Skinner, B. F. Some quantitative


properties of anxiety. Journal of ExperimentalPsychology, 1941, 29, 390-400.

Feather, N. T. Attribution of responsibility andvalence of success and failure in relation toinitial confidence and task performance. Journalof Personality and Social Psychology, 1969, 13,129-144.

Fitch, G. Effects of self-esteem, perceived per-formance, and choice on causal attributions.Journal of Personality and Social Psychology,1970,16, 311-315.

Fosco, F., & Geer, J. H. Effects of gaining controlover aversive stimuli after differing amounts ofno control. Psychological Reports, 1971, 29, 1153-1154.

Freud, S. Mourning and melancholia. In J.Strachey (Ed. & trans.), The complete psycho-logical works of Sigmund Freud (Vol. 14).London: Hogarth Press, 1957. (Originally pub-lished, 1917.)

Gatchel, R. J., Paulus, P. B., & Maples, C. W.Learned helplessness and self-reported affect.Journal of Abnormal Psychology, 1975, 84, 732-734.

Gibbon, J., Berryman, R., & Thompson, R. L.Contingency spaces and measures in classical andinstrumental conditioning. Journal of the Experi-mental Analysis of Behavior, 1974, 21, 585-605.

Glass, D. C., & Singer, J. E. Urban stress: Experi-ments on noise and social stressors. New York:Academic Press, 1972.

Golin, S., Terrell, F., & Johnson, B. Depression andthe illusion of control. Journal of Abnormal Psy-chology, 1977, 86, 440-442.

Goodkin, F. Rats learn the relationship betweenresponding and environmental events: An expan-sion of the learned helplessness hypothesis.Learning and Motivation, 1976, 7, 382-393.

Hake, H. W., & Hyman, R. Perception of thestatistical structure of a random series of binarysymbols. Journal of Experimental Psychology,1953, 43, 67-74.

Hammond, L. J. Increased responding to CS— indifferential CER. Psychonornic Science, 1966, 5,337-338.

Hammond, L. J. A traditional demonstration of theproperties of Pavlovian inhibition using differ-ential CER. Psychonomic Science, 1967, 9, 65-66.

Hammond, L. J. Retardation of fear acquisitionwhen the CS has previously been inhibitory.Journal of Comparative and Physiological Psy-chology, 1968, 66, 756-759.

Hammond, L. J., & Daniel, R. Negative contin-gency discrimination: Differentiation by rats be-tween safe and random stimuli. Journal of Com-parative and Physiological Psychology, 1970,72, 486-491.

Herrnstein, R. J. Superstition: A corollary prin-ciple of operant conditioning. In W. K. Honig(Ed.), Operant behavior: Areas of research andapplication. New York: Appleton-Century-Crofts,1966.

Hiroto, D. S. Locus of control and learned help-

lessness. Journal of Experimental Psychology,1974, 102, 187-193.

Hiroto, D. S., & Seligman, M. E. P. Generality oflearned helplessness in man. Journal of Person-ality and Social Psychology, 1975, 31, 311-327.

Huesmann, L. R. Cognitive processes and modelsof depression. Journal of Abnormal Psychology,1978, 87, 194-198.

James, W. H. Internal versus external control ofreinforcement as a basic variable in learningtheory. Unpublished doctoral dissertation, OhioState University, 1957.

James, W. H., & Rotter, J. B. Partial and onehundred percent reinforcement under chance andskill conditions. Journal of Experimental Psy-chology, 1958, 55, 397-403.

Jenkins, H. M., & Ward, W. C. Judgment of con-tingency between response and outcome. Psycho-logical Monographs, 1965, 79(1, Whole No. 594).

Johnson, T. J., Feigenbaum, R., & Weiby, M.Some determinants and consequences of theteacher's perception of causation. Journal of Edu-cational Psychology, 1964, 55, 237-246.

Kahneman, D., & Tversky, A. Subjective probabil-ity : A judgment of representativeness. CognitivePsychology, 1972, 3, 430-454.

Kahneman, D., & Tversky, A. On the psychologyof prediction. Psychological Review, 1973, 80,237-251.

Keller, R. J., Ayres, J. J. B., & Mahoney, W. J.Brief versus extended exposure to truly randomcontrol procedures. Journal of ExperimentalPsychology: Animal Behavior Processes, 1977,3, 53-65.

Kimble, G. A., & Ost, J. W. P. A conditionedinhibitory process in eyelid conditioning. Journalof Experimental Psychology, 1961, 61, 150-156.

Klein, D. C., Fencil-Morse, E., & Seligman, M. E.P. Depression, learned helplessness, and theattribution of failure. Journal of Personality andSocial Psychology, 1976, 33, 508-516.

Klein, D. C., & Seligman, M. E. P. Reversal ofperformance deficits and perceptual deficits inlearned helplessness and depression. Journal ofAbnormal Psychology, 1976,55, 11-26.

Kremer, E. F. The truly random control proced-ure: Conditioning to the static cues. Journal ofComparative and Physiological Psychology, 1974,86, 700-707.

Kremer, E. J., & Kamin, L. J. The truly randomcontrol procedure: Associative or nonassociativeeffects in rats. Journal of Comparative andPhysiological Psychology, 1971, 74, 203-210.

Kuiper, N. A. Depression and causal attributionsfor success and failure. Journal of Personalityand Social Psychology, 1978, 36, 236-246.

Langer, E. J. The illusion of control. Journal ofPersonality and Social Psychology, 1975, 32,311-328.

Langer, E. J., & Roth, J. Heads I win, tails it'schance: The illusion of control as a function ofthe sequence of outcomes in a purely chance task.


Journal of Personality and Social Psychology,1975, 32, 951-9SS.

Lewinsohn, P. M., Mischel, W., Chaplain, W., &Barton, R. Social competence and depression:The role of illusory self-perceptions ? Journal ofAbnormal Psychology, in press.

Lichtenberg, P. A definition and analysis of de-pression. Archives of Neurology and Psychiatry,1957, 77, 516-527.

Mackintosh, N. J. Stimulus selection: Learning toignore stimuli that predict no change in rein-forcement. In R. A. Hinde & J. Stevenson-Hinde(Eds.), Constraints on learning. New York:Academic Press, 1973.

Mackintosh, N. J. A theory of attention: Varia-tions in the associability of stimuli with rein-forcement. Psychological Review, 1975, 82, 276-298.

Maier, S. F., Albin, R. W., & Testa, T. J. Failureto learn to escape in rats previously exposed toinescapable shock depends on nature of escaperesponse. Journal of Comparative and Physio-logical Psychology, 1973, 85, 581-592.

Maier, S. F., & Seligman, M. E. P. Learned help-lessness : Theory and evidence. Journal of Experi-mental Psychology: General, 1976, 105, 3-46.

Maier, S. F., Seligman, M. E. P., & Solomon, R.L. Pavlovian fear conditioning and learned help-lessness. In B. A. Campbell & R. M. Church(Eds.), Punishment and aversive behavior. NewYork: Appleton-Century-Crofts, 1969.

Maier, S. F., & Testa, T. J. Failure to learn toescape by rats previously exposed to inescapableshock is partly produced by associative inter-ference. Journal of Comparative and Physio-logical Psychology, 1975, 88, 554-564.

Medow, H., & Zander, A. Aspirations for thegroup chosen by central and peripheral mem-bers. Journal of Personality and Social Psychol-ogy, 1965, 1, 224-228.

Melges, F. J., & Bowlby, J. Types of hopelessnessin psychopathological process. Archives of Gen-eral Psychiatry, 1969, 20, 690-699.

Metcalfe, M., & Goldman, E. Validation of aninventory for measuring depression. British Jour-nal of Psychiatry, 1965, 111, 240-242.

Miller, D. T. What constitutes a self-servingattributional bias? A reply to Bradley. Journalof Personality and Social Psychology, 1978, 36,1221-1223.

Miller, D. T., & Ross, M. Self-serving biases inthe attribution of causality: Fact or fiction?Psychological Bulletin, 1975, 82, 213-225.

Miller, W. R., Rosellini, R., & Seligman, M. E. P.Learned helplessness and depression. In J. D.Maser & M. E. P. Seligman (Eds.), Psycho-pathology: Experimental models. San Francisco:Freeman, 1977.

Miller, W. R., & Seligman, M. E. P. Depressionand the perception of reinforcement. Journal ofAbnormal Psychology, 1973, 82, 62-73.

Miller, W. R., & Seligman, M. E. P. Depression

and learned helplessness in man. Journal of Ab-normal Psychology, 1975, 84, 228-238.

Miller, W. R., & Seligman, M. E. P. Learned help-lessness, depression, and the perception of rein-forcement. Behaviour Research and Therapy,1976,14, 7-17.

Moskowitz, A., & LoLordo, V. M. Role of safetyin the Pavlovian backward fear conditioningprocedure. Journal of Comparative and Physio-logical Psychology, 1968, 66, 673-678.

Nelson, R. E., & Craighead, W. E. Selective recallof positive and negative feedback, self-controlbehaviors, and depression. Journal of AbnormalPsychology, 1977, 86, 379-388.

Overmier, J. B. Interference with avoidance be-havior : Failure to avoid traumatic shock. Journalof Experimental Psychology, 1968, 78, 340-343.

Overmier, J. B., & Seligman, M. E. P. Effects ofinescapable shock upon subsequent escape andavoidance learning. Journal of Comparative andPhysiological Psychology, 1967, 63, 23-33.

Padilla, A. M., Padilla, C, Ketterer, J., & Gia-calone, D. Inescapable shocks and subsequentavoidance conditioning in goldfish (Carrasiusauratus). Psychonomic Science, 1970, 20, 295-296.

Phares, E. J. Expectancy change in chance andskill situations. Journal of Abnormal and SocialPsychology, 1957, 54, 339-342.

Piaget, J., & Inhelder, B, The origin of the ideaof chance in children. New York: Norton, 1975.

Premack, D. Reinforcement theory. In D. Levine(Ed.), Nebraska Symposium on Motivation(Vol. 13). Lincoln: University of NebraskaPress, 1965.

Prokasy, W. F, Classical eyelid conditioning: Ex-perimental operations, task demands, and re-sponse shaping. In W. F. Prokasy (Ed.), Class-ical conditioning, New York: Appleton-Century-Crofts, 1965.

Quinsey, V. L. Conditioned suppression with noCS-US contingency in the rat. Canadian Journalof Psychology, 1971, 25, 69-82.

Reiss, S., & Wagner, A. R. CS habitation pro-duces a "latent inhibition effect" but no activeconditioned inhibition. Learning and Motivation,1972, 3, 237-245.

Rescorla, R. A. Pavlovian conditioning and itsproper control procedures. Psychological Review,1967, 74, 71-79.

Rescorla, R. A. Probability of shock in the pres-ence and absence of the CS in fear conditioning.Journal of Comparative and Physiological Psy-chology, 1968, 66, 1-5.

Rescorla, R. A. Conditioned inhibition of fearresulting from negative CS-US contingencies.Journal of Comparative and Physiological Psy-chology, 1969, 67, 504-509. (a)

Rescorla, R. A. Establishment of a positive rein-forcer through contrast with shock. Journal ofComparative and Physiological Psychology, 1969,67, 260-263. (b)

Rescorla, R. A. Pavlovian conditioned inhibition.Psychological Bulletin, 1969, 72, 77-94. (c)


Rescorla, R. A. Summation and retardation testsof latent inhibition. Journal of Comparative andPhysiological Psychology, 1971, 75, 77-81.

Rescorla, R. A. Informational variables in Pav-lovian conditioning. In G. H. Bower (Ed.),The psychology of learning and motivation(Vol. 6). New York: Academic Press, 1972.

Rescorla, R. A., & LoLordo, V. M. Inhibition ofavoidance behavior. Journal of Comparative andPhysiological Psychology, 1965, 59, 406-412.

Rescorla, R. A., & Wagner,,A. R. A theory ofPavlovian conditioning: Variations in the effec-tiveness of reinforcement and nonreinforcement.In A. H. Black & W. F. Prokasy (Eds.),Classical conditioning II: Current theory andresearch. New York: Appleton-Century-Crofts,1972.

Rizley, R. Depression and distortion in the attribu-tion of causality. Journal of Abnormal Psychol-ogy, 1978, 87, 32-48.

Roth, S., & Bootzin, R. R. Effects of experimen-tally induced expectancies of external control:An investigation of learned helplessness. Journalof Personality and Social Psychology, 1974, 29,253-264.

Roth, S., & Kubal, L. The effects of noncontingentreinforcement on tasks of differing importance:Facilitation and learned helplessness effects.Journal of Personality and Social Psychology,1975, 32, 680-691.

Rotter, J. B. Generalized expectancies of internalversus external control of reinforcement. Psycho-logical Monographs, 1966, 80(1, Whole No. 609).

Rotter, J. B., Liverant, S., & Crowne, E. P. Thegrowth and extinction of expectancies in chancecontrolled and skilled tasks. Journal of Psychol-ogy, 1961, 52, 161-177.

Rozensky, R. H., Rehm, L. P., Pry, G., & Roth, D.Depression and self-reinforcement behavior inhospitalized patients. Journal of Behavior Ther-apy and Experimental Psychiatry, 1977, 8, 35-38.

Seligman, M. E. P. Depression and learned help-lessness. In R. J. Friedman & M. M. Katz(Eds.), The psychology of depression: Contem-porary theory and research. New York: Wiley,1975. (a)

Seligman, M. E. P. Helplessness: On depression,development, and death. San Francisco: Freeman,1975. (b)

Seligman, M. E. P., Abramson, L. Y., Semmel, A.,& von Baeyer, C, Depressive attributional style.Journal of Abnormal Psychology, 1979, 88, 242-247.

Seligman, M. E. P., & Beagley, G. Learned help-lessness in the rat. Journal of Comparative andPhysiological Psychology, 1975, 88, 534-541.

Seligman, M. E. P., & Groves, D. Non-transientlearned helplessness. Psychonomic Science, 1970,19, 191-192.

Seligman, M. E. P., Klein, D. C., & Miller, W. R.Depression. In H. Leitenberg (Ed.), Handbookof behavior modification and behavior therapy,Englewood Cliffs, NJ.: Prentice-Hall, 1976.

Seligman, M. E. P., & Maier, S. F. Failure toescape traumatic shock. Journal of ExperimentalPsychology, 1967, 74, 1-9.

Seligman, M. E. P., Maier, S. F., & Solomon, R.L. Unpredictable and uncontrollable aversiveevents. In F. R. Brush (Ed.), Aversive con-ditioning and learning. New York: AcademicPress, 1971.

Seligman, M. E. P., Rosellini, R. A., & Kozak,M. J. Learned helplessness in the rat: Timecourse, immunization, and reversibility. Journalof Comparative and Physiological Psychology,1975, 88, 542-547.

Seward, J. P., & Humphrey, G. L. Avoidancelearning as a function of pretraining in the cat.Journal of Comparative and Physiological Psy-chology, 1967, 63, 338-341.

Siegel, S., & Domjan, M. Backward conditioningas an inhibitory procedure. Learning and Moti-vation, 1971, 2, 1-11.

Skinner, B, F. "Superstition" in the pigeon. Journalof Experimental Psychology, 1948, 38, 168-172.

Smedslund, J. The concept of correlation in adults.Scandinavian Journal of Psychology, 1963, 4,165-173.

Staddon, J. E. R., & Simmelhag, V. J. The super-stition experiment: A re-examination of itsimplications for the study of adaptive behavior.Psychological Review, 1971, 78, 3-43.

Starr, B. J., & Katkin, E. S. The clinican as anaberrant actuary: Illusory correlation and theincomplete sentence blank. Journal of AbnormalPsychology, 1969, 74, 670-675.

Streufert, S., & Streufert, S. C. The effects ofconceptual structure, failure, and success onattributions of causality and interpersonal atti-tudes. Journal of Personality and Social Psy-chology, 1969, 11, 138-147.

Thornton, J. W., & Jacobs, P. D. Learned helpless-ness in human subjects. Journal of ExperimentalPsychology, 1971, 87, 369-372.

Tversky, A., & Kahneman, D. Belief in the law ofsmall numbers. Psychological Bulletin, 1971, 76,105-110.

Tversky, A., & Kahneman, D. Availability: A heu-ristic for judging frequency and probability.Cognitive Psychology, 1974, 5, 207-232.

Wagner, A. R., & Rescorla, R. A. Inhibition inPavlovian conditioning: Application of a theory.In R. A. Boakes & S. Halliday (Eds.), Inhibi-tion and learning. New York: Academic Press,1972.

Ward, W. C., & Jenkins, H. M. The display ofinformation and the judgment of contingency.Canadian Journal of Psychology, 1965, 19, 231-241.

Weiner, B., et al. Perceiving the causes of successand failure. Morristown, N. J.: General Learn-ing Press, 1971.

Weiner, B., Nierenberg, R., & Goldstein, M. Sociallearning (locus of control) versus attributional(causal stability) interpretations of expectancy


of success. Journal of Personality, 1976, 44, 52-68.

Weisman, R. G., & Litner, J. S. The causes ofPavlovian excitation and inhibition of fear inrats. Journal of Comparative and PhysiologicalPsychology, 1969, 69, 667-672.

Weisman, R. G., & Litner, J. S. Role of the inter-trial interval in Pavlovian differential condition-ing of fear in rats. Journal of Comparative andPhysiological Psychology, 1972, 74, 211-218.

Weiss, J. M. Effects of coping responses on stress.Journal of Comparative and Physiological Psy-chology, 1968, 65, 251-260.

Williams, J. L., & Maier, S. F. Transnationalimmunization and therapy of learned helplessnessin the rat. Journal of Experimental Psychology:Animal Behavior Processes, 1977, 3, 240-252.

Winer, B. J. Statistical principles in experimentaldesign. New York: McGraw-Hill, 1962.

Wortman, C. B. Some determinants of perceivedcontrol. Journal of Personality and Social Psy-chology, 1975, 31, 282-294.

Wortman, C. B., Costanzo, R. P., & Witt, J. R.Effects of anticipated performance on the attribu-tions of causality to self and others. Journal ofPersonality and Social Psychology, 1973, 27,372-381.

Wright, J. C. Consistency and complexity ofresponse sequences as a function of schedules ofnoncontingent reward. Journal of ExperimentalPsychology, 1962, 63, 601-609.

Zuckerman, M., & Lubin, B. Manual for the Mul-tiple Affect Adjective Check List. San Diego,Calif.: Education and Industrial Testing Service,1965.

Zung, W. K. A self-rating depression scale.Archives of General Psychiatry, 1965, 12, 63-70.

Received June 23, 1978 •

Documents

Alloy & Abramson - University of Nottingham · Y. Abramson while both authors were at the University of Pennsylvania. National Institute of Mental Health Grant MH 19604 to M. E. P