Telling More Than We Can Know

Psychological Review<f Copyright © 1977 C—J by the American Psychological Association, Inc.

V O L U M E 8 4 N U M B E R 3 M A Y 1 9 7 7

Telling More Than We Can Know:Verbal Reports on Mental Processes

Richard E. Nisbett and Timothy DeCamp WilsonUniversity of Michigan

Evidence is reviewed which suggests that there may be little or no direct intro-spective access to higher order cognitive processes. Subjects are sometimes (a)unaware of the existence of a stimulus that importantly influenced a response,(b) unaware of the existence of the response, and (c) unaware that the stimulushas affected the response. It is proposed that when people attempt to report ontheir cognitive processes, that is, on the processes mediating the effects of astimulus on a response, they do not do so on the basis of any true introspection.Instead, their reports are based on a priori, implicit causal theories, or judg-ments about the extent to which a particular stimulus is a plausible cause of agiven response. This suggests that though people may not be able to observedirectly their cognitive processes, they will sometimes be able to report accu-rately about them. Accurate reports will occur when influential stimuli are salientand are plausible causes of the responses they produce, and will not occur whenstimuli are not salient or are not plausible causes.

"Why do you like him?" "How did you solvethis problem?" "Why did you take that job?"

In our daily lives we answer many such ques-tions about the cognitive processes underlyingour choices, evaluations, judgments, and be-havior. Sometimes such questions are asked bysocial scientists. For example, investigatorshave asked people why they like particular po-

The writing of this paper, and some of the researchdescribed, was supported by grants GS-40085 andBNS75-23191 from the National Science Foundation.The authors are greatly indebted to Eugene Borgida,Michael Kruger, Lee Ross, Lydia Temoshok) and AmosTversky for innumerable ideas and generous and con-structive criticism. John W. Atkinson, Nancy Bellows,Dorwin Cartwright, Alvin Goldman, Sharon Gurwitz,Ronald Lemley, Harvey London, Hazel Markus, Wil-liam R. Wilson, and Robert Zajonc provided valuablecritiques of earlier drafts of the paper.

Requests for reprints should be sent to Richard E.Nisbett, Research Center for Group Dynamics, Insti-tute for Social Research, University of Michigan, AnnArbor, Michigan 48109.

litical candidates (Gaudet, 1955) or detergents(Kornhauser & Lazarsfeld, 1935), why theychose a particular occupation (Lazarsfeld,1931), to go to graduate school (Davis, 1964)or to become a juvenile delinquent (Burt, 1925),why they got married or divorced (Goode,1956) or joined a voluntary organization (Sills,1957) or moved to a new home (Rossi, 1955)or sought out a psychoanalyst (Kadushin,1958), or failed to use a contraceptive tech-nique (Sills, 1961). Social psychologists rou-tinely ask the subjects in their experiments whythey behaved, chose, or evaluated as they did.Indeed, some social psychologists have advo-cated the abandonment of the social psychol-ogy experiment and its deceptive practices andhave urged that subjects simply be asked howtheir cognitive processes would work if theywere to be confronted with particular stimulussituations (Brown, 1962; Kelman, 1966).

Recently, however, several cognitive psy-chologists (Mandler, 1975a, 1975b; Miller,

231

232 RICHARD E. NISBETT AND TIMOTHY DECAMP WILSON

1962; Neisser, 1967) have proposed that wemay have no direct access to higher order men-tal processes such as those involved in evalua-tion, judgment, problem solving, and theinitiation of behavior. The following quotationswill serve to indicate the extent to which theseinvestigators doubt people's ability to observedirectly the workings of their own minds. "Itis the result of thinking, not the process ofthinking, that appears spontaneously in con-sciousness" (Miller, 1962, p. 56). "The con-structive processes [of encoding perceptualsensations] themselves never appear in con-sciousness, their products do" (Neisser, 1967,p. 301). And in Neisser's next paragraph: "Thisgeneral description of the fate of sensory in-formation seems to fit the higher mental pro-cesses as well" (p. 301). Handler's (1975a) sug-gestions are still more sweeping: "The analysisof situations and appraisal of the environment. . . goes on mainly at the nonconscious level"(p. 241). "There are many systems that cannotbe brought into consciousness, and probablymost systems that analyze the environment inthe first place have that characteristic. In mostof these cases, only the products of cognitiveand mental activities are available to conscious-ness" (p. 245). And finally: "unconscious pro-cesses . . . include those that are not availableto conscious experience, be they feature analy-zers, deep syntactic structures, affective ap-praisals, computational processes, languageproduction systems, action systems of manykinds" (p. 230).

It is important to note that none of thesewriters cites data in support of the view thatpeople have no direct access to higher ordermental processes. In fact, when the above quo-tations are read in context, it is clear that thesource of the speculations is not research onhigher order processes such as "thinking,""affective appraisal," and "action systems,"but rather research on more basic processes ofperception and memory. Recent research hasmade it increasingly clear that there is almostno conscious awareness of perceptual and mem-orial processes. It would be absurd, for example,to ask a subject about the extent to which herelied on parallel line convergence when mak-ing a judgment of depth or whether he storedthe meanings of animal names in a hierarchicaltree fashion or in some other manner. Miller

(1962) has provided an excellent example ofour lack of awareness of the operation of mem-orial processes. If a person is asked, "What isyour mother's maiden name?", the answer ap-pears swiftly in consciousness. Then if the per-son is asked "How did you come up withthat?", he is usually reduced to the inarticulateanswer, "I don't know, it just came to me."

It is a substantial leap, however, from re-search and anecdotal examples concerning per-ception and memory to blanket assertionsabout higher order processes. In the absenceof evidence indicating that people cannot cor-rectly report on the cognitive processes under-lying complex behaviors such as judgment,choice, inference, and problem solving, socialscientists are not likely to abandon their prac-tice of quizzing their subjects about such pro-cesses. The layman is even less likely toabandon his habit of asking and answering suchquestions.

A second problem with the new anti-intro-spectivist view is that it fails to account for thefact, obvious to anyone who has ever ques-tioned a subject about the reasons for his be-havior or evaluations, that people readilyanswer such questions. Thus while people usu-ally appear stumped when asked about per-ceptual or memorial processes, they are quitefluent when asked why they behaved as theydid in some social situation or why they like ordislike an object or another person. It wouldseem to be incumbent on one who takes a posi-tion that denies the possibility of introspectiveaccess to higher order processes to account forthese reports by specifying their source. If it isnot direct introspective access to a memory ofthe processes involved, what is the source ofsuch verbal reports?

Finally, a third problem with the anti-intro-spectivist view is that it does not allow for thepossibility that people are ever correct in theirreports about their higher order mental pro-cesses. It seems intuitively unlikely that suchreports are always inaccurate. But if people aresometimes accurate, several questions arise, (a)What is the basis of these accurate reports? (b)Are accurate reports fundamentally differentin kind from inaccurate ones? (c) Is it possibleto specify what sorts of reports will be accurateand what sorts will be inaccurate?

The first part of this article is concerned with

VERBAL REPORTS ON MENTAL PROCESSES 233

a review of the evidence bearing on the accur-acy of subjective reports about higher mentalprocesses. The second part of the paper pre-sents an account of the basis of such reports.We shall argue for three major conclusions.

1. People often cannot report accurately onthe effects of particular stimuli on higher order,inference-based responses. Indeed, sometimesthey cannot report on the existence of criticalstimuli, sometimes cannot report on the exist-ence of their responses, and sometimes cannoteven report that an inferential process of anykind has occurred. The accuracy of subjectivereports is so poor as to suggest that any intro-spective access that may exist is not sufficientto produce generally correct or reliable reports.

2. When reporting on the effects of stimuli,people may not interrogate a memory of thecognitive processes that operated on the stim-uli ; instead, they may base their reports on im-plicit, a priori theories about the causal con-nection between stimulus and response. If thestimulus psychologically implies the responsein some way (Abelson, 1968) or seems "repre-sentative" of the sorts of stimuli that influencethe response in question (Tversky & Kahne-man, 1974), the stimulus is reported to haveinfluenced the response. If the stimulus doesnot seem to be a plausible cause of the response,it is reported to be noninfluential.

3. Subjective reports about higher mentalprocesses are sometimes correct, but even theinstances of correct report are not due to directintrospective awareness. Instead, they are dueto the incidentally correct employment of apriori causal theories.

Verbal Reports on Cognitive Processes inDissonance and Attribution Studies

Much of the evidence that casts doubt on theability of people to report on their cognitiveprocesses comes from a study of the literaturethat deals with cognitive dissonance and self-perception attribution processes. Or rather, theevidence comes from a consideration of whatwas not published in that literature. A reviewof the nonpublic, sub rosa aspects of these in-vestigations leads to three conclusions: (a)Subjects frequently cannot report on the exist-ence of the chief response that was produced

by the manipulations; (b) even when they areable to report the existence of the responses,subjects do not report that a change process oc-curred, that is, that an evaluational or attitudi-nal response underwent any alterations; and(c) subjects cannot correctly identify thestimuli that produced the response.

Awareness of the Existenceof the Response

The central idea of insufficient justificationor dissonance research is that behavior that isintrinsically undesirable will, when performedfor inadequate extrinsic reasons, be seen asmore attractive than when performed for ade-quate extrinsic reasons. In the view of Fest-inger (1957) and other dissonance theorists, at-titude change occurs because the cognition "Ihave done something unpleasant without ade-quate justification" is dissonant and thereforepainful; and the person revises his opinionabout the behavior in order to avoid thepsychic discomfort.

The central idea of attribution theory is thatpeople strive to discover the causes of attitudi-nal, emotional, and behavioral responses (theirown and others), and that the resulting causalattributions are a chief determinant of a host ofadditional attitudinal and behavioral effects.Thus, for example, if someone tells us that aparticular Western movie is a fine film, our ac-ceptance of that opinion, and possibly our sub-sequent behavior, will be determined by ourcausal analysis of the person's reasons for theevaluation: Does he like all movies? All West-erns? All John Wayne movies? Do other peoplelike the movie? Does this person tend to likemovies that other people do not like?

Many insufficient-justification studies andmany attribution studies where the subjectmakes inferences about himself have employedbehavioral dependent variables. Substantialeffects have been shown on behavior of inherentinterest and with significant social implications,including pain, hunger and thirst tolerance,psychopathology, task perseverance, and ag-gressive behavior. Two examples will serve toillustrate research with behavioral conse-quences.

Zimbardo, Cohen, Weisenberg, Dworkin,and Firestone (1969) asked subjects to accept


a series of painful electric shocks while per-forming a learning task. When the task wascompleted, subjects were asked to repeat it.Some subjects were given adequate justifica-tion for performing the task a second time andaccepting another series of shocks (the researchwas very important, nothing could be learnedunless the shocks were given again), while othersubjects were given insufficient justification(the experimenter wanted to satisfy his moreor less idle curiosity about what would happen).Subjects with insufficient justification for ac-cepting the shocks showed lower GSR re-sponses and better learning performance onthe second task than subjects with sufficientjustification. The explanation offered for thisfinding is that insufficient-justification subjectssought to justify taking the shocks, which theydid by deciding that the shocks were not allthat painful. Thus the evaluation of the pain-fulness of the shocks was lowered, and physio-logical and behavioral indicators reflected thisevaluation.

A study by Valins and Ray (1967) will illu-strate the attribution paradigm. These inves-tigators asked snake-phobic subjects to watchslides while receiving occasional electric shocks.Subjects were wired for what they believedwere recordings of heart rate. They were al-lowed to hear a rhythmic pattern of soundswhich, they were told, was the amplified soundof their own heart beats. Subjects were showna series of slides of snakes interspersed withslides of the word "SHOCK." Following each pre-sentation of the shock slide, subjects weregiven an electric shock. After a few such pair-ings, the appearance of the shock slide was ac-companied by an increased rate of "heart-beats." Snake slides were never accompaniedby any change in apparent heart rate. Follow-ing this procedure, subjects were requested toapproach, and if possible, to touch, a 30-inch(76.2-cm) boa constrictor. Such subjects ap-proached the snake more closely than subjectswho had gone through the identical procedurebut who believed that the "heartbeats" weresimply "extraneous sounds" (which, of course,they actually were). The finding is explained asfollows. Subjects in the heart rate conditionlearned that their "heart rate" indicated theywere appropriately frightened when they sawthe shock slide, because of the electric shock it

portended, but that they were not frightenedby the snake slides. If they were not frightenedby the snake slides, perhaps they were not asafraid of live snakes as they had thought.Armed with this new self-attribution of snakefearlessness, they were more willing to approachthe boa.

The two experiments just described share acommon formal model. Verbal stimuli in theform of instructions from the experimenter, to-gether with the subject's appraisal of the stim-ulus situation, are the inputs into a fairly com-plicated cognitive process which results in achanged evaluation of the relevant stimuli andan altered motivational state. These motiva-tional changes are reflected in subsequent phys-iological and behavioral events. Thus: stimuli—»cognitive process —» evaluative and motiva-tional state change —> behavior change. Follow-ing traditional assumptions about higher men-tal processes, it has been tacitly assumed byinvestigators that the cognitive processes inquestion are for the most part verbal, consciousones. Thus the subject consciously decides howhe feels about an object, and this evaluationdetermines his behavior toward it. As severalwriters (Bern, 1972; Nisbett & Valins, 1972;Storms & Nisbett, 1970; Weick, 1966) havepointed out, there is a serious problem with thisimplicit assumption: Typically, behavioral andphysiological differences are obtained in theabsence of verbally reported differences in evalu-ations or motive states. For example, in thestudy by Zimbardo, Cohen, Weisenberg,Dworkin, and Firestone (1969), experimentalsubjects given inadequate justification for tak-ing shock learned much more quickly and showedmuch less GSR reactivity to the shock than didcontrol, adequate-justification subjects, but theformer did not report the shock to be signifi-cantly less painful than did the latter. And sub-jects in the Valins and Ray (1967) experimentwho had "inferred" that they were not veryfrightened of snakes, as indicated by their will-ingness to approach the boa constrictor, showedno evidence of any such inference when asked adirect question about how frightened they wereof snakes.

We have reviewed all the insufficient-justi-fication and attribution studies we have beenable to find that meet the following criteria: (a)behavioral or physiological effects were ex-


amined, and (b) at approximately the sametime, verbal reports of evaluations and motiva-tional states were obtained. Studies that didnot permit a clear, uncontroversial comparisonof the strength of behavioral and self-report in-dicators were not included (e.g., Brehm, Back,& Bogdonoff, 1969; Schachter & Singer, 1962),nor were studies that employed controversial,poorly understood techniques such as hypnosis(e.g., Brock & Grant, 1969).

Three striking generalizations can be madeabout these studies:

1. In the majority of studies, no significantverbal report differences were found at all.This applies to studies by Cohen and Zimbardo(1969), Cottrell and Wack (1967), Davisonand Valins (1969), Ferdinand (1964), Freed-man (1965), Grinker (1969), Pallak (1970),five experiments by Pallak, Brock, and Kiesler(1967), Experiment 1 in Pallak and Pittman(1972), Schachter and Wheeler (1962), Snyder,Schultz, and Jones (1974), Storms and Nisbett(1970), Valins and Ray (1967), Waterman(1969), Weick and Penner (1969), Weick andPrestholdt (1968), and Zimbardo, Cohen, Weis-enberg, Dworkin, and Firestone (1969).

2. In the remainder of studies, the behavi-oral effects were in most cases stronger (i.e.,more statistically reliable) than the verbal re-port effects (Berkowitz & Turner, 1974; Kru-glanski, Friedman, & Zeevi, 1971; Schlachet,1969; Nisbett & Schachter, 1966; Experiment2 of Pallak and Pittman, 1972; and Weick,1964). Exceptions to this are reports by Brehm(1969), Freedman (1963), Mansson (1969),and Zimbardo, Weisenberg, Firestone, andLevy (1969).

3. In two studies where it was reported, thecorrelation between verbal report about motivestate and behavioral measures of motive statewas found to be nil (Storms & Nisbett, 1970;Zimbardo, Cohen, Weisenberg, Dworkin, &Firestone, 1969). The rest of the literature inthis area is strangely silent concerning thecorrelations between verbal report and be-havior. Since positive correlations would haveconstituted support for investigators' hypothe-ses, while zero or negative correlations wouldhave been difficult to understand or interpretin terms of prevailing assumptions about thenature of the cognitive processes involved, the

failure to report the correlations constitutespresumptive evidence that they were not posi-tive. In order to check on this possibility, wewrote to the principal investigators of thestudies described above, asking for the correla-tions between verbal report and behavior.Only three investigators replied by saying thatthey still had the data and could provide thecorrelations. In all three instances, the corre-lations were in fact nonsignificant and close tozero (Davison & Valins, 1969; Freedman,1965; Snyder, Schultz, & Jones, 1974).

The overall results thus confound any as-sumption that conscious, verbal cognitive pro-cesses result in conscious, verbalizable changesin evaluations or motive states which thenmediate changed behavior. In studies where thedata are available, no association is found be-tween degree of verbal report change and de-gree of behavior change in experimental groups.And in most studies no evidence is found thatexperimental subjects differ from control sub-jects in their verbal reports on evaluations andmotivational states.

What of the studies that do find differencesin the verbal reports of experimental and con-trol subjects? (It should be noted that this in-cludes many studies not reviewed here wherethe only dependent measure was a verbal oneand where differences between experimentaland control groups were obtained.) Shouldthese studies be taken as evidence that thetraditional model sometimes works, that sub-jects are sometimes aware of the cognitive pro-cesses that occur in these experiments? Evi-dence to be discussed below casts doubt on sucha conclusion.

Awareness of the Existenceof a Change Process

There is an important difference betweenawareness of the existence of an evaluation ormotive state and awareness of a changed evalu-ation or motive state. The former sort of aware-ness does not imply true recognition of the pro-cess induced by insufficient justification andattribution manipulations—which in fact al-ways involves a change in evaluations. Thus ifit could be shown that subjects cannot reporton the fact that a change has taken place as a


consequence of such manipulations, this wouldsuggest that they are not aware of the occur-rence of a process.

Bern and McConnell (1970) contrived an ex-periment to demonstrate that in fact, subjectsdo not experience a subjective change in theirevaluations in response to insufficient-justifica-tion manipulations. A stock-in-trade of the dis-sonance tradition is the counterattitudinaladvocacy experiment. In this type of experi-ment, subjects are asked to write an essay op-posing their own views on some topic and arethen asked what their attitudes are toward thetopic. Subjects who are coerced (or heavilybribed) into writing the essays show no changein evaluation of the topic. Subjects who aregiven insufficient justification for writing theessay, or who are manipulated into believingthat they had free choice in the matter, typic-ally shift their evaluations in the direction ofthe position advocated in the essay. On the faceof it, this would seem to indicate that subjectsare aware of the existence of a change processsince the means employed for assessing the re-sponse is a verbal report, and this reportchanges from premanipulation measures topostmanipulation measures.

Bern and McConnell contested this assump-tion by the simple expedient of asking the sub-jects, at the time of the postmanipulation mea-sure, what their attitude had been 1 week earlier,at the time of the premanipulation measure.Control subjects had no difficulty reporting ac-curately on their previous opinions. In contrast,though the postmanipulation attitudes of ex-perimental subjects were substantially differ-ent from their premanipulation attitudes, theyreported that their current attitudes were thesame as their premanipulation attitudes. Thussubjects apparently changed their attitudes inthe absence of any subjective experience ofchange. This suggests that though subjects cansometimes report on the existence of the newevaluation, they may still be unaware of thefact that the evaluation has changed. If so,then they cannot be aware of the nature of thecognitive process that has occurred, becausethey are not even aware of the fact that a pro-cess has occurred at all.

Such a conclusion gains credence in view of atruly stunning demonstration of the same phe-nomenon by Goethals and Reckman (1973).

These investigators assessed the opinions ofhigh school students on 30 social issues, in-cluding attitudes toward busing of schoolchildren to achieve racial integration. One totwo weeks later, students were called and askedto participate in a group discussion of the bus-ing issue. Each group was composed of threesubjects whose pretest opinions indicated thatthey were all pro-busing or all anti-busing, plusone high school student confederate who' wasarmed with a number of persuasive opinionsand whose job it was to argue persistentlyagainst the opinion held by all other groupmembers. He was highly successful in this task.Following the discussion, subjects indicatedtheir opinions on the busing issue—on a scaledifferent in form from the original measure.The original anti-busing subjects had theiropinions sharply moderated in a pro-direction.Most of the pro-busing subjects were actuallyconverted to an anti-busing position. ThenGoethals and Reckman asked their subjects torecall, as best they could, what their originalopinions on the busing question had been. Sub-jects were reminded that the experimenterswere in possession of the original opinion scaleand would check the accuracy of the subjects'recall. Control subjects were able to recall theiroriginal opinions with high accuracy. In con-trast, among experimental subjects, the originalanti-busing subjects "recalled" their opinionsas having been much more pro-busing thanthey actually were, while the original pro-bus-ing subjects actually recalled their originalopinions as having been, on the average, anti-busing! In fact, the original pro-busing sub-jects recalled that they had been more anti-busing than the original anti-busing subjectsrecalled that they had been.

It would appear that subjects in the Goe-thals and Reckman (1973) study did not actu-ally experience these enormous shifts as opinionchange:

Some subjects listened carefully to the course of thediscussion and began to nod their heads in agreementwith the confederate's arguments. They seemed to cometo agree with him without any awareness of their earlierattitude. In the debriefing they gave every indicationthat the position they adopted after the discussion wasthe position they had basically always held. . . .Mostcommented that the discussion had served to broadentheir awareness of the issues involved or had provided


support for their original position. No subject reportedthat the discussion had had any effect in changing ormodifying his position, (p. 499)

Thus research in the insufficient-justificationand attribution traditions seems to indicatethat (a) subjects sometimes do not report theevaluational and motivational states producedin these experiments; and (b) even when theycan report on such states, they may not reportthat a change has taken place in these states.

It may have occurred to the reader that themost direct approach to the question of accur-acy of subjects' reports in these experimentswould be simply to ask subjects why they be-haved as they did and listen to what they haveto say about their own cognitive processes.This would indeed be a fruitful approach, andit is discussed below.

Reports A bout Cognitive Processes

A literal reading of the literature would givethe impression that researchers working in theareas of insufficient-justification and attributionhave not bothered to ask their subjects abouttheir thought processes. We have been able tofind only a single report of the results of suchquestioning. This is the terse and intriguingreport by Ross, Rodin, and Zimbardo (1969)in their experiment on reattribution of arousalsymptoms that the subjects "never explicitlymentioned any conflict about, or searching for,the 'explanation' for their arousal. This sug-gests that attribution may never have beenconsciously debated by these subjects" (p.287). Fortunately, additional unpublisheddata, collected from subjects following theirparticipation in attribution experiments byNisbett and Schachter (1966) and Storms andNisbett (1970), are available. These data areconsistent with the description supplied byRoss et al.

In the experiment by Nisbett and Schachter(1966), subjects were requested to take a seriesof electric shocks of steadily increasing intens-ity. Prior to exposure to the shock, some of thesubjects were given a placebo pill which, theywere told, would produce heart palpitations,breathing irregularities, hand tremor, and but-terflies in the stomach. These are the physicalsymptoms most often reported by subjects as

accompanying the experience of electric shock.It was anticipated that when subjects withthese instructions were exposed to the shock,they would attribute their arousal symptomsto the pill, and would therefore be willing totolerate more shock than subjects who couldonly attribute these aversive symptoms to theshock. And, in fact, the pill attribution subjectstook four times as much amperage as shockattribution subjects.

Following his participation in the experi-ment, each subject in the pill attribution groupwas interviewed following a Spielberger-type(1962) graded debriefing procedure, (a) Ques-tion: "I notice that you took more shock thanaverage. Why do you suppose you did?" Typi-cal answer: "Gee, I don't really know. . . .Well, I used to build radios and stuff when Iwas 13 or 14, and maybe I got used to electricshock." (b) Question: "While you were takingthe shock, did you think about the pill at all?"Typical answer: "No, I was too worried aboutthe shock." (c) Question: "Did it occur to youat all that the pill was causing some physicaleffects?" Typical answer: "No, like I said, Iwas too busy worrying about the shock." Inall, only 3 of 12 subjects reported having madethe postulated attribution of arousal to the pill.(d) Finally, the experimenter described the hy-pothesis of the study in detail, including thepostulated process of attribution of symptomsto the pill. He concluded by asking the subjectif he might have had any thoughts like thosedescribed. Subjects typically said that the hy-pothesis was very interesting and that manypeople probably would go through the processthat the experimenter described, but so far asthey could tell, they themselves had not.

A similar blank wall was discovered byStorms and Nisbett (1970) in their experimenton the reattribution of insomnia symptoms.In that experiment, insomniac subjects wereasked to report, for 2 consecutive nights, onthe time they had gone to bed and the time theyhad finally gotten to sleep. Arousal conditionsubjects were then given a placebo pill to takeIS minutes before going to bed for the next 2nights. These subjects were told that the pillwould produce rapid heart rate, breathing ir-regularities, bodily warmth, and alertness—the physical and emotional symptoms, in otherwords, of insomnia. Relaxation subjects were


told that their pills would produce the oppositesymptoms—lowered heart rate, breathing rate,body temperature, and a reduction in alertness.It was anticipated that subjects in the arousalcondition would get to sleep more quickly onthe nights they took the pills because theywould attribute their arousal symptoms to thepills rather than to emotionally laden cogni-tions concerning work or social life. Relaxationsubjects were expected to take longer to get tosleep since they would infer that their emotionalcognitions must be particularly intense becausethey were as fully aroused as usual even thoughthey had taken a pill intended to lower arousal.These were in fact the results. Arousal subjectsreported getting to sleep 28% quicker on thenights with the pills, and relaxation subjects re-ported taking 42% longer to get to sleep. Sleeponset was unaffected for control subjects.

In the interview following completion of theexperiment, it was pointed out to subjects inexperimental conditions that they had reportedgetting to sleep more quickly (or more slowly)on experimental nights than on the previousnights, and they were asked why. Arousal sub-jects typically replied that they usually foundit easier to get to sleep later in the week, or thatthey had taken an exam that had worried thembut had done well on it and could now relax, orthat problems with a roommate or girlfriendseemed on their way to a resolution. Relaxationsubjects were able to find similar sorts of rea-sons to explain their increased sleeplessness.When subjects were asked if they had thoughtabout the pills at all before getting to sleep,they almost uniformly insisted that after tak-ing the pills they had completely forgottenabout them. When asked if it had occurred tothem that the pill might be producing (or count-eracting) their arousal symptoms, they reiter-ated their insistence that they had not thoughtabout the pills at all after taking them. Finally,the experimental hypothesis and the postulatedattribution processes were described in detail.Subjects showed no recognition of the hypothe-sized processes and (unlike subjects in the Nis-bett and Schachter study) made little pretenseof believing that any subjects could have gonethrough such processes.

Since many skilled and thorough investiga-tors have worked in the dissonance tradition,it seemed highly unlikely that the silence in

that literature concerning subjects' reports ontheir mental processes was due to simple failureto ask subjects the relevant questions. Instead,it seemed more likely that subjects had beenasked, and asked often, but that their answershad failed to reflect any ability to report theinferences that investigators believed to haveoccurred. If so, those investigators, like Nisbettand his colleagues, might have failed to reportthe answers because they made little sense interms of the traditional assumptions about theconscious, verbalizable nature of cognitiveprocesses. Accordingly, we contacted two of themost prolific and innovative researchers in thattradition—E. Aronson and P. Zimbardo—andasked them if they had ever quizzed their sub-jects about their mental processes. They hadindeed, with results similar to those describedabove.

Aronson (Note 1) responded as follows:

We occasionally asked our subjects why they had re-sponded as they did. The results were very disappoint-ing. For example, in the initiation experiment (Aronson& Mills, 1959), subjects did a lot of denying when askedif the punishment had affected their attitudes towardthe group or had entered into their thinking at all. WhenI explained the theory to the subjects, they typicallysaid it was very plausible and that many subjects hadprobably reasoned just the way I said, but not theythemselves.

Zimbardo (Note 2) gave a similar account:

Pretest subjects were routinely asked why they had be-haved as they did. I don't remember any subject whoever described anything like the process of dissonancereduction that we knew to have occurred. For example,in the shock experiment (Zimbardo, Cohen, Weisen-berg, Dworkin, & Firestone, 1969), we pointed out toexperimental subjects that they had learned morequickly the second time. A typical response would havebeen, "I guess maybe you turned the shock down." Or,in the grasshopper experiment (Zimbardo, Weisenberg,Firestone, & Levy, 1969), we asked subjects why theyhad been willing to eat a grasshopper. A typical responsewould have been, "Well, it was just no big deal whetherI ate a grasshopper or not."

Thus the explanations that subjects offer fortheir behavior in insufficient-justification andattribution experiments are so removed fromthe processes that investigators presume tohave occurred as to give grounds for consider-able doubt that there is direct access to theseprocesses. This doubt would remain, it shouldbe noted, even if it were eventually to be shown


that processes other than those posited by in-vestigators were responsible for the results ofthese experiments. Whatever the inferentialprocess, the experimental method makes itclear that something about the manipulatedstimuli produces the differential results. Yetsubjects do not refer to these critical stimuli inany way in their reports on their cognitiveprocesses.

As a final point, we note Kelley's (1967) ob-servation that the results of insufficient-justi-fication experiments could never be obtained ifsubjects were aware of the critical role playedby the social pressure from the experimenter.If subjects realized that their behavior was pro-duced by this social pressure, they would notchange their attitudes so as to move them intoline with their behavior, because they wouldrealize that their behavior was governed by thesocial pressure and not by their attitudes. Weconcur with Kelley's view that this fundamen-tal unawareness of the critical role of the ex-perimenter's behavior is essential to the errone-ous attitude inferences obtained in theseexperiments.

Other Research on Verbal ReportsAbout Cognitive Processes

There are at least five other literatures bear-ing on the question of the ability of subjects toreport accurately about the effects of stimulion complex, inferential responses: (a) Thelearning-without-awareness literature, (b) theliterature on subject ability to report accuratelyon the weights they assign to particular stimu-ulus factors in complex judgment tasks (re-viewed by Slovic & Lichtenstein, 1971), (c)some of the literature on subliminal perception,(d) the classic Maier (1931) work on awarenessof stimuli influencing problem solving, and (e)work by Latane and Darley (1970) on aware-ness of the effect of the presence of other peopleon helping behavior.

We shall discuss the first two areas of re-search in a later context, and it would take usfar afield to review the subliminal perceptionliterature in its entirety. Brief mention of thecurrent status of the subliminal perceptionquestion is in order, however, since it bears di-rectly on the issue of subject ability to reportaccurately on the effects of stimuli. If, as some

writers claim, stimuli can be responded to inthe literal absence of awareness of their exist-ence, then it logically follows that they couldnot possibly report on the influence of thosestimuli on their responses.

Subliminal Perception

The subliminal perception question has hada stormy, controversial history, chronicled byDixon (1971). It is fair to say, however, thatthe basic question of whether people can re-spond to a stimulus in the absence of the abilityto verbally report on its existence would todaybe answered in the affirmative by many moreinvestigators than would have been the case adecade ago. The reasons for this have been re-viewed by Dixon (1971) and Erdelyi (1974).The new acceptance rests on (a) methodologi-cal innovations in the form chiefly of signal de-tection techniques and dichotic listening pro-cedures and (b) persuasive theoretical argu-ments by Erdelyi (1974) and others that havesucceeded in deriving the subliminal perceptionphenomenon from the notion of selective at-tention and filtering (Broadbent, 1958; Moray,1969;Treisman, 1969).

An example of recent research employingsignal detection and dichotic listening proce-dures is provided by W. R. Wilson (1975).Wilson played tone sequences into the unat-tended auditory channel while subjects trackeda human voice in the attended channel. Sub-jects subsequently reported having heard notones, in fact, nothing at all, in the unattendedchannel. Moreover, in a signal detection task,subjects were presented (binaurally) with tonesequences which were either new or which hadpreviously been presented up to five times inthe unattended channel. Subjects were unableto discriminate new from old stimuli at a levelexceeding chance. Despite this fact, subjectsshowed the traditional familiarity effect onliking of the tone stimuli (Zajonc, 1968)."Familiar" tone sequences, that is, tone se-quences previously presented to the unattendedchannel, were preferred to novel stimuli. Wil-son argued that the experiment provides evi-dence that affective processes are triggered byinformation that is too weak to produce sub-sequent verbal recognition.

Results such as those provided by Wilson are


well understood in terms of recent theoreticaldevelopments in the field of attention and mem-ory. It is now generally recognized (Erdelyi,1974; Mandler, 1975b) that many more stimuliare apprehended than can be stored in short-term memory or transferred to long-term mem-ory. Thus, subliminal perception, once widelyregarded as a logical paradox ("How can weperceive without perceiving"), may be derivedas a logical consequence of the principle of se-lective filtering. We cannot perceive withoutperceiving, but we can perceive without remem-bering. The subliminal perception hypothesisthen becomes theoretically quite innocuous:Some stimuli may affect ongoing mental pro-cesses, including higher order processes ofevaluation, judgment, and the initiation of be-havior, without being registered in short-termmemory, or at any rate without being trans-ferred to long-term memory.

Thus if recent data and theory are correct intheir implications, it follows that subjects some-times cannot report on the existence of influen-tial stimuli. It therefore would be quite impos-sible for them to describe accurately the roleplayed by these stimuli in influencing their re-sponses ; and any subsequent verbal report bysubjects about the cause of their responseswould be at least partially in error.

Reports on Problem-Solving Processes

There is a striking uniformity in the waycreative people—artists, writers, mathemati-cians, scientists, and philosophers—speakabout the process of production and problemsolving. Ghiselin (1952) has collected into onevolume a number of essays on the creative pro-cess by a variety of creative workers fromPoincare to Picasso. As Ghiselin accurately de-scribed the general conclusion of these workers,"Production by a process of purely consciouscalculation seems never to occur" (p. 15). In-stead, creative workers describe themselves al-most universally as bystanders, differing fromother observers only in that they are the first towitness the fruits of a problem-solving processthat is almost completely hidden from consci-ous view. The reports of these workers arecharacterized by an insistence that (a) the in-fluential stimuli are usually completely obscure

—the individual has no idea what factorsprompted the solution; and (b) even the factthat a process is taking place is sometimes un-known to the individual prior to the pointthat a solution appears in consciousness.

Some quotations from Ghiselin's (1952) col-lection will serve to illustrate both these points.The mathematician Jacques Hadamard re-ports that "on being very abruptly awakenedby an external noise, a solution long searchedfor appeared to me at once without the slight-est instant of reflection on my part . . . andin a quite different direction from any of thosewhich I previously tried to follow" (p. 15).Poincare records that "the changes of travelmade me forget my mathematical work. Hav-ing reached Coutances, we entered an omnibusto go some place or other. At the moment whenI put my foot on the step the idea came to me,without anything in my former thoughts seem-ing to have paved the way for it, that the trans-formations I had used to define the Fuchsianfunctions were identical with those of non-Euclidean geometry" (p. 37).

Whitehead writes of "the state of imagina-tive muddled suspense which precedes success-ful inductive generalization" (Ghiselin, 1952,p. 15), and Stephen Spender describes "a dimcloud of an idea which I feel must be condensedinto a shower of words" (p. 15). Henry Jamesspeaks of his deliberate consignment of an ideato the realm of the unconscious where it can beworked upon and realized: "I was charmedwith my idea, which would take, however,much working out; and because it had so muchto give, I think, must I have dropped it for thetime into the deep well of unconscious cerebra-tion : not without the hope, doubtless, that itmight eventually emerge from that reservoir,as one had already known the buried treasureto come to light, with a firm iridescent surfaceand a notable increase of weight" (p. 26).

That mundane problem-solving in everydaylife differs little, in its degree of consciousness,from the problem-solving of creative geniuses,is indicated by the very elegant work of Maier,done some 45 years ago. In Maier's (1931)classic experiment, two cords were hung fromthe ceiling of a laboratory strewn with manyobjects such as poles, ringstands, clamps, pliers,and extension cords. The subject was told thathis task was to tie the two ends of the cords


together. The problem in doing so was that thecords were placed far enough apart that thesubject could not, while holding onto one cord,reach the other. Three of the possible solutions,such as tying an extension cord to one of theceiling cords, came easily to Maier's subjects.After each solution, Maier told his subjects"Now do it a different way." One of the solu-tions was much more difficult than the others,and most subjects could not discover it on theirown. After the subject had been stumped forseveral minutes, Maier, who had been wander-ing around the room, casually put one of thecords in motion, Then, typically within 45seconds of this cue, the subject picked up aweight, tied it to the end of one of the cords,set it to swinging like a pendulum, ran to theother cord, grabbed it, and waited for the firstcord to swing close enough that it could beseized. Immediately thereafter, Maier askedthe subject to tell about his experience of get-ting the idea of a pendulum. This question eli-cited such answers as "It just dawned on me.""It was the only thing left." "I just realized thecord would swing if I fastened a weight to it."A psychology professor subject was more in-ventive: "Having exhausted everything else,the next thing was to swing it. I thought of thesituation of swinging across a river. I hadimagery of monkeys swinging from trees. Thisimagery appeared simultaneously with the so-lution. The idea appeared complete."

Persistent probing after the free report suc-ceeded in eliciting reports of Maier's hint andits utilization in the solution of the problemfrom slightly less than a third of the subjects.This fact should be quickly qualified, however,by another of Maier's findings. Maier was ableto establish that one particular cue—twirling aweight on a cord—was a useless hint, that is,subjects were not aided in solving the problemby exposure to this cue. For some of the sub-jects, this useless cue was presented prior to thegenuinely helpful cue. All of these subjectsreported that the useless cue had been helpfuland denied that the critical cue had played anyrole in their solution. These inaccurate reportscast doubt on any presumption that even thethird of Maier's subjects who accurately re-ported that they used the helpful cue were re-porting such use on the basis of genuinely in-sightful introspection, since when they were

offered a false "decoy" cue they preferred it asan explanation for their solution.

Reports on the Effects of the Presence of Otherson Helping Behavior

Latane and Barley (1970) have shown, in alarge number of experiments in a wide varietyof settings, that people are increasingly lesslikely to help others in distress as the numberof witnesses or bystanders increases. Thus, forexample, the more people who overhear an in-dividual in another room having what soundslike an epileptic seizure, the lower the prob-ability that any given individual will rush tohelp. Latane and Darley early became in-trigued by the fact that their subjects seemedutterly unaware of the influence of the pres-ence of other people on their behavior. Accord-ingly, they systematically asked the subjectsin each of their experiments whether theythought they had been influenced the the pres-ence of other people. "We asked this questionevery way we knew how: subtly, directly, tact-fully, bluntly. Always we got the same answer.Subjects persistently claimed that their be-havior was not influenced by the other peoplepresent. This denial occurred in the face of re-sults showing that the presence of others didinhibit helping" (p. 124). It should also benoted that when Latane and Darley describedtheir experiments in detail to other subjectsand asked these subjects to predict how others,and they themselves, would behave when aloneor with other people present, these observersubjects uniformly agreed that the presence ofother people would have no effect on their ownor other people's behavior.

Thus the literature contains evidence fromdomains other than insufficient-justificationand attribution research suggesting that peoplemay have little ability to report accuratelyabout their cognitive processes. The subliminalperception literature suggests that people maysometimes be unable to report even the exist-ence of influential stimuli, and anecdotal re-ports of creative workers suggest that this mayfrequently be the case in problem-solving. Inaddition, these anecdotal reports suggest themost extreme form of inaccessibility to cogni-tive processes—literal lack of awareness that a


process of any kind is occurring until the mo-ment that the result appears. The work ofMaier and of Latane and Darley additionallysuggests that even when subjects are thor-oughly cognizant of the existence of the rele-vant stimuli, and of their responses, they maybe unable to report accurately about the in-fluence of the stimuli on the responses.

Demonstrations of Subject Inability to ReportAccurately on the Effects of Stimuli

on Responses

Though the evidence we have reviewed isconsistent with the skepticism expressed bycognitive psychologists concerning people'sability to introspect about their cognitive pro-cesses, the evidence is limited in several re-spects. Dissonance and attribution processesmay be unique in important ways. For example,deceptive practices are often employed instructuring the stimulus situations in such ex-periments, and these practices may result inpeople being misled in ways that do not nor-mally occur in daily life. The subliminal per-ception literature is controversial, and thoughthe new data and theoretical arguments haveproved to be convincing to many investigators,the critical response to these new developmentshas not been formulated, and it may yet proveto be as devastating as the previous wave ofcriticism was to older evidence and formula-tions. The evidence on problem-solving pro-cesses is anecdotal, except for one series of ex-periments employing a single type of problem.That particular problem, moreover, was aspatial one, and subjects may find special dif-ficulty in reporting verbally about spatial rea-soning. The Latane and Darley findings areimpressive, but they deal with awareness ofonly a single type of response. Moreover, sub-jects and even observers may be highly moti-vated to deny the role of such a trivial factor asthe presence of others in such an importantethical domain as the rendering of help to an-other human in distress.

In order to fill in the gaps in the literature,we have performed a series of small studies in-vestigating people's ability to report accuratelyon the effects of stimuli on their responses.

They were designed with several criteria inmind:

1. The cognitive processes studied were of aroutine sort that occur frequently in daily life.Deception was used minimally, and in only afew of the studies.

2. Studies were designed to sample a widerange of behavioral domains, including evalu-ations, judgments, choices, and predictions.

3. Care was taken to establish that subjectswere thoroughly cognizant of the existence ofboth the Critical stimulus and their ownresponses.

4. With two exceptions, the critical stimuliwere verbal in nature, thus reducing the pos-sibility that subjects could be cognizant of therole of the critical stimulus but simply unableto describe it verbally.

5. Most of the stimulus situations were de-signed to be as little ego-involving as possible sothat subjects would not be motivated ongrounds of social desirability or self-esteemmaintenance to assert or deny the role of par-ticular stimuli in influencing their responses.

The reader is entitled to know that the stim-ulus situations were chosen in large part be-cause we felt that subjects would be wrongabout the effects of the stimuli on their re-sponses. We deliberately attempted to studysituations where we felt that a particular stim-ulus would exert an influence on subjects' re-sponses but that subjects would be unable todetect it, and situations where we felt a parti-cular stimulus would be ineffective but subjectswould believe it to have been influential. It iseven more important to note, however, thatwe were highly unsuccessful in this attemptedbias. In general, we were no more accurate inour predictions about stimulus effects than thesubjects proved to be in their reports aboutstimulus effects. Most of the stimuli that weexpected to influence subjects' responses turnedout to have no effect, and many of the stimulithat we expected to have no effect turned outto be influential.

In all of the studies, some component of acomplex stimulus situation was manipulatedand the impact of this stimulus component onresponses could thus be assessed. Subjects, asit turned out, were virtually never accurate intheir reports. If the stimulus component had a


significant effect on responses, subjects typic-ally reported that it was noninfluential; if thestimulus component had no significant effect,subjects typically reported that it had beeninfluential.

Failure to Report the Influence ofEffective Stimulus Factors

Erroneous Reports about Stimuli InfluencingAssociative Behavior

The phenomenon of verbal associationseemed a fruitful one for illustrating an inabilityto report accurately about the role of influentialstimuli. For example, it seems likely that simul-taneous associative behavior—when two peoplespeak the same thought or begin humming thesame tune at the same time—may occur be-cause of the presence of some stimulus whichsets off identical associative processes in thetwo people. Then, because these associativeprocesses are hidden from conscious view, bothparties are mystified about the occurrence ofthe "coincidental" mutual behavior.

In order to test subject ability to report in-fluences on their associative behavior, we had81 male introductory psychology studentsmemorize a list of word pairs. Some of theseword pairs were intended to generate associa-tive processes that would elicit certain targetwords in a word association task to be per-formed at a later point in the experiment. Forexample, subjects memorized the word pair"ocean-moon" with the expectation that whenthey were later asked to name a detergent, theywould be more likely to give the target "Tide"than would subjects who had not previouslybeen exposed to the word pairs. In all, eightword pair cues were employed, and all eight didin fact have the effect of increasing the prob-ability of target responses in the word associa-tion task. The average effect of the semanticcuing was to double the frequency of targetresponses, from 10% to 20% (p < .001). Im-mediately following the word association task,subjects were asked in open-ended form whythey thought they had given each of their re-sponses in the word association task. Despitethe fact that nearly all subjects could recallnearly all of the words pairs, subjects almostnever mentioned a word pair cue as a reason for

giving a particular target response. Instead,subjects focused on some distinctive feature ofthe target ("Tide is the best-known deter-gent"), some personal meaning of it ("Mymother uses Tide"), or an affective reaction toit ("I like the Tide box"). When specificallyasked about any possible effect of the wordcues, approximately a third of the subjects didsay that the words had probably had an effect,but there is reason to doubt that these reportswere indications of any true awareness. An"awareness ratio" was calculated for each tar-get word. This was the number of subjects whoreported an influence of the cues divided by thenumber of subjects who were influenced by thecues. This latter number was an estimate,based on the number of cued subjects who gavethe target response minus the number of un-cued subjects who gave the target response.These awareness ratios for the eight targetwords ranged from 0 to 244%. This means thatfor some of the target words, none of the sub-jects reported any influence of the word cues,and for others, many more subjects reported aninfluence than were probably influenced.

Erroneous Reports about Position Effects onAppraisal and Choice

We conducted two studies that serendipit-ously showed a position effect on evaluation ofan array of consumer goods. (We had at-tempted, unsuccessfully, to manipulate thesmell of garments in the array.) In both studies,conducted in commercial establishments underthe guise of a consumer survey, passersby wereinvited to evaluate articles of clothing—fourdifferent nightgowns in one study (378 sub-jects) and four identical pairs of nylon stock-ings in the other (52 subjects). Subjects wereasked to say which article of clothing was thebest quality and, when they announced achoice, were asked why they had chosen thearticle they had. There was a pronounced left-to-right position effect, such that the right-most object in the array was heavily over-chosen. For the stockings, the effect was quitelarge, with the right-most stockings being pre-ferred over the left-most by a factor of almostfour to one. When asked about the reasons fortheir choices, no subject ever mentioned spon-taneously the position of the article in the


array. And, when asked directly about a pos-sible effect of the position of the article, virtu-ally all subjects denied it, usually with aworried glance at the interviewer suggestingthat they felt either that they had misunder-stood the question or were dealing with a mad-man.

Precisely why the position effect occurs isnot obvious. It is possible that subjects carriedinto the judgment task the consumer's habitof "shopping around," holding off on choice ofearly-seen garments on the left in favor of later-seen garments on the right.

Erroneous Reports about Anchoring Ejectson Predictions

In an unpublished study by E. Borgida, R.Nisbett, and A. Tversky, subjects (60 intro-ductory psychology students) were asked toguess what the average behavior of Universityof Michigan students would be in three differ-ent experiments. Some subjects were given an"anchor" in the form of knowledge about thebehavior of a particular "randomly chosen sub-ject." Of the anchor subjects, some were givenonly information about the individual's be-havior, while others were also shown a briefvideotaped interview with the individual. Ithad been anticipated that the videotape wouldincrease the salience and vividness of the an-chor and that subjects who were exposed to itwould show a greater anchoring effect, that is,that their estimates of the average behavior ofthe sample would cluster more closely aboutthe anchor value. Only weak support for theprediction was found, and anchoring effectsacross the experiments described to subjectsranged from huge and highly statistically sig-nificant ones down to actual "anti-anchoring"effects (i.e., somewhat greater variance of es-timates of the sample average for anchor con-ditions than for the no-anchor condition). Thisrange of effects, however, made possible a testof subjects' ability to report on their utilizationof the anchor value. Immediately after makingtheir estimates of average sample behavior,subjects were asked about the extent to whichthey had relied on knowledge about the par-ticular individual's behavior in making theseestimates. Subjects reported moderate utiliza-tion of the anchor value in all conditions. Thus

they reported the same degree of utilization ofthe anchor value for experiments where it hadnot been used at all as they did for experimentswhere it had heavily influenced their estimates.

Erroneous Reports about the Influence of anIndividual's Personality on Reactions to hisPhysical Characteristics

Perhaps the most remarkable of the demon-strations is one we have described in detailelsewhere (Nisbett & Wilson, in press). Thisstudy, an experimental demonstration of thehalo effect, showed that the manipulatedwarmth or coldness of an individual's person-ality had a large effect on ratings of the at-tractiveness of his appearance, speech, andmannerisms, yet many subjects actually in-sisted that cause and effect ran in the oppositedirection. They asserted that their feelingsabout the individual's appearance, speech, andmannerisms had influenced their liking of him.

Subjects were shown an interview with a col-lege teacher who spoke English with a Euro-pean accent. The interview dealt with teachingpractices and philosophy of education. Half thesubjects saw the teacher answering the ques-tions in a pleasant, agreeable, and enthusiasticway (warm condition). The other half saw anautocratic martinet, rigid, intolerant, and dis-trustful of his students (cold condition). Sub-jects then rated the teacher's likability andrated also three attributes that were by theirnature essentially invariant across the two ex-perimental conditions: his physical appearance,his mannerisms, and his accent. Subjects whosaw the warm version of the interview liked theteacher much better than subjects who saw thecold version of the interview, and there was avery marked halo effect. Most of the subjectswho saw the warm version rated the teacher'sappearance, mannerisms, and accent as at-tractive, while a majority of subjects who sawthe cold version rated these qualities as irritat-ing. Each of these differences was significantat the .001 level.

Some subjects in each condition were askedif their liking for the teacher had influencedtheir ratings of the three attributes, and somewere asked if their liking for each of the threeattributes had influenced their liking of theteacher. Subjects in both warm and cold condi-


tions strongly denied any effect of their overallliking for the teacher on ratings of his attri-butes. Subjects who saw the warm version alsodenied that their liking of his attributes had in-fluenced their overall liking. But subjects whosaw the cold version asserted that their dislik-ing of each of the three attributes had loweredtheir overall liking for him. Thus it would ap-pear that these subjects precisely inverted thetrue causal relationship. Their disliking of theteacher lowered their evaluation of his appear-ance, his mannerisms, and accent, but subjectsdenied such an influence and asserted insteadthat their dislike of these attributes had de-creased their liking of him!

Reporting the Influence of IneffectiveStimulus Factors

Three of our demonstrations involved themanipulation of stimulus factors that turnedout to have no effect on subjects' judgments.In each of these studies, subjects reported thatat least some of these actually ineffective fac-tors had been highly influential in theirjudgments.

Erroneous Reports about the Emotional Impactof Literary Passages

In the first of these studies, 152 subjects (in-troductory psychology students) read a selec-tion from the novel Rabbit, Run by John Up-dike. The selection described an alcoholichousewife who has just been left by her hus-band and who is cleaning up her filthy home inpreparation for a visit by her mother. Whiledrunkenly washing her infant girl, she acci-dentially allows the child to drown. The selec-tion is well written and has a substantial emo-tional impact even when read out of the contextof the rest of the novel. There were four con-ditions of the experiment. In one condition,subjects read the selection as it was written. Ina second condition, a passage graphically de-scribing the messiness of the baby's crib wasdeleted. In a third condition, subjects read theselection minus a passage physically describingthe baby girl. In the fourth condition, bothpassages were deleted.

After reading the selection, all subjects wereasked what emotional impact it had had. Then

the manipulated passages were presented, andsubjects were asked how the presence of thepassage had affected (or would have affected,for subjects for whom the passage was deleted),the emotional impact of the selection. As itturned out, there was no detectable effect onreported emotional impact due to inclusionversus deletion of either passage. (Both pairsof means differed by less than .10 on a 7-pointscale.) Subjects reported, however, that thepassages had increased the impact of the selec-tion. Subjects exposed to the passage describingthe messiness of the baby's crib were virtuallyunanimous in their opinion that the passagehad increased the impact of the selection: 86%said the passage had increased the impact. Twothirds of the subjects exposed to the physicaldescription of the baby reported that the pas-sage had had an effect, and of those who re-ported it had an effect, two and a half times asmany subjects said it had increased the impactof the selection as said it decreased the impact.The subjects who were not exposed to the pas-sages on the initial reading predicted that bothpassages would have increased the impact ofthe selection had they been included. Predictedeffects by these subjects were in fact extremelyclose to the pattern of (erroneous) reportedeffects by subjects who were exposed to thepassages.

Erroneous Reports about the Effects ofDistractions on Reactions to a Film

In another study, 90 subjects (introductorypsychology students) were asked to view abrief documentary on the plight of the Jewishpoor in large cities. Some subjects viewed thefilm while a distracting noise (produced by apower saw) occurred in the hall outside. Othersubjects viewed the film while the focus waspoorly adjusted on the projector. Control sub-jects viewed the film under conditions of no dis-traction. After viewing the film, subjects ratedit on three dimensions—how interesting theythought it was, how much they thought otherpeople would be affected by it, and how sym-pathetic they found the main character to be.Then, for experimental conditions, the experi-menter apologized for the poor viewing con-ditions and asked subjects to indicate next toeach rating whether he had been influenced by


the noise or poor focus. Neither the noise northe poor focus actually had any detectableeffect on any of the three ratings. (Ratings werein general trivially higher for distraction sub-jects.) In the first and only demonstration ofreasonably good accuracy in subject report ofstimulus effects we found, most of the subjectsin the poor focus condition actually reportedthat the focus had not affected their ratings(although 27% of the subjects reported thatthe focus had lowered at least one rating, a pro-portion significantly different from zero). Amajority of subjects in the noise condition,however, erroneously reported that the noisehad affected their ratings. Fifty-five per cent ofthese subjects reported that the noise had low-ered at least one of their ratings.

Erroneous Reports about the Effects ofReassurance on Willingness to Take ElectricShocks

In a third study, 75 subjects (male intro-ductory psychology students) were asked topredict how much shock they would take in anexperiment on the effects of intense electricshocks. One version of the procedural protocolfor the experiment included a "reassurance"that the shocks would do "no permanentdamage." The other version did not includethis "reassurance." Subjects receiving the firstversion were asked if the phrase about perman-ent damage had affected their predictions aboutthe amount of shock they would take, and sub-jects receiving the second version were asked ifthe phrase would have affected their predic-tions, had it been included. Inclusion of thephrase in fact had no effect on predicted shocktaking, but a majority of subjects reported thatit did. Of those reporting an effect, more than80% reported it had increased their predictions.Subjects who had not received the phrase weresimilarly, and erroneously, inclined to say thatit would have increased their willingness totake shock had it been included.

Taken together, these studies indicate thatthe accuracy of subject reports about higherorder mental processes may be very low. Wewish to acknowledge that there are method-ological and interpretive problems with someof the individual studies, however. Althoughthe magnitude of effects induced by effective

critical stimuli ranged from a ratio of 2:1 overcontrol values to a ratio of 4:1, the criticalstimuli may often have been merely necessaryand not sufficient causes of the responses inquestion. Therefore subjects may often havebeen correct in asserting that some other stim-ulus was a more important determinant of theirresponses. In studies where the manipulatedstimuli were ineffective (e.g., the literary pas-sage and distraction studies), it is conceivablethat perceived experimenter demands couldhave contributed to the results. And finally, insome of the studies it could be argued that thesubjects denied the role of the influentialstimulus in order to avoid looking silly or fool-ish (e.g., the position effect study), and not be-cause they were unaware of its causal role.

We also wish to acknowledge that the studiesdo not suffice to show that people could never beaccurate about the processes involved. To doso would require ecologically meaningless buttheoretically interesting procedures such asinterrupting a process at the very moment itwas occurring, alerting subjects to pay carefulattention to their cognitive processes, coachingthem in introspective procedures, and so on.What the studies do indicate is that such intro-spective access as may exist is not sufficient toproduce accurate reports about the role ofcritical stimuli in response to questions askeda few minutes or seconds after the stimuli havebeen processed and a response produced.

The Origin of Verbal Reports AboutCognitive Processes

The Fount That Never Was

In summary, it would appear that peoplemay have little ability to report accurately ontheir cognitive processes:

1. Sometimes, as in many dissonance andattribution studies, people are unable to reportcorrectly even about the existence of the evalu-ative and motivational responses produced bythe manipulations.

2. Sometimes, as in dissonance and attribu-tion studies, and in the reports of creativeartists and scientists, people appear to be un-able to report that a cognitive process hasoccurred.


3. Sometimes, as in the subliminal percep-tion literature and the reports of creativeworkers, people may not be able to identify theexistence of the critical stimulus.

4. Even when people are completely cogni-zant of the existence of both stimulus and re-sponse, they appear to be unable to report cor-rectly about the effect of the stimulus on theresponse. This is true in dissonance and attribu-tion studies, in the subliminal perception liter-ature, in the reports of creative workers, and inthe work by Maier (1931), Latane and Barley(1970), and in our own studies described above.

In addition, we might point out that at leastsome psychological phenomena probably wouldnot occur in the first place if people were awareof the influence of certain critical stimuli. Forexample, if people were aware of the effects ofthe presence of other people on their tendencyto offer help to a person in distress, they wouldsurely strive to counteract that influence, andwould therefore not show the typical effect.Similarly, if people were aware of positioneffects on their evaluations, they would attemptto overcome these effects. A number of otherphenomena would seem to depend on lack ofawareness of the role played by certain criticalfactors, for example, halo effects, contrast ef-fects, and order effects. If people knew thattheir judgments were subject to influence fromother judgments made about an object or fromjudgments just previously made about otherobjects, or from the order in which the objectwas examined, then they would correct for suchinfluences and these effects would not exist.

Polanyi (1964) and others (e.g., Gross, 1974)have argued persuasively that "we can knowmore than we can tell," by which it is meantthat people can perform skilled activities with-out being able to describe what they are doingand can make fine discriminations without be-ing able to articulate their basis. The researchdescribed above suggests that the converse isalso true—that we sometimes tell more thanwe can know. More formally, people sometimesmake assertions about mental events to whichthey may have no access and these assertionsmay bear little resemblance to the actualevents.

The evidence reviewed is then consistentwith the most pessimistic view concerning

people's ability to report accurately abouttheir cognitive processes. Though methodolog-ical implications are not our chief concern, weshould note that the evidence indicates it maybe quite misleading for social scientists to asktheir subjects about the influences on theirevaluations, choices, or behavior. The relevantresearch indicates that such reports, as well aspredictions, may have little value except forwhatever utility they may have in the study ofverbal explanations per se.

More importantly, the evidence suggeststhat people's erroneous reports about their cog-nitive processes are not capricious orhaphazard, but instead are regular and syste-matic. Evidence for this comes from the factthat "observer" subjects, who did not partici-pate in experiments but who simply read ver-bal descriptions of them, made predictionsabout the stimuli which were remarkably simi-lar to the reports about the stimuli by subjectswho had actually been exposed to them. Inexperiments by Latane and Darley (1970), andin several of our own studies, subjects wereasked to predict how they themselves, or howother people, would react to the stimulus situa-tions that had actually been presented to othersubjects. The observer subjects made predic-tions that in every case were similar to the er-roneous reports given by the actual subjects.Thus Latane and Barley's original subjectsdenied that the presence of other people hadaffected their behavior, and observer subjectsalso denied that the presence of others wouldaffect either their own or other people's be-havior. When our word association study wasdescribed to observer subjects, the judgmentsof the probability that particular word cueswould affect particular target responses werepositively correlated with the original subjects'"introspective reports" of the effects of theword cues on the target responses. (Both sub-ject reports and observer predictions wereslightly negatively correlated with true cuingeffects.) In two of our other studies, subjectswere asked to predict how they would have re-sponded to stimuli that were actually presentedto subjects in another condition. In both cases,predictions about behavior were very similarto the inaccurate reports of subjects who hadactually been exposed to the conditions. Thus,whatever capacity for introspection exists, it


does not produce accurate reports about stim-ulus effects, nor does it even produce reportsthat differ from predictions of observers operat-ing only with a verbal description of the stimu-lus situation. As Bern (1967) put it in a similarcontext, if the reports of subjects do not differfrom the reports of observers, then it is unnec-cessary to assume that the former are drawingon "a fount of privileged knowledge" (p. 186).It seems equally clear that subjects and ob-servers are drawing on a similar source for theirverbal reports about stimulus effects. Whatmight this be?

A Priori Causal Theories

We propose that when people are asked toreport how a particular stimulus influenced aparticular response, they do so not by consult-ing a memory of the mediating process, but byapplying or generating causal theories aboutthe effects of that type of stimulus on that typeof response. They simply make judgments, inother words, about how plausible it is that thestimulus would have influenced the response.These plausibility judgments exist prior to, orat least independently of, any actual contactwith the particular stimulus embedded in aparticular complex stimulus configuration.Causal theories may have any of severalorigins.

1. The culture or a subculture may have ex-plicit rules stating the relationship between aparticular stimulus and a particular response("I came to a stop because the light started tochange." "I played a trump because I had nocards in the suit that was led").

2. The culture or a subculture may supplyimplicit theories about causal relations. InAbelson's (1968) terms, the presence of a par-ticular stimulus may "psychologically imply"a particular response ("Jim gave flowers toAmy [me]; that's why she's [I'm] actingpleased as punch today"). In Kelley's (1972)terms, people growing up in a given culturelearn certain "causal schemata," psychologicalrules governing likely stimulus-response rela-tions ("The ballplayer [I] was paid to endorseAqua-Velva, that's the only reason he [I]endorsed it").

3. An individual may hold a particularcausal theory on the basis of empirical observa-

tion of covariation between stimuli of the gen-eral type and responses of the general type.("I'm grouchy today. I'm always grouchywhen I don't break 100 in golf.") There is rea-son to suspect, however, that actual covariationmay play less of a role in perceived or reportedcovariation than do theories about covariation.The Chapmans (Chapman, 1967; Chapman &Chapman, 1967, 1969) have shown that pow-erful covariations may go undetected when theindividual lacks a theory leading him to suspectcovariation and, conversely, that the individualmay perceive covariation where there is noneif he has a theory leading him to expect it. Thepresent position, of course, leads to the expecta-tion that people would be as subject to theory-induced errors in self-perception as in the per-ception of convariation among purely externalevents.

4. In the absence of a culturally suppliedrule, implicit causal theory, or assumptionabout covariation, people may be able to gen-erate causal hypotheses linking even novelstimuli and novel responses. They may do soby searching their networks of connotative re-lations surrounding the stimulus descriptionand the response description. If the stimulus isconnotatively similar to the response, then itmay be reported as having influenced the re-sponse. To the extent that people share similarconnotative networks they would be expectedto arrive at similar judgments about the like-lihood of a causal link between stimulus andresponse.

We do not wish to imply that all or evenmost a priori causal theories are wrong. Verbalreports relying on such theories will typicallybe wrong not because the theories are in errorin every case but merely because they are incor-rectly applied in the particular instance.

The tools that people employ when asked tomake judgments about causality are analogousto the "representativeness heuristic" describedby Tversky and Kahneman (1973, 1974;Kahneman & Tversky, 1973). These writershave proposed that when making judgmentsabout the probability that an individual is, say,a librarian, one does so by comparing his in-formation about the individual with the con-tents of his stereotype concerning librarians. Ifthe information is representative of the con-


tents of the stereotype concerning librarians,then it is deemed "probable" that the individ-ual is a librarian. Information that is morepertinent to a true probability judgment, suchas the proportion of librarians in the popula-tion, is ignored. We are proposing that a similarsort of representativeness heuristic is employedin assessing cause and effect relations in self-perception. Thus a particular stimulus will bedeemed a representative cause if the stimulusand response are linked via a rule, an implicittheory, a presumed empirical covariation, oroverlapping connotative networks.

In the experiments reviewed above, then,subjects may have been making simple repre-sentativeness judgments when asked to intro-spect about their cognitive processes. Worryand concern seem to be representative, plaus-ible reasons for insomnia while thoughts aboutthe physiological effects of pills do not. Seeinga weight tied to a string seems representative ofthe reasons for solving a problem that requirestying a weight to a cord, while simply seeingthe cord put into motion does not. The plightof a victim and one's own ability to help himseem representative of reasons for intervening,while the sheer number of other people presentdoes not. The familiarity of a detergent andone's experience with it seem representative ofreasons for its coming to mind in a free associ-ation task, while word pairs memorized in averbal learning experiment do not. The knit,sheerness, and weave of nylon stockings seemrepresentative of reasons for liking them, whiletheir position on a table does not. And a reas-surance that electric shock will cause no per-manent damage seems representative of reasonsfor accepting shock; reading about the behaviorof a particular experimental subject (the "an-chor" value) seems representative of the rea-sons for choosing a similar behavior as theaverage value for the subject population as awhole; a passage graphically describing thephysical characteristics of a child seems repre-sentative of reasons for being emotionally af-fected by a literary selection ending with thedeath of a child, and a distracting noise seemsrepresentative of reasons for not liking a film.

When subjects were asked about their cog-nitive processes, therefore, they did somethingthat may have felt like introspection but whichin fact may have been only a simple judgment

of the extent to which input was a representa-tive or plausible cause of output. It seemslikely, in fact, that the subjects in the presentstudies, and ordinary people in their daily lives,do not even attempt to interrogate their mem-ories about their cognitive processes when theyare asked questions about them. Rather, theymay resort in the first instance to a pool ofculturally supplied explanations for behaviorof the sort in question or, failing in that, begina search through a network of connotative re-lations until they find an explanation that maybe adduced as psychologically implying the be-havior. Thus if we ask another person why heenjoyed a particular party and he respondswith "I liked the people at the party," we maybe extremely dubious as to whether he hasreached this conclusion as a result of anythingthat might be called introspection. We are justi-fied in suspecting that he has instead askedhimself Why People Enjoy Parties and hascome up with the altogether plausible hypo-thesis that in general people will like parties ifthey like the people at the parties. Then, hisonly excursion into his storehouse of private in-formation would be to make a quick check toverify that his six worst enemies were not at theparty. If not, he confidently asserts that thepeople-liking was the basis of his party-liking.He is informationally superior to observers, inthis account, only by virtue of being able tomake this last-minute check of his enemies list,and not by virtue of any ability to examine di-rectly the effects of the stimuli (the people) onhis response (enjoyment).

The present view carries two important im-plications that go beyond a merely anti-intro-spectivist position: (a) People's reports willsometimes be correct, and it should be possibleto predict when they will be likely to be cor-rect, (b) People's reports about their highermental processes should be neither more norless accurate, in general, then the predictionsabout such processes made by observers. Anexperiment by Nisbett and Bellows (Note 3),reported below, tested both these implications.

Accuracy of Subject Reports andObserver Predictions

The above analysis implies that it should bepossible to demonstrate accuracy and inaccu-


racy in verbal reports in the same experimentby simply asking subjects to make two sorts ofjudgments—those for which the influentialfactors are plausible and are included in a pri-ori causal theories, and others which are in-fluenced by implausible factors not included insuch theories. In the former case, both subjectsand observers should be accurate; in the latter,neither should be accurate.

Nisbett and Bellows (Note 3) asked femalesubjects to read a lengthy description of awoman who was applying for a job as acounselor in a crisis intervention center. Sub-jects read what they believed was the applica-tion portfolio, a lengthy document including aletter of recommendation and a detailed reportof an interview with the center's director. Fivestimulus factors were manipulated, (a) Thewoman's appearance was either described insuch a way as to make it clear that she wasquite physically attractive, or nothing was saidabout her appearance, (b) The woman waseither described as having superb academiccredentials, or nothing was said about her aca-demic credentials, (c) The woman was de-scribed as having spilled a cup of coffee overthe interviewer's desk, or nothing was saidabout any such incident, (d) The woman wasdescribed as having been in a serious auto ac-cident, or nothing was said about an accident.(e) Subjects were either told that they wouldmeet the woman whose folder they were read-ing, or they were told that they would meetsome other applicant. These stimuli weremanipulated factorially.

After reading the portfolio, subjects wereasked to make four judgments about thewoman: (a) how much they liked her, (b) howsympathetic they thought she would be towardclients' problems, (c) how intelligent theythought she was, and (d) how flexible theythought she would be in dealing with clients'problems. Then subjects were asked how eachof the factors (ranging from 0 for some subjectsup to 5 for others) had influenced each of thefour judgments.

In addition, "observer" subjects were askedto state how each of the five factors would in-fluence each of the four judgments. These sub-jects did not read any portfolio and, indeed, thefactors were described only in summary form(e.g., "Suppose you knew that someone was

quite physically attractive. How would that in-fluence how much you would like the person?").Both observers and subjects answered thesequestions on 7-point scales ranging from "in-crease^) my liking a great deal" to "de-crease (d) my liking a great deal."

Two predictions about the results of thestudy follow from the present analysis.

1. Subjects should be much more accurate intheir reports about the effects of the stimulusfactors on the intelligence judgment than intheir reports about the effects of the factors ontheir other judgments. This is because the cul-ture specifies more clearly what sorts of factorsought to influence a judgment of intelligence,and in what way they should do so, than it doesfor judgments such as liking, sympathy towardothers, or flexibility. In fact, the other factor-judgment combinations were chosen withmalice aforethought. Recent work by socialpsychologists has shown that several of thefactors have implausible effects on several ofthe judgments, for example, people tend togive more favorable ratings on a number of di-mensions to people whom they believe they areabout to meet than to people whom they do notexpect to meet (Barley & Berschied, 1967).

2. Whether subjects are generally accuratein reports about the effects of the factors on agiven judgment or generally inaccurate, theiraccuracy will be equalled by observers workingfrom impoverished descriptions of the factors.

The results gave the strongest possible sup-port to both predictions. Mean subject reportsabout the effects of the factors, mean observerreports, and mean actual effects (experimentalminus control means) were compared for eachof the judgments. The most remarkable resultwas that subject and observer reports of factorutilization were so strongly correlated for eachof the judgments that it seems highly unlikelythat subjects and observers could possibly havearrived at these reports by different means.Mean subject and observer reports of factorutilization were correlated .89 for the likingjudgment, .84 for the sympathy judgment, .99for the intelligence judgment, and .77 for theflexibility judgment. Such strong correspond-ence between subject and observer reports sug-gests that both groups produced these reports


via the same route, namely by applying orgenerating similar causal theories.

As anticipated, subject accuracy was ex-tremely high for the intelligence judgment.Subject reports about the effects of the factorswere correlated .94 with true effects of thefactors. Also as anticipated, however, observerpredictions were fully as accurate as subject re-ports : Observer predictions were correlated .98with true effects of the factors on the intelli-gence judgment.

For the other judgments, the accuracy ofsubject reports was literally nil. Subject reportswere correlated —.31 with true effects on theliking judgment, .14 with true effects on thesympathy judgment, and .11 with true effectson the flexibility judgment. Once again, ob-servers were neither more nor less accurate thansubjects. Correlations of their predictions withtrue effects were highly similar to the correla-tions of subject reports with true effects.

It should be noted that the experiment pro-vides good justification for requiring a changein the traditional empirical definition of aware-ness. "Awareness" has been equated with "cor-rect verbal report." The Nisbett and Bellowsexperiment and the present analysis stronglysuggest that this definition is misleading andovergenerous. The criterion for "awareness"should be instead "verbal report which exceedsin accuracy that obtained from observers pro-vided with a general description of the stimu-lus and response in question." Even highly ac-curate reports, therefore, provide no evidenceof introspective awareness of the effects of thestimuli on responses if observers can equal thatlevel of accuracy.

Accuracy and Inaccuracy in VerbalExplanations

When Will We Be Wrong In OurVerbal Reports?

It is possible to speculate further about thecircumstances that should promote accuracy inreports about higher mental processes and thosethat should impair accuracy. We will need tocall on another Tversky and Kahneman (1973)concept to help describe these circumstances.These writers proposed that a chief determin-ant of judgments about the frequency and

probability of events is the availability in mem-ory of the events at the time of judgment.Events are judged as frequent in proportion totheir availability, and their availability is de-termined by such factors as the salience of theevents at the time they were encountered, thestrength of the network of verbal associationsthat spontaneously call the events to mind, andinstructional manipulations designed to makethe events more salient at the time of judgment.

The representativeness and availabilityheuristics are undoubtedly intertwined in theappraisal of cause and effect relations. If a par-ticular stimulus is not available, then it willnot be adduced in explanation of a given effect,even though it might be highly representativeor plausible once called to mind. Similarly, therepresentativeness heuristic may be a chiefdeterminant of availability in cause-effectanalysis: A particular stimulus may be avail-able chiefly because it is a highly representativecause of the effect to be explained.

It is possible to describe many circumstancesthat would serve to reduce the availability of agiven causal candidate that is in fact influen-tial, or to enhance the availability of a causalcandidate that is in fact noninfluential. Simi-larly, influential causes will sometimes benonrepresentative of the effects they produce,and noninfluential factors nevertheless will behighly representative causes. Any of these cir-cumstances should promote error in verbalreports.

Removal in time. Perhaps chief among thecircumstances that should decrease accuracy inself-report is a separation in time between thereport and the actual occurrence of the process.In almost all the research described above, sub-jects were asked about a cognitive process im-mediately after its occurrence, often withinseconds of its occurrence. While the presentviewpoint holds that there may be no directaccess to process even under these circum-stances, it is at least the case that subjects areoften cognizant of the existence of the effectivestimuli at this point. Thus subjects have somechance of accurately reporting that a particularstimulus was influential if it happens to seemto be a plausible cause of the outcome. At somelater point, the existence of the stimulus may beforgotten, or become less available, and thusthere would be little chance that it could be


correctly identified as influential. Similarly, thevagaries of memory may allow the invention offactors presumed to be present at the time theprocess occurred. It is likely that such inventedfactors would be generated by use of causaltheories. Thus it would be expected that themore removed in time the report is from theprocess, the more stereotypical should be thereported explanation.

Mechanics of judgment.1 There is a class ofinfluential factors to which we should be par-ticularly blind. That class may be described asthe mechanics of judgment factors—for ex-ample, serial order effects, position effects, con-trast effects, and many types of anchoring ef-fects. Such factors should seem particularlyimplausible as reasons for liking or disliking anobject, or for estimating its magnitude on somedimension as high or low. Indeed, it seems out-rageous that such a judgment as one concerningthe quality of a nightgown might be affected byits position in a series, or that the estimation ofthe size of an object should be affected by thesize of a similar object examined just previ-ously.

Context. Generally, it should be the case thatwe will be blind to contextual factors, or at anyrate be particularly poor at disentangling theeffects of the stimulus from the context inwhich it was encountered. Contextual cues arenot likely to be spontaneously salient when weare asked, or ask ourselves, why we evaluatedan object as we did. Any question about an ob-ject is likely to focus our attention on the prop-erties of the object itself and to cause us toignore contextual cues. When a question aboutcontext is asked directly, on the other hand, aswhen we questioned our subjects about the ef-fects of noise on their reactions to a film, con-textual factors might well be reported as in-fluential even when they are not. Unlikemechanics of judgment factors, many contextfactors, once they are made available, shouldseem highly plausible causes.

Nonevents. Ross (in press) has pointed outthat many judgments and evaluations probablyare based at least in part on the nonoccurrenceof certain events. Thus one person may cor-rectly perceive that another person does notlike him, and this perception may be basedlargely on the nonoccurrence of friendly be-haviors rather than on the outright manifesta-

tion of hostility. There is good reason to suspectthat nonbehavior will be generally less avail-able and salient than behavior, and thereforeshould rarely be reported as influential. Butthe effect will still require explanation, andthus noninfiuential events will often be invokedin preference to influential nonevents.

Nonverbal behavior. In evaluating other peo-ple, we probably rely heavily on nonverbal cuessuch as posture, distance, gaze, and the volumeand tone of voice (Argyle, Solter, Nicholson,Williams, & Burgess, 1970). Yet it seems likelythat such nonverbal cues would be less avail-able than verbal behavior, if only becauseverbal labels for nonverbal behavior are fewand impoverished. To the extent that we relyon verbal memory to explain our evaluations ofother people there will be proportionately moreverbal behaviors to serve as causal candidatesthan nonverbal behaviors. To the extent thatnonverbal behaviors are important to evalua-tions, relative to verbal behaviors, they will bewrongly overlooked.

Discrepancy between the magnitudes of causeand effect. In general, we would expect thatfactors will be perceived as causal to the degreethat their magnitudes resemble the magnitudeof the effects they are adduced to explain. Inthe development of causal schemata, both thenotion that large causes can produce large ef-fects and the notion that small causes can pro-duce small effects probably precede the de-velopment of the notion that large causes canproduce small effects. The notion that smallcauses can produce large effects probably de-velops very late and never attains very greatstability. It is likely that conspiracy theoriesoften feed on the discrepancy between offici-ally provided causal explanations and the largeeffects they are invoked to explain. It is out-rageous that a single, pathetic, weak figure likeLee Harvey Oswald should alter world history.When confronted with large effects, it is tocomparably large causes that we turn for ex-planations. Thus when Storms and Nisbett(1970) interviewed insomniacs and asked themwhy they slept so little, both on particular oc-casions and in general, they were inclined toexplain their insomnia in terms of the stress of

1 We are indebted to Amos Tversky for this idea.


their current life situation or even in terms ofneurosis or chronic anxiety. Smaller causes,such as an overheated room, a tendency to workor exercise or smoke just before going to bed, ora tendency to keep irregular hours, were over-looked. Many judgments of the plausibility ofcause and effect relations are probably basedat least in part on the fittingness of cause andeffect magnitudes. Thus both mechanics ofjudgment factors and nonevents should oftenbe perceived as implausible causes simply be-cause of their smallness and seeming inconse-quentiality.

When Will We Be Correct In OurVerbal Reports?

The present analysis corresponds to commonsense in that it allows that we will often beright about the causes of our judgments and be-havior. If a stranger walks up to a person,strikes him, and walks away, and the person islater asked if he likes the stranger, he will replythat he does not and will accurately report thereason. The interaction he has had with thestranger will be highly salient and a highlyplausible reason for disliking someone. And, ingeneral, the conditions that promote accuracyin verbal report will be the opposite of those de-scribed previously. These conditions may besummarized briefly by saying that reports willbe accurate when influential stimuli are (a)available and (b) plausible causes of the re-sponse, and when (c) few or no plausible butnoninfluential factors are available.

There is, in fact, some evidence in the litera-ture that people can sometimes accurately re-port on the stimuli that influenced particularcognitive processes. Ironically, both of theareas where this had been systematically dem-onstrated have been developed by investigatorswho were seeking to show lack of awareness forthe cognitive processes concerned. These arethe literatures on (a) learning without aware-ness and (b) awareness of factors influencingcomplex judgments.

Learning Without (?) Awareness

Most of the literature concerning people'sability to report on the factors that influencelearning, or the increased emission of an oper-

ant, has focused on the Greenspoon (1955) orrelated phenomena. In this paradigm, subjectssay the names of words that come to mind, orgenerate sentences employing particular words.After a baseline, no-reinforcement period, sub-jects are systematically reinforced (by "uhhuhs" or "goods") for a particular class of re-sponses (e.g., plural nouns, or sentences em-ploying first-person pronouns). The reinforce-ment typically elicits an increased rate ofresponse for the reinforced class. Early investi-gators reported that subjects were unaware ofthis influence on their behavior. Later investi-gators (see e.g., Dulany, 1962; Erikson, 1962;Spielberger, 1962) insisted that subjects hadbeen inadequately questioned, and that exten-sive probing revealed that all subjects whoshowed learning were also aware of the experi-menter's reinforcements and the link betweenthese reinforcements and their own increasedoutput of reinforced responses.

Many writers have proposed that the sub-jects' "awareness" is due to nothing more thana Heisenberg-type effect. That is, the measure-ment procedure itself may suggest to the sub-ject a connection that was not apparent to himbefore. Be that as it may, the present analysismakes it clear that there is every reason to ex-pect that subjects in these experiments shouldbe able to accurately report about cause andeffect, (a) The response possibilities allowed thesubject are extremely constrained. He is per-mitted very little latitude in the sorts of be-havior he may emit, (b) The stimulus situationis even more fixed and static. In fact, virtuallythe only stimulus that occurs is the experi-menter's "uh huh" or "good." (c) Finally, thecausal connection between this critical stimu-lus or reinforcement and the increased frequencyof a particular response class should be a highlyplausible one.

It is thus hardly surprising that subjects re-port, or at least can be induced to report, aconnection between the experimenter's stimu-lus and their own responses. Devotees of learn-ing-without-awareness could scarcely have de-signed a paradigm more likely to result inaccurate verbal report if they had set out de-liberately to do so. There is some evidence, infact, that when even relatively minor steps aretaken to disguise the connection between stim-ulus and response, subjects will fail to report


such a connection. This comes from the so-called "double agent" study by Rosenfeld andBaer (1969) in which the subject believed him-self to be the experimenter and the response(for which he was reinforced by the "subject"confederate) was not so focal as for the tradi-tional Greenspoon subject. Under these cir-cumstances, subjects reported no awareness,even under extensive probing, of the connectionbetween the confederate's behavior and theirown.

The Correspondence Between Actual andSubjective Weights in Judgment Tasks

Slovic and Lichtenstein (1971) have re-viewed the literature concerning the ability ofsubjects to report accurately on the weightsthey assign to various stimulus factors in mak-ing evaluations. Most of the investigations ofthis question have employed either clinicalpsychologists or stockbrokers as subjects, andthe judgmental domain has been largely limitedto clinical diagnoses and assessments of thefinancial soundness of stocks. Subjects areasked to diagnose patients using MinnesotaMultiphasic Personality Inventory (MMPI)scores or to assess stocks using such indicatorsas growth potential and earnings ratio. Thensubjects are asked to state the degree of theirreliance on various factors. These subjectiveweights are then compared to the objectiveweights derived from regression of the subject'sjudgments on the various factors. Slovic andLichtenstein (1971) concluded that self-insightwas poor and that of the studies which allowedfor a comparison of perceived and actual cueutilization, "all found serious discrepancies be-tween subjective and objective relativeweights" (p. 684). While this is a fair assess-ment of this literature, what strikes one as im-pressive from the present vantage point is thatalmost all the studies reviewed by Slovic andLichtenstein (1971) found evidence of at leastsome correspondence between subjective andobjective weights. This is almost the sole evi-dence we have been able to uncover, outside thelearning-without-awareness literature, thatpeople can be at all accurate in reporting aboutthe effects of stimuli on their responses.

The present framework is useful in under-standing this lonely outcropping of accuracy.

Clinical psychologists and stockbrokers under-take a formal study of the decision processesthey should employ. They are taught explicitlyhow various factors should be weighed in theirevaluations. Thus, for example, elevation of theschizophrenia scale will seem to be a highlyplausible reason for a judgment of severe path-ology because this is an association that clini-cians are formally taught. It seems likely, infact, that clinicians and stockbrokers could as-sign accurate weights prior to making the seriesof judgments in these experiments simply bycalling on the stored rules about what suchjudgments should reflect. If so, one wouldscarcely want to say they were engaging in pro-spective introspection, but merely that theyremember well the formal rules of diagnosis orfinancial counseling they were taught.

And in general, we may say that people willbe accurate in reports about the causes of theirbehavior and evaluations wherever the culture,or a subculture, specifies clearly what stimulishould produce which responses, and especiallywhere there is continuing feedback from theculture or subculture concerning the extent towhich the individual is following the prescribedrules for input and output. Thus university ad-missions officials will be reasonably accurateabout the weights they assign to various typesof information in admissions folders, and automechanics will be reasonably accurate aboutthe weights they assign to various factors indeciding whether a car has ignition or carbure-tor troubles. But such accuracy cannot be re-garded as evidence of direct access to processesof evaluation. It is evidence for nothing morethan the ability to describe the formal rules ofevaluation.

The implication of this analysis is that thejudgment studies lack what might be called"causal theory controls." Subjects' reportedweights should not be compared directly totheir actual weights. Instead, investigatorsshould examine the increment in prediction ofactual weights that is obtained by asking sub-jects about their subjective weights over theprediction that is obtained by (a) asking sub-jects about their subjective weights prior totheir examination of the data set in question;or (b) asking subjects about their beliefs con-cerning the weights employed by the average,or ideal, or some particular, clinician or stock-


broker; or even (c) by simply asking subjectsabout the weights they were taught to employin such judgments.

In the present view, little or no "awareness"might be found in studies employing such con-trols. That is, the subject's actual weights mightbe as well predicted by his subjective weightsfor his neighboring stockbroker as by the sub-ject's reports about the weights he himself used.

Why are We Unaware of OurUnawareness?

There is of course a problem with any char-acterization of "introspection" as nothing morethan judgments of plausibility. It does not feellike that at all. While we may sometimes admitto confusion about how we solved a particularproblem or why we evaluated a person in agiven way, it is often the case that we feel asthough we have direct access to cognitive pro-cesses. We could retreat behind our data andassert that there is by now enough evidencediscrediting introspective reports to allow us toignore any argument based on introspection.But there is more that can be said than this.

It seems likely that there are regularitiesconcerning the conditions that give rise to in-trospective certainty about cognitive processes.Confidence should be high when the causalcandidates are (a) few in number, (b) perceptu-ally or memorially salient, (c) highly plausiblecauses of the given outcome (especially wherethe basis of plausibility is an explicit culturalrule), and (d) where the causes have been ob-served to be associated with the outcome in thepast. In fact, we appeal to introspection to sup-port this view. Does the reader feel there is any-thing beyond factors such as these that need beadduced to account for occasions of subjectivecertainty?

The above view, it should be noted, is emi-nently testable. It should be possible to showthat subjective certainty is great when causalcandidates are salient and highly plausible, butare in reality noninfluential. Subjective cer-tainty should be lower when the causal candi-dates are actually influential, but are not sali-ent, not plausible, or compete with more salientor plausible but noninfluential causal candi-dates.

There are several factors that may help to

sustain the illusion of introspective awareness.These are sketched below.

Confusion Between Content and Process

An important source of our belief in intro-spective awareness is undoubtedly related tothe fact that we do indeed have direct access toa great storehouse of private knowledge.Jones and Nisbett (1972) have enumerated alist of types of privately held knowledge thatbears repeating in the present context. Theindividual knows a host of personal historicalfacts; he knows the focus of his attention atany given point in time; he knows what hiscurrent sensations are and has what almost allpsychologists and philosophers would assertto be "knowledge" at least quantitativelysuperior to that of observers concerning hisemotions, evaluations, and plans. Given thatthe individual does possess a great deal of ac-curate knowledge and much additional "knowl-edge" that is at least superior to that of anyobserver, it becomes less surprising that peoplewould persist in believing that they have, inaddition, direct access to their own cognitiveprocesses. The only mystery is why people areso poor at telling the difference between privatefacts that can be known with near certaintyand mental processes to which there may be noaccess at all.

A related point is that we are often capableof describing intermediate results of a series ofmental operations in such a way as to promotethe feeling that we are describing the operationsthemselves. Thus, for example, it is undoubt-edly true that Maier's (1931) psychology pro-fessor subject had "imagery of monkeys swing-ing from trees." It is even conceivable that thatimagery preceded and even facilitated the finalsteps in the mental operations that resulted inthe cord-swinging solution. But is it scarcelyreasonable to propose that such imagery was theprocess by which the problem was solved. Asecond example of the confusion of intermedi-ate output with process was provided by anacquaintance of the authors' who was asked tointrospect about the process by which he hadjust retrieved from memory his mother'smaiden name. "I know just what the processwas," he said. "I first thought of my uncle'slast name, and since that happens to be my


mother's maiden name, I had the solution."This only pushes the process question back astep further, of course, and our acquaintance'sanswer would appear to reflect a confusion ofintermediate results with the process by whichthe final result was obtained.

It should be noted that the individual's pri-vate access to content will sometimes allowhim to be more accurate in his reports about thecauses of his behavior than an observer wouldbe. The occasions when the individual is corrector at least not provably wrong, and the ob-server is manifestly wrong, should serve tosustain the individual's sense of privileged ac-cess to process. Even these instances of super-iority to observers, we would argue, are basednot on access to process but access to content.It is possible to describe three kinds of contentaccess that will allow for the individual's super-ior accuracy on occasion.

Knowledge of prior idiosyncratic reactions toa stimulus category. We have argued that per-ceived covariation between stimuli and re-sponses is determined more by causal theoriesthan by actual covariation. There are probablysome cases, however, where individuals haveidiosyncratic reactions to a particular stimulusthat only they have knowledge of. For example,a person may believe that he generally loathesstrangers who slap him on the back, and thisbelief may make him superior to observers inexplaining his feelings in such a situation. Wewould suggest that such cases may be rare,however, and that the vast majority of per-ceived covariations between stimuli and re-sponses may be determined by causal theoriesshared by both actors and observers.

Differences in causal theories between subcul-tures. It should be obvious that the individ-ual's reports will be superior to those of ob-servers when the observer is from a subculturethat holds different causal theories. Thus, afterattending a party at which a lively, high deci-bel band was playing, an 18-year-old wouldprobably accurately say that the music in-creased his liking of the party. If a 40-year-oldwere asked to predict how much he would en-joy such a party and why, he would be moreapt to say that the music would decrease hisenjoyment. Thus, actors' reports about stim-ulus effects will differ from observers' predic-tions about their own reactions to that stimu-

lus whenever the actors and observers belongto subcultures that have different causal the-ories about the effects of that stimulus. Thesereports will be similar whenever the actor andobserver share causal theories about the effectsof the stimulus in question, or whenever ob-servers are asked to predict the effects of thestimulus on a member of a subculture withwhich they are familiar. Thus, whereas the40-year-old's predictions about the influence ofthe music on himself will not correspond to theself-reports of the 18-year-old, his predictionsabout the influence of the music on the 18-year-old would probably correspond well to the lat-ter's own reports.

Attentional and intentional knowledge. An in-dividual may know that he was or was not at-tending to a particular stimulus or that he wasor was not pursuing a particular intention. Anobserver, lacking such private knowledge ofcontent, might often be more prone to error inhis assumptions about the causes of an individ-ual's behavior than the individual himself. Onthe other hand, private access to such contentmay also serve to mislead the individual. Oc-casionally, noninfluential stimuli may be morevivid and available to the individual than toan outside observer, for example, and thus theobserver might sometimes be more accurate byvirtue of disregarding such salient but non-influential stimuli.

Inadequate Feedback

It seems likely that another important rea-son for our belief in introspective awarenessstems from lack of feedback. Disconnrmationof hypotheses about the workings of our mindsis hard to come by. If an insomniac believesthat he is unable to get to sleep because of thestress of his life situation, he will always be ableto find evidence supporting the view that hislife situation is currently stressful. Indeed, theinsomnia should be proof enough of the stress-fulness of his life situation! And should he, inthe midst of the very most stressful episode ofhis life, get a good night's sleep, he scarcelyneed abandon his sensible hypothesis about thecause of his insomnia in general. He can simplyinfer that the unusual stress must have left himso exhausted that it conquered his insomniamomentarily.


Motivational Reasons

A final factor that may serve to sustain ourbelief in direct introspective awareness is mo-tivational. It is naturally preferable, from thestandpoint of prediction and subjective feelingsof control, to believe that we have such access.It is frightening to believe that one has no morecertain knowledge of the workings of one's ownmind than would an outsider with intimateknowledge of one's history and of the stimulipresent at the time the cognitive processoccurred.

Reference Notes

1. Aronson, E. Personal communication, 1975.2. Zimbardo, P. Personal communication, 1975.3. Nisbett, R. E., & Bellows, N. Accuracy and inaccur-

acy in verbal reports about influences on evaluations.Unpublished manuscript, University of Michigan,1976.

References

Abelson, R. P. Psychological implication. In R. P.Abelson et al. (Eds.), Theories of cognitive consist-ency: A sourcebook, Chicago: Rand McNally, 1968.

Argyle, M., Salter, U., Nicholson, H., Williams, M., &Burgess, P. The communication of inferior and super-ior attitudes by verbal and nonverbal signals. BritishJournal of Social and Clinical Psychology, 1970, 9,222-231.

Aronson, E., & Mills, J. The effect of severity of initia-tion on liking for a group. Journal of Abnormal andSocial Psychology, 1959, 59, 177-181.

Bern, D. J. Self-perception: An alternative interpreta-tion of cognitive dissonance phenomena. Psychologi-cal Review, 1967, 74, 183-200.

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

Bern, D. J., & McConnell, H. K. Testing the self-per-ception explanation of dissonance phenomena: On thesalience of premanipulation attitudes. Journal of Per-sonality and Social Psychology, 1970, 14, 23-31.

Berkowitz, L., & Turner, C. Perceived anger level, in-stigating agent, and aggression. In H. London & R.E. Nisbett (Eds.), Tliought and feeling: Cognitive al-teration of feeling stales. Chicago: Aldine-Atherton,1974.

Brehm, J. W. Modification of hunger by cognitive dis-sonance. In P. G. Zimbardo, The cognitive control ofmotivation. Glenview, 111.: Scott, Foresman, 1969.

Brehm, M. L., Back, K. W., & Bogdonoff, M. D. Aphysiological effect of cognitive dissonance underfood deprivation and stress. In P. G. Zimbardo, Thecognitive control of motivation. Glenview, 111.: Scott,Foresman, 1969.

Broadbent, D. E. Perception and communication. Lon-don, Pergamon Press, 1958.

Brock, T. C., & Grant, L. D. Dissonance, awareness,and thirst motivation. In P. G. Zimbardo, The cogni-tive control of motivation. Glenview, 111.: Scott, Fores-man, 1969.

Brown, R. Models of attitude change. In R. Brown, E.Galanter, E. H. Hess, & G. Handler (Eds.), New di-rections in psychology (Vol. 1). New York: Holt,Rinehart & Winston, 1962.

Burt, C. L. The young delinquent (4th ed.). London:University of Toronto Press, 1925.

Chapman, L. J. Illusory correlation in observational re-port. Journal of Verbal Learning and Verbal Behavior,1967, 6, 151-155.

Chapman, L. J., & Chapman, J. P. Genesis of popularbut erroneous diagnostic observations. Journal of Ab-normal Psychology, 1967, 72, 193-204.

Chapman, L. J., & Chapman, J. P. Illusory correlationas an obstacle to the use of valid psychodiagnosticsigns. Journal of Abnormal Psychology, 1969, 74,271-280.

Cohen, A. R., & Zimbardo, P. G. Dissonance and theneed to avoid failure. In P. G. Zimbardo, The cogni-tive control of motivation. Glenview, 111.: Scott, Fores-man, 1969.

Cottrell, N. B., & Wack, D. L. The energizing effect ofcognitive dissonance on dominant and subordinateresponses. Journal of Personality and Social Psychol-ogy, 1967, 6, 132-138.

Darley, J. M., & Berscheid, E. Increased liking as a re-sult of the anticipation of personal contact. HumanRelations, 1967, 20, 29-40.

Davis, J. A. Great aspirations: The graduate school plansof America's college students. Chicago: Aldine, 1964.

Davison, G. C., & Valins, S. Maintenance of self-at-tributed and drug-attributed behavior change.Journal of Personality and Social Psychology, 1969,11,25-33.

Dixon, N. F. Subliminal perception: The nature of a con-troversy. London: McGraw-Hill, 1971.

Dulany, D. C. The place of hypotheses and intention:An analysis of verbal control in verbal conditioning.In C. W. Eriksen (Ed.), Behavior and awareness.Durham, N.C.: Duke University Press, 1962.

Erdelyi, M. H. A new look at the new look: Perceptualdefense and vigilance. Psychological Review, 1974,81, 1-25.

Eriksen, C. W. Figments, fantasies, and foibles: Asearch for the unconscious mind. In C. W. Eriksen(Ed.), Behavior and awareness. Durham, N.C.: DukeUniversity Press, 1962.

Ferdinand, P. R. The effect of forced compliance on recog-nition. Unpublished master's thesis, Purdue Uni-versity, 1964.

Festinger, L. Cognitive dissonance. Stanford, Calif.:Stanford University Press, 1957.

Freedman, J. L. Attitudinal effects of inadequate justi-fication. Journal of Personality, 1963, 31, 371-385.

Freedman, J. L. Long-term behavioral effects of cogni-tive dissonance. Journal of Experimental Social Psy-chology, 1965, 1, 145-155.

Gaudet, H. A model for assessing changes in voting in-tention. In. P. F. Lazarsfeld & M. Rosenberg (Eds.),


The language of social research. New York: FreePress of Glencoe, 1955.

Ghiselin, B. The creative process. New York: Mentor,1952.

Goethals, G. R., & Reckman, R. F. The perception ofconsistency in attitudes. Journal of ExperimentalSocial Psychology, 1973, 9, 491-501.

Goode, W. J. After divorce. New York: Free Press ofGlencoe, 1956.

Greenspoon, J. The reinforcing effect of two spokensounds on the frequency of two responses. AmericanJournal of Psychology, 1955, 68, 409-416.

Grinker, J. Cognitive control of classical eyelid condi-tioning. In P. G. Zimbardo, The cognitive control ofmotivation. Glenview, 111.: Scott, Foresman, 1969.

Gross, L. Modes of communication and the acquisitionof symbolic competence. Seventy-third Yearbook of theNational Society for the Study of Education. Chicago:University of Chicago Press, 1974.

Jones, E. E., & Nisbett, R. E. The actor and the ob-server: Divergent perceptions of the causes of be-havior. In E. E. Jones et al. (Eds), Attribution: Per-ceiving the causes of behavior. Morristown, N.J.: Gen-eral Learning Press, 1972.

Kadushin, C. Individual decisions to undertake psy-chotherapy. Administrative Science Quarterly, 1958,3, 379-411.

Kahneman, D., & Tversky, A. On the psychology of pre-diction. Psychological Review, 1973, 80, 237-251.

Kelley, H. H. Attribution theory in social psychology.In D. Levine (Ed.), Nebraska Symposium on Motiva-tion (Vol. 15). Lincoln: University of Nebraska Press,1967.

Kelley, H. H. Attribution in social interaction. In E. E.Jones et al. (Eds.), Attribution: Perceiving the causesof behavior. Morristown, N.J.: General LearningPress, 1972.

Kelman, H. Deception in social research. Transaction,1966, 3, 20-24.

Kornhauser, A., & Lazarsfeld, P. F. The analysis of con-sumer actions. In P. F. Lazarsfeld & M. Rosenberg(Eds.), The language of social research. Glencoe, 111.:Free Press, 1955.

Kruglanski, A. W., Friedman, I., & Zeevi, G. Theeffects of extrinsic incentive on some qualitative as-pects of task performance. Journal of Personality,1971, 39, 606-617.

Latan£, B., & Darley, J. M. The unresponsive bystander:Why doesn't he help? New York: Appleton-Century-Crofts, 1970.

Lazarsfeld, P. F. (Ed.), Jugend und Beruf. Jena, Ger-many: Fischer, 1931.

Maier, N. R. F. Reasoning in humans: II. The solutionof a problem and its appearance in consciousness.Journal of Comparative Psychology, 1931, 12, 181-194.

Handler, G. Consciousness: Respectable, useful andprobably necessary. In R. Solso (Ed.), Informationprocessing and cognition: The Loyola Symposium. Hills-dale, N.J.: Erlbaum, 1975. (a)

Mandler, G. Mind and emotion. New York: Wiley,1975. (b)

Mansson, H. H. The relation of dissonance reduction tocognitive, perceptual, consummatory, and learningmeasures of thirst. In P. G. Zimbardo, The cognitive

control of motivation. Glenview, 111.: Scott, Foresman,1969.

Miller, G. A. Psychology: The science of mental life. NewYork: Harper & Row, 1962.

Moray, N. Attention: Selective processes in vision andhearing. London: Hutchinson Educational, 1969.

Neisser, U. Cognitive psychology. New York: Appleton-Century-Crofts, 1967.

Nisbett, R. E., & Schachter, S. Cognitive manipulationof pain. Journal of Experimental Social Psychology,1966, 2, 227-236.

Nisbett, R. E., & Valins, S. Perceiving the causes ofone's own behavior. New York: General LearningPress, 1972.

Nisbett, R. E., & Wilson, T. D. The halo effect: Evi-dence for unconscious alteration of judgments.Journal of Personality and Social Psychology, in press.

Pallak, M. S. The effects of unexpected shock and rele-vant or irrelevant dissonance on incidental retention.Journal of Personality and Social Psychology, 1970,14, 271.

Pallak, M. S., Brock, T. C., & Kiesler, C. A. Disso-nance arousal and task performance in an incidentalverbal learning paradigm. Journal of Personality andSocial Psychology, 1967, 7, 11-21.

Pallak, M. S., & Pittman, T. S. General motivationaleffects of dissonance arousal. Journal of Personalityand Social Psychology, 1972, 21, 349-358.

Polanyi, M. Personal knowledge: Toward a post-criticalphilosophy. New York: Harper, 1964.

Rosenfeld, H. M., & Baer, D. M. Unnoticed verbal con-ditioning of an aware experimenter by a more awaresubject: The double-agent effect. Psychological Re-view, 1969, 76, 425-432.

Ross, L. D. The intuitive psychologist and his short-comings: Distortions in the attribution process. InL. Berkowitz (Ed.), Advances in experimental socialpsychology. New York: Academic Press, in press.

Ross, L., Rodin, J., & Zimbardo, P. G. Toward an at-tribution therapy: The reduction of fear through in-duced cognitive-emotional misattribution. Journal ofPersonality and Social Psychology, 1969, 12, 279-288.

Rossi, P. H. Why families move: A study in the social psy-chology of urban residential mobility. New York: FreePress of Glencoe, 1955.

Schachter, S., & Singer, J. E. Cognitive, social, andphysiological determinants of emotional state. Psy-chological Review, 1962, 69, 379-399.

Schachter, S., & Wheeler, L. Epinephrine, chlorpro-mazine, and amusement. Journal of Abnormal andSocial Psychology, 1962, 65, 121-128.

Schlachet, P. J. The motivation to succeed and thememory for failure. In P. G. Zimbardo, The cognitivecontrol of motivation. Glenview, 111.: Scott, Foresman,1969.

Sills, D. L. The volunteers: Means and ends in a nationalorganization. Glencoe, 111.: Free Press, 1957.

Sills, D. L. On the art of asking "Why not?" Some prob-lems and procedures in studying acceptance of familyplanning. In All India Conference on Family Plan-ning (Fourth, 1961) Report of the Proceedings: 29thJanuary-3rd February 1961, Hyderabad. Bombay:Family Planning Association of India, 1961.


Slovic, P., & Lichtenstein, S. Comparison of Bayesianand regression approaches to the study of informationprocessing in judgment. Organizational Behavior andHuman Performance, 1971, 6, 649-744.

Snyder, M., Schulz, R., & Jones, E. E. Expectancy andapparent duration as determinants of fatigue. Journalof Personality and Social Psychology, 1974, 29, 426-434.

Spielberger, C. D. The role of awareness in verbal con-ditioning. In C. W. Eriksen (Ed.), Behavior andawareness. Durham, N.C.: Duke University Press,1962.

Storms, M. D., & Nisbett, ,R. E. Insomnia and the at-tribution process. Journal of Personality and SocialPsychology, 1970, 2, 319-328.

Treisman, A. M. Strategies and models of selective at-tention. Psychological Review, 1969, 76, 282-299.

Tversky, A., & Kahneman, D. Availability: A heuristicfor judging frequency and probability. Cognitive Psy-chology, 1973, 5, 207-232.

Tversky, A., & Kahneman, D. Judgment under uncer-tainty: Heuristics and biases. Science, 1974, 184,1124-1131.

Valins, S., & Ray, A. A. Effects of cognitive desensitiza-tion on avoidance behavior. Journal of Personalityand Social Psychology, 1967, 1, 345-350.

Waterman, C. K. The facilitating and interfering effectsof cognitive dissonance on simple and complex paired-associate learning tasks. Journal of ExperimentalSocial Psychology, 1969, 5, 31-42.

Weick, K. E. Reduction of cognitive dissonance throughtask enhancement and effort expenditure. Journal ofAbnormal and Social Psychology, 1964, 68, 533-539.

Weick, K. E. Task acceptance dilemmas: A site for re-search on cognition. In S. Feldman (Ed.), Cognitiveconsislencey. New York: Academic Press, 1966.

Weick, K. E., & Penner, D. D. Discrepant membershipas an occasion for effective cooperation. Sociometry,1969, 32, 413-424.

Weick, K. E., & Prestholdt, P. Realignment of discre-pant reinforcement value. Journal of Personality andSocial Psychology, 1968, 8, 180-187.

Wilson, W. R. Unobtrusive induction of positive attitudes.Unpublished doctoral dissertation, University ofMichigan, 1975.

Zajonc, R. B. The attitudinal effects of mere exposure.Journal of Personality and Social Psychology, 1968,8 (2, Pt. 2).

Zimbardo, P. G., Cohen, A., Weisenberg, M., Dworkin,L., & Firestone, I. The control of experimental pain.In P. G. Zimbardo, The cognitive control of motivation.Glenview, 111.: Scott, Foresman, 1969.

Zimbardo, P. G., Weisenberg, M., Firestone, L, & Levy,B. Changing appetites for eating fried grasshopperswith cognitive dissonance. In P. G. Zimbardo, Thecognitive control of motive lion. Glenview, 111.: Scott,Foresman, 1969.

Received December 6, 1976 •

Documents

Telling More Than We Can Know