When People Ask Why Questions, and the Heuristics of ... · When People Ask "Why" Questions, and the Heuristics of Attributional Search ... Central to attribution theory is the as-

Journal of Personality and Social Psychology1981, Vol.40, No. 4, 650-663

Copyright 1981 by the Amiicrican Psychological Association, Inc.0022-3514/81 /4004-0650S00.75

When People Ask "Why" Questions, and theHeuristics of Attributional Search

Paul T. P. WongTrent University, Peterborough,

Ontario, Canada

Bernard WeinerUniversity of California, Los Angeles

Five experiments making use of a self-probe methodology in both simulated andreal conditions demonstrated that individuals do engage in spontaneous attri-butional search. This search is most likely when the outcome of an event isnegative and unexpected. Content analysis of attributional questions also sug-gested that causal search is biased toward internality after failure but towardexternality following success. This reverse of the oft-reported hedonic bias im-plicates the adaptive function of causal search. The data also revealed that themost commonly used heuristic in attributional search is to center on the locusand control dimensions of causality. The importance of heuristics in causal searchand the advantages of the self-probe methodology employed in these investiga-tions are discussed.

Central to attribution theory is the as-sumption that people spontaneously engagein attributional activities. But there is littleor no published evidence to substantiate thisclaim (Bern, 1972; Wortman & Dintzer,1978). In the literature of attribution re-search, subjects typically are asked to makeattributions either by completing a fixednumber of rating scales or by providingopen-ended explanations for events. Bothmethods are highly reactive. In the absenceof adequate methodology, the issue ofwhether lay people engage in spontaneousattributional activities remains unsolved. Inthis article, we re-examine this issue, identifythe major preconditions for attributionalsearch, propose a new self-probe methodol-ogy, and then present data yielded by thismethodology.

A presumption guiding this article is thatit is more fruitful to ask when, rather thanif, attribution occurs. No one has proposedthat the attribution process goes on at alltimes. To the contrary, many investigators

This research was supported by a Leave Fellowshipfrom the Social Sciences and Humanities ResearchCouncil of Canada to Paul T. P. Wong and a grant fromthe Spencer Foundation to Bernard Weiner.

The authors are grateful to Alan Worthington forstatistical consultation.

Reprint requests should be sent to Paul T. P. Wong,Department of Psychology, Trent University, Peterbor-ough, Ontario K9J 7B8 Canada.

in the attribution area have contended thatindividuals carry with them sets of beliefs,schemas, or presuppositions as to how var-ious causes and effects are related (see Kel-ley & Michela, 1980). If our experiencesconform to our beliefs and expectations, thenthere is no need to search for explanations.For example, the conviction that "aptitude"is a relatively stable characteristic is gen-erally accepted in our culture; students withproven aptitudes in, for example, math orartistic endeavors are expected to do well inthose areas. Given this belief and expecta-tion, success in these activities should notcall for explanation or elicit attributionalsearch.

A corollary of the above reasoning is thatattributional search will take place whenone's experiences cannot be readily assimi-lated into one's existing belief system. A fre-quently encountered difficulty in the assim-ilation of information results from discon-firmation of existing beliefs and relatedexpectancies. Examples of this type ofdisconfirmation abound in real life: cheatingby a person with a reputation for honestyand integrity; failure by a student known tobe competent; or rejection of a manuscriptsubmitted by an established author. In eachof these cases, the disconfirmed expectationsare based on the belief that perceived dis-positions such as honesty, competence, and

650

HEURISTICS OF ATTRIBUTIONAL SEARCH 651

creativity are relatively stable. Disconfir-mation of expectations based on consensusinformation, such as task difficulty, also islikely to trigger the attribution process. Forexample, failure at an "easy" task is incon-sistent with the expectations generated bythe concept of "easy."

The hypothesis that expectancy disconfir-mation instigates attribution processes hasbeen alluded to by a number of investigators(see Lau & Russell, 1980; Pyszczynski &Greenberg, 1981). For example, it has beensuggested that atypical events are morelikely to elicit multiple causality than typicalevents (Kelley, 1971). Furthermore, it hasbeen demonstrated that novel or unexpectedevents promote exploration (Berlyne, 1960);attributional search can be considered oneinstance of the more general class of ex-ploratory behaviors.

In addition to expectancy disconfirmation,frustration (failure) is hypothesized to be asecond potent instigator of the attributionprocess. The law of effect dictates that or-ganisms are motivated to terminate or pre-vent a negative state of affairs. But effectivecoping importantly depends on locating thecause(s) of failure. In this case, attributionserves an adaptive function. In support ofthis line of reasoning, there is evidence thatrejection in an affiliative context is morelikely to elicit attributional search and in-formation seeking than is acceptance (Folkes,in press). Furthermore, it has been docu-mented that failure in instrumental learningand at achievement-related tasks promotesexploration (Wong, 1979). Finally, there isevidence to suggest that people are moti-vated to preserve their self-esteem; attribu-tion may also serve a defensive functionwhen self-esteem is threatened (Zuckerman,1979). In sum, it is hypothesized thatexpectancy disconfirmation (unexpectedevents) and frustration (nonattainment of agoal) will give rise to attributional search.

Unfortunately, the reactive methodologiescurrently in use do not permit an unambig-uous test of these hypotheses. Diener andDweck (1978) overcame the reactive issueand have reported a measure of spontaneousattributions. In one of their investigations,children were instructed to verbalize "whatthey were thinking about" while performing.But this procedure has a number of limita-

tions. When one is still engaged in problemsolving, it is only natural to be preoccupiedwith possible ways and means of solving theproblem rather than explaining the antici-pated outcome. The absence of attributionalcognitions in their mastery-oriented subjectsmight be so explained.

The present self-probe methodology is amodification of Diener and Dweck's proce-dure. Instead of asking subjects to verbalizewhat they are thinking while performing, weasked subjects to report what questions, ifany, they would ask themselves following aparticular outcome. Since causal explana-tions are answers to "why" questions, self-questioning seems to be a direct and naturalway to gauge the extent of attributionalsearch, and it at least has the face validityof measuring the presence and depth of one'ssearch for causal understanding. To broadenthe sample of cognitions reported by the sub-jects, in our final experiment we instructedthem to report whatever questions orthoughts came into their minds following anevent.

Experiment 1

The main purpose of this experiment wasto document that individuals do sponta-neously ask "why" questions and that theextent of causal search is determined by thenature of the outcome (success vs. failure)and expectancy (expected vs. unexpectedoutcome). We predicted that both frustra-tion (failure) and expectancy disconfirma-tion (unexpected outcome) would instigatemore attributional search than success andan expected outcome.

Given the above contexts, most of thequestions that people spontaneously ask wereexpected to be related to attributional search.However, individuals also may raise action-oriented questions (e.g., "What can I doabout the situaton?"). Studies of coping withstress (e.g., Folkman, Schaefer, & Lazarus,1979; Lazarus, 1966) have documented theexistence of "secondary appraisal," a processof evaluating one's coping resources and op-tions. It was predicted that failure would alsoelicit more of such action-oriented questionsthan would success.

Finally, it was anticipated that unex-pected outcomes would evoke more reeval-

652 PAUL T. P. WONG AND BERNARD WEINER

uative questions than expected outcomes.Since in the present study expectancy wasprimarily based on the belief that one wasa strong or weak student, expectancy dis-confirmation should have resulted in a stateof imbalance (Feather, 1971; Heider, 1958)or dissonance (Festinger, 1957) in regard tooneself. To maintain self-consistency, onemay either "explain away" the unexpectedoutcome or modify one's beliefs to accom-modate the outcome. The latter strategymay give rise to questions related to the reas-sessment of one's competence (e.g., "Am Ismarter than I think?").

Method

Seventy students (41 females and 29 males) partici-pated in the sudy as part of their course requirementfor introductory psychology at the University of Cali-fornia, Los Angeles (UCLA). They were tested in twogroups of approximately equal size.

The subjects were given a questionnaire containingfour hypothetical situations (2 levels of outcome X 2 lev-els of expectancy). They were asked to imagine thatthey expectedly or unexpectedly succeeded or failed ata midterm test. For example, in the unexpected failurecondition, they were to believe that they were "strong"in a subject, but they unexpectedly failed the midtermexam. The order of presentation of the four conditionswas randomized, and the subjects were instructed thatthey could work on these conditions in any order. Afterthe description of each condition, the subjects wereasked, "What questions, if any, would you most likelyask yourself?" They were told not to write any questionsif such inquiries would not characterize their thinking.No other instructions were given.

Results and Discussion

The responses were classified into fourmutually exclusive categories: attribution,action, re-evaluation, and miscellaneous, withan interjudge agreement of 94%. Attribu-tion questions are "why" questions con-cerned with the possible causes of the out-come (e.g., "Why did this happen?" "DidI study hard enough?"); action questions areconcerned with possible courses of actionand generally have a future orientation (e.g,"What can I do to pass?" "Shall I get atutor?"); and re-evaluation questions areconcerned with the reassessment of one'sability or aspiration (e.g., "Have I under-estimated myself?"). Miscellaneous ques-tions include any that cannot be classified

into these three categories. (The distinctionsand examples for each category were givento the judges prior to coding.)

Figure 1 shows the total number of ques-tions asked (top panel) as well as the numberof questions asked in the attribution (middlepanel) and action and re-evaluation (bottompanels) categories. Only 5% of the ques-tions fell in the miscellaneous category; theyare not included in the figure. Concerningcausal ascriptions, failure and unexpectedoutcomes generated more attributional ques-tions than did success and expected out-comes, respectively, F(l, 69) = 79.34, p <.001; F(l, 69) = 80.50, p < .001. There alsowas a significant Outcome X Expectancy in-teraction, F(l, 69) = 9.69, /K.001, pri-marily due to the very low rate of respondingin the expected success condition.

As hypothesized, failure generated moreaction-oriented questions than success, F( 1,69) = 32.35, p< .001. Expected failure gaverise to the greatest number of instrumentalquestions, resulting in a significant Out-come X Expectancy interaction, F(l, 69) =31.29, p<.001. In fact, expected failuregenerated four times more instrumentalquestions than did unexpected failure. Per-haps expected failure (i.e., prolonged frus-tration) poses a greater threat and calls formore instrumental considerations. There isindeed some evidence that expected failureis more stressful, because 29% of the instru-mental questions in the expected conditiondealt with escape/avoidance (e.g., Should Idrop the course?), whereas all of the instru-mental questions in the unexpected conditionwere related to mastery-oriented coping ac-tions (e.g., working harder, changing one'sstudy habits, etc.).

Re-evaluation questions also conformedto prediction, occurring only following un-expected outcomes. There is some suggestionof a positivity bias in that people were morelikely to re-evaluate themselves favorablyfollowing unexpected success than to con-sider downgrading themselves following un-expected failures, F(l, 69) = 5.80, p < .05.

In sum, it appears that individuals do en-gage in attributional search and are morelikely to do so following failure and unex-pected outcomes. There is also evidence ofaction-oriented questions, particularly after

HEURISTICS OF ATTRIBUTIONAL SEARCH

Experiment 1 Experiment 2 Experiments

653

oh-

O

fi-

4-

Totc

aen

r~

1 Questions -

FailureSuccess

•' "'•.••:•••&

->>,:

-

61

_o 4-ccLjJQQ

Attribution

UJ

2-

2-I -

Act on

-

Re-evaluation -

1 frSBil'j^ jj.''i"i'>j

n-r-H-,EXPECTED UN- SELF OTHER

EXPECTEDSELF OTHER

Figure 1. Mean number of responses in the four categories, as a function of the experimental conditions.

expected failure, and re-evaluation questionsgiven unexpected events, particularly afterunexpected success. Nearly all the questionsthat subjects asked can be classified into thethree categories of attribution, action, andre-evaluation. But attributional questionscomprised the largest proportion of the totalquestions asked.

Experiments 2 and 3People do apparently engage in attribu-

tional activities, but it is not known how they

search for causation. By causal or attribu-tional search we simply mean the process ofsearching for causal understanding. Al-though it has long been recognized that howpeople search for causal ascription is vitalto our understanding of the attribution pro-cess (Kelley, 1967, 1971), not much isknown about causal search. We assume thatthis search probably takes the form of a se-ries of implicit self-directed questions, forexample, "Is it because of me?" Such ques-tions are essentially hypotheses formulated


by an individual concerning possible causes.(How these hypotheses are tested is anotherfundamental issue in attributional researchthat will not be dealt with in this article.)We also assume that individuals use heuris-tic rules that restrict causal search to se-lected areas of the total possible solutions(see Simon & Newell, 1971). That is, causalsearch is neither random nor exhaustive butis guided by a set of heuristics. We concep-tualize these heuristics as various focuses ofattention that guide individuals to formulatehypotheses and seek relevant information intheir search for causal understanding.

The psychological significance of per-ceived locus of causality (Lefcourt, 1976;Rotter, 1966) and perceived control (Ban-dura, 1977; Langer, 1975; Seligman, 1975)has been demonstrated in a variety of situ-ations. It is hypothesized here that the pro-cess of causal search will first focus on thesource or locus of causality (whether thecause resides within the person or in the ex-ternal world) and then shift to the control-lability of the cause (whether it is subjectto personal influence). Finally, attention willturn toward causal stability (whether thecause is likely to change). These three fo-cuses of attention correspond to the threeprimary dimensions or properties of causesspecified by Weiner (1979). Intention andgenerality (globality) also have been sug-gested as possible dimensions of causality(Abramson, Seligman, & Teasdale, 1978;Rosenbaum, 1972). At present, there is noempirical evidence concerning which of thecausal dimensions is of primary considera-tion in attributional search.

Each causal dimension is conceptualizedas a continuum with opposite poles. For ex-ample, focusing on the locus dimension maybe oriented toward the internal or the ex-ternal pole, just as the focus on the controldimension may be toward the controllableor the uncontrollable pole. In the search forcausality, if an individual implicitly asks "Isit because of me?" followed by "Could Ihave prevented it from happening?", thenthis sequence of self-directed questions re-flects the heuristic of an initial search forinternal causes, followed by a focus on con-trollable causes. Focus of attention in causalsearch may influence the kinds of causal ex-

planations reached as well as the perceiveddimensional properties of a given causal as-cription.

Given the above assumptions concerningcausal search, the self-questioning method-ology is suited for the investigation ofsearching heuristics. However, the problemremains as to how to identify the dimen-sional focus of each self-directed question.The traditional approach of dimensional cat-egorization is to have raters code the re-sponses. But one obvious shortcoming of thismethodology is that the rater's coding maynot correspond with that of the subject's. Tocircumvent this problem, subjects in Exper-iments 2 and 3 coded their own verbaliza-tions into different dimensional focuses.Self-coding may be faulted as being just asartificial as coding by raters, but at least self-coding reflects the subject's own perceptionsand phenomenological experience.

The self-probe methodology used here,consisting of both self-questioning and self-coding, provided evidence concerning thepriority or temporal hierarchy of focuses ofattention. In addition, it provided data con-cerning relative dimensional salience, whichis here operationalized as the frequency ofoccurrence of dimensional focuses. Thus, theself-probe methodology revealed the relativepriority and salience of various dimensionalfocuses in causal search.

Because causal search may differ for ac-tors and observers (see Ross, 1977), per-spective (self vs. other) was included as anantecedent condition in Experiments 2 and3. Inasmuch as expected success yielded sofew attributional questions in Experiment 1,an "expected" condition was not included inthese experiments. Hence, Experiments 2and 3 combined unexpected success and un-expected failure with a self- and other-per-spective. In Experiment 2, a within-subjectsdesign was used, whereas in Experiment 3the four conditions were manipulated in abetween-subjects design.

Method

The subjects in Experiment 2 were 56 introductorypsychology students at UCLA (36 females and 20males) who participated for course credit. The subjectsin Experiment 3 were 86 females and 74 males from thesame population, randomly assigned to one of the four


treatment conditions. All subjects were tested in a groupsetting.

The subjects first received a questionnaire similar tothat used in Experiment 1. In the other-failure condition,for example, subjects were told, "You know your friendis strong in a subject, yet he failed at the midtermexam." Following each description, the subjects wrotethe questions they would most likely ask themselves. Aminimum of five questions in each condition was re-quired so that the temporal sequence of reportedthoughts could be examined.

After completion of the questionnaire, the concept ofcausal dimensions was introduced. The subjects weretold, "In seeking an explanation for success or failure,people often ask themselves certain questions regardingpossible causes. Generally, these questions can be de-scribed in terms of five different dimensions, represent-ing five different focuses of concern." The dimensions,presented in different predetermined random orders,were described as follows:

1. The locus dimension is concerned with the sourceof causality, this is, whether the cause resides in you,some other people, or in the situation.

2. The control dimension is concerned with the extentof one's control or mastery over various causal factors.

3. The intention dimension concerns responsibilityand purpose.

4. The stability dimension is concerned with predic-tion, that is, whether a causal factor will persist orchange over time.

5. The generality dimension is concerned with thegeneralizability of a causal factor to other situationsor to other people.

Each of the dimensions was illustrated with specificexamples. In Experiment 2, the examples were drawnfrom achievement-related situations, whereas in Exper-iment 3, the examples were unrelated to achievement.For instance, the examples for the Intention dimensionwere "Did the teacher fail me on purpose?" (Experiment2) and "Did he break my window on purpose?" (Ex-periment 3).

The subjects were instructed that a question could beclassified into any number of dimensions, according tothe focuses of concern that initially prompted the ques-tion. To facilitate the self-coding process, a grid wasprovided with the headings of the five dimensions ran-domly assigned to different columns, while the rows rep-resented the order of the questions. Subjects coded theresponses by placing a check mark in the appropriatedimensional column for each question (row). It was in-dicated that each question was to be coded indepen-dently of the preceding question.


The questions were again classified by twojudges into the four categories used in Ex-periment 1, with an interjudge agreement of

97%. To further examine the contents ofattributional questions, all attributionalquestions were then coded into different spe-cific causes of ability, attitude, cheating, ef-fort, emotion, error, general, help, knowl-edge, luck, motivation, physical conditions,situation, task, study method, and teacher.Several examples of each of the causal as-criptions were given to the judges. For in-stance, the "general" attributions were de-scribed as "nonspecific questions that areconcerned with seeking explanations for onoutcome," (e.g., "Why did that happen?""How did I get an A?"). Interjudge agree-ment on this classification was 95%.

The mean number of questions in thethree major categories for the four experi-mental conditions is shown in Figure 1.Analyses of the attributional questions re-vealed a main effect of outcome in both Ex-periments 2 and 3: F(l, 55)= 14.60, p<.001, and F(l, 156) = 19.93, p < .001, re-spectively. These results support the predic-tion that people ask more "why" questionsafter failure than after success. Only in Ex-periment 2 was there a significant Out-come X Perspective interaction, F(l, 55) =17.11, p < .001, indicating that the outcomeeffect was more pronounced in the self thanin the other condition. And, only in Exper-iment 3 was there a significant main effectof perspective, F(\, 156) = 11.46,/? < .001,with more attributional questions raised inthe other than in the self condition, In sum,the consistent finding across both studies, aswas reported for Experiment 1, is that frus-tration (failure) is more likely to instigateattributional search than is goal attainment(success).

In the action category, the prediction thatfailure leads to more instrumental questionsthan does success was clearly confirmed inExperiment 3, F(l, 156) = 9.89, p<.01,and approached an acceptable level of sig-nificance in Experiment 2, F(l, 55) = 3.67,p < .06. Considering all three experiments,instrumental concern was consistently greaterin the failure than in the success condition.

With respect to re-evaluation, the resultsagain are consistent with Experiment 1. Suc-cess led to more re-evaluation questions thandid failure: F(l, 55) = 4.89, p< .05 (Ex-periment 2), and F(\, 156) =16.08, p<


.001 (Experiment 3). The positivity bias inreassessment is therefore reliably demon-strated. This bias may be more than self-serving, for it also is evident from the per-spective of the observer. However, in Ex-periment 2 only, the tendency for greaterreassessment given success rather than fail-ure was higher in the self than in the othercondition: Outcome X Perspective interac-tion, F(l, 55) = 6.09, p < .05. Thus, thereis suggestive evidence that individuals maybe especially concerned with improving theirown self-concept or self-esteem (Zucker-man, 1979).

We turn next to the dimensional issues.The first topic to be examined was the tem-poral order of the search process. To deter-mine this, the initial occurrence of a dimen-sional concern was ascertained. If, forexample, the first question was judged ashaving a "control" focus, then control re-ceived a score of one, and so on. Given mul-tiple classification of the same question (i.e.,there was more than one focus of concern),all the dimensions involved received thesame score. And if a dimensional focus wasnot used to code any question raised withina condition, then it received an arbitraryscore of the total number of questions in thatcondition plus one.

The mean priority or sequence of eachdimension is portrayed in Figure 2. The maineffect of dimensions was significant in Ex-periments 2 and 3: F(4, 220) = 85.87,p < .001, and F(4, 624) = 41.70, p < .001,respectively. Linear contrasts confirmed theprediction that locus and control signifi-cantly (p < .01) differed from the other threedimensions. These results were exhibitedacross all the experimental conditions. InExperiment 2 only, locus was temporallyprior to control in the attributional search(p < .01). The inconsistency between the twoexperiments can be traced to the fact thatin Experiment 2, subjects tended to use sin-gle-dimensional coding for each question(M = 1.23), whereas in Experiment 3 therewas more double-dimensional coding (M =2.17). Given this inconsistency, it is not clearwhether a double-focus model is a more ac-curate description of the initiation of theattributional process than is a single-focusmodel. According to the former, the focus

is on the locus and control dimensions si-multaneously ("Am I personally responsiblefor what happened?"). According to the lat-ter, there is a focus on one dimension at atime, with inquiry about locus ("Is it becauseof me?") preceding thoughts about control-lability ("Did I have any control over whathappened?"). However, the data clearlydemonstrate that these two dimensions havethe highest priority in attributional search.

In addition to the temporal hierarchy ofdimensions, their relative salience was de-termined by the frequency of their occur-rence. The more frequently a causal dimen-sion is used to classify questions, the moresalient this dimension is presumed to be. Thefrequency data yielded a pattern of resultsidentical to the sequence findings, with locusand control significantly more salient (p <.01) than the other three dimensions. It ap-pears that a dimensional focus occurringearly in the attributional search also tendsto be repeated.

Although in causal search the focus isprimarily on locus and control regardless oftreatment conditions, the contents of attri-bution questions are dependent on both theoutcome and the perspective. Both experi-ments were very consistent in showing that

Locus Control Stobility Intention Generality

;n

:n

— 1

1

i —— 1 —

Experiment2

3

nnFigure 2. Mean temporal sequence of occurrence of thefive causal dimensions (low values indicate high priorityin the sequence). (For Experiment 4, the values arebased on self-ratings.)


several causal ascriptions are associated al-most exclusively with only one or two treat-ment conditions. More specifically, in Ex-periment 2, questions regarding error(11.4%), emotion (7.9%), and physical con-dition (4.9%) were associated with failure,but not with success (where the frequencyof these attributions was either 0 or less than1%). Questions about luck (15%), on theother hand, were associated with success, butnot with failure. Further, cheating (13.2%)and help (8.3%) were associated with others'success, but not with self-success. The samepattern of asymmetry was also obtained inExperiment 3, where error (11.6%), emotion(9.2%), and physical condition (8%) wereexclusively connected with failure, luck(11.5%) was associated with success, andconcerns about cheating (23.9%) and help(13%) were linked with others' success.

Effort, task, teacher, and general attri-butions were more evenly distributed acrossdifferent treatment conditions. However,analyses of variance (ANOVA) revealed thatin both Experiments 2 and 3, questions con-cerning effort were considered more fre-quently after failure (26.4% and 24%, re-spectively) than after success (17.2% and16.8%): F(\, 55) = 22.01,p< .001, and F(l,156) = 25.00, p < .001. Conversely, taskattributions were considered more fre-quently following success (22.8% and 27%)than following failure (7.8% and 13.3%);F(\, 55) = 9.17, p < .01, and F(l, 156) =11.63, p < .001. Other causal attributions,such as ability and attitude, occurred so in-frequently (less than 5% in all conditions)that they will not be discussed here.

The priority of causal ascriptions was de-termined by their initial occurrence in theseries of questions raised by the subjects.Once again priority data were identical tofrequency data. For example, in the failurecondition, effort had the highest frequencyas well as the highest temporal priority (themean initial occurrences being 2.6 and 2.4,respectively, for Experiments 2 and 3). AN-OVAS showed that in both experiments, effortwas considered earlier after failure than af-ter success: F(l, 55) = 24.60, p < .001, andF(l, 156) = 26.87, p< .001, respectively.Task attribution on the other hand, was con-sidered earlier after success than after fail-

ure: F(l, 55)= 9.17, p< .01, and F(l,156)= 23.48, p<. 001.

The above data suggest that both the sa-lience (frequency) and the temporal priority(sequence) of specific causal ascriptions aredependent on particular conditions, whereasthe dimensions (focuses) remain invariantacross conditions. This is logically possiblebecause given the same focus, the directionor orientation of the focus may be biased bythe outcome. For example, given the locusdimension, it may be oriented to internalcauses following failure but to externalcauses following success. To examine thispossibility, all causal attributions were clas-sified on the basis of logical analysis intoeither internal causes (e.g., effort, ability)or external causes (e.g., task, luck). ANOVAwere performed on the total frequencies ofall internal and external causes. In both ex-periments, failure gave rise to greater inter-nal orientation than success, F(l, 55) =101.57, /x.OOl, and F(l, 156) = 47.72,p < .001, whereas success resulted in greaterexternal orientation than failure, F(\, 55) =62.42, p< .001, and F(l, 156)= 14.18,p < .001. (It should be noted that the inter-nal cause considered most frequently—ef-fort—is also controllable, whereas the mostfrequently cited external cause—task ease—is uncontrollable.) This result lends somecredence to the notion that the orientationof a focus in causal search may be biasedby the nature of the outcome.

The above attributional bias is in oppo-sition to the well-known hedonic bias hy-pothesis, which posits that individuals inter-nalize success but externalize failure (seeBradley, 1978; Wong, Waiters, & Sproule,1978; Zuckerman, 1979). In the introduc-tion, it was suggested that attribution mayserve either an adaptive or a defensive func-tion. Perhaps defensive functioning predom-inates when one is publicly asked to give anexplanation of a task already completed,whereas adaptive functioning prevails whenthere is a search for a solution to problemsthat may recur. In our data, questions aboutinternal and controllable causes (effort) forfailure typically were followed by questionsabout possible coping actions. The adaptiveadvantage of this kind of bias in causal


search is that one is motivated to plan con-structive coping actions only when the causeis perceived as controllable by the actor.

Experiment 4

The prior experiments demonstrated thatlocus and control have the highest priorityand salience among the five attributionaldimensions. These results were based on self-coding of generated questions in an achieve-ment setting. In Experiment 4, the generalityof these findings was examined when thecontext in which the positive and negativeoutcomes occurred was not specified. In ad-dition, in the prior investigations the impor-tance of dimensional focus was determinedafter self-directed questions had been re-ported. In Experiment 4, subjects were askedto indicate the priority and salience of thedifferent causal dimensions during the pro-cess of causal search.

Method

Sixty-one introductory psychology students at UCLA(35 females and 26 males) participated in the experi-ment for course credit. They were tested in a groupsetting.

The subjects received a questionnaire containing, ina fixed random order, the four experimental conditions(2 levels of outcome X 2 levels of perspective) used inExperiments 2 and 3. The self-success condition, forexample, stated: "Suppose you just experienced an un-expected positive outcome. As you seek an explanation,what are your focuses of concern?" The subjects thenwere introduced to the five attributional dimensions dis-cussed in the prior experiments; again the dimensionswere characterized as focuses one might use in searchingfor an explanation. As in Experiment 3, the specificexamples used to clarify the dimensions were unrelatedto achievement events. In each of the four treatmentconditions, the five dimensions were presented in a fixedrandom order, and the subjects rated the priority andthe salience of each dimension in each condition. Prioritywas defined as "the order or sequence in which variousfocuses are considered by you." Salience was describedas "the extent to which a dimension or focus of concernis persistent or prominent in your mind." For each di-mension, rating scales ranging from 1 to 7 were pro-vided.


The mean priority (sequence) ratings areshown in the bottom panel of Figure 2. AnANOVA revealed a significant main effect

for dimensions, F(4, 240) = 17.01, p> < .001.Again, the locus and control dimensions hada significant higher priority (p < .05) thanthe three remaining dimensions across all theconditions. The mean salience ratings wereconsistent with the priority ratings: Locusand control had a greater salience than theother three dimensions, as confirmed by asignificant main effect for dimensions, F(4,240) = 16.97, p< .001, and orthogonal com-parisons contrasting the combination of lo-cus and control with the average of the otherthree dimensions (p < .05 .for all treat-ments). These data clearly replicated thefindings in the prior investigations, eventhough in the present experiment the contextof the outcomes was not specified and thedata were based on direct dimensional rat-ings. In the present study, we again failedto find any significant difference betweenlocus and control; thus, there is additionalsupport for a double-focus model of attri-butional search.

The correlations between the priority andthe salience ratings, considered separatelyfor each of the four conditions and for eachof the five dimensions, yielded correlationcoefficients ranging from .46 to .84, with allps < .001. This finding is consistent with ourprior results that dimensions considered ear-lier in time also persist longer in thought.

Experiment 5

Experiments 1-3 demonstrated that in-dividuals do ask "why" questions, even whennot specifically directed to do so. The findingthat most of the questions generated wereattributional indicates that causal search isprominent in people's minds. To increase theconfidence in these findings, in Experiment5 the effects of outcome and expectancy dis-confirmation were examined in a more nat-uralistic setting, with subjects asked to re-port whatever questions or thoughts cameinto their minds regarding their exam re-sults.

Students were recruited as subjects afterthey had completed a series of midterm ex-ams and had received exam feedback onmost, if not all, of their tests. Since therewas considerable difficulty in finding stu-


dents who actually expected to fail at theirmidterm tests, the words success and failurewere defined as "doing well" and "not doingso well." In addition, because of the diffi-culty in obtaining students' expectanciesprior to the midterm exams and maintainingtheir anonymity, we simply asked subjectsat the time of experimental testing to indi-cate whether their overall midterm resultswere expected or unexpected.

To further document that individuals ini-tially focus on the locus and control dimen-sions in causal search, in the presentinvestigation self-coding was replaced byan objective, information-seeking behavior.Subjects were informed that prior researchhad produced information that might helpthem determine the causes of their perfor-mance on tests and that this information(contained in five separate envelopes) wasorganized into five categories, each reflect-ing a different dimension of causality. Sub-jects were asked to choose which envelopesthey wished to examine first. The rationalefor this new procedure was that in causalsearch, people use heuristics to formulatehypotheses (i.e., ask questions) as well as toseek out relevant information. Thus, the typeof information they seek may reflect the heu-ristics they use. For example, if their heu-ristic is to focus on locus and control di-mensions first, they will naturally first askfor the information pertinent to those twodimensions. This information-seeking be-havior seems to be a more objective way ofdetermining the heuristics of causal searchthan the self-coding method used in Exper-iments 2 and 3.

In sum, the present study was designed toreplicate and extend the major findings ofthe preceding experiments in a more natu-ralistic achievement situation, with a broaderself-probe methodology to sample attribu-tional cognitions and a more behavioristicway of identifying the heuristics of causalsearch.

Method

One hundred volunteers from the introductory psy-chology class of Trent University were recruited as sub-jects. They were tested in small groups, with an averagesize of 15 subjects. The participants were first given aquestionnaire to complete and were initially asked to

indicate their own criteria for success and failure in thefollowing manner: "For me personally, doing well meansa grade of or higher; doing not so well meansa grade of or lower." They were then instructedto "reflect on and assess your performance on all yourmidterm tests. According to your own criterion, do youthink that you have done well or that you have not doneso well? Was your overall midterm result expected orunexpected?" Subjects indicated their responses by cir-cling a choice between doing well and not well and be-tween expected and unexpected. Then they were toldto write down in sequence what questions or thoughts,if any, came to their mind given the outcomes of theirmidterm exams.

In 'the second part of the experiment, the subjectswere informed that they might ask for informationyielded by past research to help them determine thecause(s) of their midterm exam performance. This in-formation was contained in five different envelopes rep-resenting the five causal dimensions. The dimensionswere introduced and defined in the usual manner on asheet of paper. Several bundles of envelopes were ob-servable to the subjects. The subjects were asked to in-dicate which envelope(s) they wished to examine firstby circling the appropriate dimension(s). They were alsotold that they could examine the rest of the envelopeslater. Following their choice, they approached the ex-perimenter for the envelope(s). Each envelope actuallycontained information pertinent to that causal dimen-sion. For example, the envelope on stability includedthe statement: "When people attribute success or failureto causes that are relatively stable, they tend to havestrong expectancy of having the same outcome again inthe future." Subjects looked at the information, returnedit, and asked for their remaining selection(s).


Of the 100 subjects, 17 indicated thattheir failure was expected and 24 that it wasunexpected; 43 reported expected successand 16 reported unexpected success. Thesubjective criterion for success revealed nosignificant differences between the outcomeand expectancy groups (F < 1).

Responses were classified into the usualfour categories, with an interjudge reliabilityof 95%. These data are shown in Figure 3.In view of unequal cells, planned orthogonalcomparisons were executed. As predicted,failure produced a greater number of re-sponses than success, F(l, 96) = 23.00,p < .001 (left panel). In addition, failure wasassociated with more attributional re-sponses, F(l. 96) =17.87, /x.OOl, andmore action-oriented responses, F(l, 96) =18.94, p<.001, than success. Concerningthe expectancy variable, unexpected out-comes were related to a greater number oftotal responses, F(l, 96) = 36.81, p < .001,


and more attributional responses, F(l,96) = 9.75, p<.001, than expected out-comes. Also in accordance with the findingsin Experiment 1, there was an Outcome XExpectancy interaction regarding action re-sponses, F(\, 96) = 5.01, p < .05, with ex-pected failure generating the most instru-mental responses. There was, however, nosignificant difference in re-evaluation. Sinceexpectancy was not explicitly derived frombeliefs about one's competence, as in Ex-periments 1 to 3, it is not necessary to reviseone's self-concept to accommodate the un-expected outcome; therefore, very rarely didsubjects reassess their own competence. Whatseems to be unique to the present naturalisticstudy is that nearly all of the re-evaluation

questions had to do with one's aspirations,values, or goals, as illustrated by these ques-tions: "Is university what I really want?","Is it worth my while to stay in university?","What can I get out of a university educa-tion?". Apparently, subjects had some sec-ond thoughts (i.e., re-evaluation) about thevalue of a university education; this type ofconcern was not evident in our prior simu-lated studies.

The responses were then subdivided intoquestions and statements (thoughts). Thepatterns of data were quite similar, with the"thought" data somewhat less sensitive tothe outcome and expectancy variables thanwere the "question" data. The contents ofstatements (thoughts) were very similar to

Total Responses_Tota I Quest ions_Total Statements

Eiil Failure

|""~1 Success

i/tZoa

tt

O 2.

ui 1'a

Z<

Attribution

Act ion

Re-evaluation

Expected Unexpected Expected Unexpected Expected Unexpected

Figure 3. Mean number of total responses (left panel), questions (middle panel), and statements orthoughts (right panel) in the four conditions.


those of questions except for the finding thatemotional expressions occurred almost ex-clusively in the form of a statement (e.g.,"I feel like getting violent with the marker,""I feel like crying," "Surprised how well Iam doing").

The data pertaining to information seek-ing are depicted in Figure 4. The upper panelof Figure 4 portrays the data of all individ-uals (n = 65) seeking only one kind of in-formation. The hypothesis that there isgreatest choice of locus or control informa-tion is supported by a binomial test (z =3.79, p < .001). The lower panel consists ofthe data of all individuals (n = 28) askingfor two envelopes. Here again the predictionthat most individuals would select the com-bination of locus and control evidence wasconfirmed by a binomial test (z = 8.30,p < .001). The seven individuals asking formore than two but fewer than five envelopesall included both locus and control in theirrequest.

General Conclusions

zo

10

16

11

1

4

0

Control Intentionlarolltv — .

in•On

ne

n

.030O

91

10

19 '

10.

3 •

Locoi Stability General

^p«

-

wmm

n nII n H

0 *

I I

Figure 4. Frequency of responses in seeking differenttypes of information. (Data represent number of indi-viduals.)

The five experiments presented here pro-vide evidence regarding several fundamentalissues in attributional research. First, it canbe concluded with reasonable confidencethat people do ask "why" questions, evenwhen they are not specifically directed to doso. The finding that most of their queriespertain to attributions indicates that causalquestions are prominent in thought. In Ex-periment 1, subjects were instructed to re-port whatever questions, if any, they wouldmost likely ask themselves in certain hypo-thetical situations. In Experiment 5, subjectswere asked to report whatever questions orthoughts came to their minds. In these ex-periments, subjects were recruited to partic-ipate in experiments designed to study cog-nitive processes, and it was made clear tothem in the instructions that they did nothave to make any verbal responses. Theseprocedures should not bias subjects towardasking attributional questions. The fact thatsubjects also asked other kinds of questions(i.e., instrumental and re-evaluation) indi-cates that they felt free to ask any kinds ofquestions. The only experimental constraint

was that subjects were exposed to a specifiedset of outcomes. Such a constraint was nec-essary because we were primarily interestedin whether this particular set of precondi-tions gives rise to attributional search.

Second, our results are very consistentin substantiating that frustration (failure)and expectancy disconfirmation (unexpectedoutcomes) promote attributional search.However, they are by no means the onlypreconditions for attributional search. Onecould readily identify a number of other pre-conditions. For example, stressful events(personal tragedy, interpersonal conflict,natural catastrophes, etc.) are likely to bepotent instigators of attribution. Novel andunknown events may have a similar insti-gating effect. For example, young childrenare generally inquisitive, not only becausethey have not yet developed an adequatestructure of causal beliefs and knowledge butalso because many experiences are still newto them. Events of great personal importancemay also be an effective antecedent for at-tribution.

Third, we have strong evidence that at-


tributional search primarily is focused on thelocus and control dimensions of causality.This finding is unlikely to derive from di-rective cues provided by our instructions orcontexts, because it was obtained in threedifferent experimental procedures: self-cod-ing (Experiments 2 and 3), self-rating (Ex-periment 4), and information seeking (Ex-periment 5). We also find that the focus wasoriented toward internal and controllablecauses (i.e., effort) after failure, but towardexternal and uncontrollable causes (i.e., taskease) following success, suggesting that theorientation of the search focus is dependenton the outcome. The significance of infor-mation on the locus and controllability ofcause(s) is quite apparent, inasmuch as cop-ing actions are very dependent on such in-formation.

The present studies have also raised sev-eral new issues. First, the attributional biasin causal search is in direct contrast to tra-ditional findings of success-failure bias thatare based on overt explanations. This reversehedonic issue may be resolved by testing thehypothesis that motivational forces at workin causal search may be different from thoseused to provide a public explanation. Re-search is needed to determine whether a con-trol or competence motive operates primarilyin causal search, whereas the self-enhancingor defensive motive predominates in publicexplanations.

A second unresolved issue is the extent towhich the orientation of dimensional focusdetermines the causal explanation reached.It seems reasonable to assume that what wefind depends to a great extent on where welook. If the heuristic used is one of searchingfor external and uncontrollable causes firstbefore considering internal and controllablefactors, it is likely that we will arrive at someacceptable external and uncontrollable causesif we search hard enough. A great void stillexists regarding the heuristics that peopleuse in attributional search and the effectsof using different heuristics on causal as-criptions. The understanding of the attri-bution process will depend on further anal-ysis of the heuristics of causal search.

We have demonstrated that the self-probemethodology is a sensitive and reliable wayof monitoring the presence and the extent

of spontaneous attributional search. It seemsonly logical that given any precondition oroutcome, one must first establish the pres-ence of spontaneous causal search beforeproceeding to investigate other aspects of theattribution process. The present simple andunobtrusive method has also provided evi-dence that action and re-evaluation relatedcognitions are sensitive to the major ante-cedents (i.e., outcome and expectancy) ma-nipulated. Also, the self-probe methodologyseems well suited to investigating the heu-ristics and the temporal course of attribu-tional search. In sum, the present method-ology has proven fruitful in unravelling thecognitive processes of causal search, and ithas taken us at least one step closer to tap-ping the subject's phenomenological expe-riences than have the more reactive meth-odologies commonly employed in attributionresearch.

ReferencesAbramson, L. Y., Seligman, M. E. P., & Teasdale,

J. D. Learned helplessness in humans: Critique andreformulation. Journal of Abnormal Psychology,1978,57, 49-74.

Bandura, A. Self-efficacy: Toward a unifying theory ofbehavioral change. Psychological Review, 1977, 84,191-215.

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

Berlyne, D. E. Conflict, arousal, and curiosity. NewYork: McGraw-Hill, 1960.

Bradley, G. W. Self-serving biases in the attributionprocess: A reexamination of the fact or fiction ques-tion. Journal of Personality and Social Psychology,1978, 36, 56-71.

Diener, C. T., & Dweck, C. S. An analysis of learnedhelplessness: Continuous changes in performance,strategy, and achievement cognitions following fail-ure. Journal of Personality and Social Psychology,1978, 36, 451-462.

Feather, N. T. Organization and discrepancy in cog-nitive structures. Psychological Review, 1971, 78,355-379.

Festinger, L. A theory of cognitive dissonance. Stan-ford, Calif.: Stanford University Press, 1957.

Folkes, V. S. An attributional analysis of affiliative be-havior. Journal of Experimental Social Psychology,in press.

Folkman, S., Schaefer, C., & Lazarus, R. S. Cognitiveprocesses as mediators of stress and coping. In V.Hamilton & D. M. Warburton (Eds.), Human stressand cognition: An information-processing approach.London: Wiley, 1979.

Heider, F. The psychology of interpersonal relations.New York: Wiley, 1958.


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

Kelley, H. H. Attribution in social interaction. Morris-town, N.J. General Learning Press, 1971.

Kelley, H. H., & Michela, J. L. Attribution theory andresearch. Annual Review of Psychology, 1980, 31,457-501.

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

Lau, R. R., & Russell, D. Attributions in the sportspages: A field test of some current hypotheses in at-tribution research. Journal of Personality and SocialPsychology, 1980, 39. 29-38.

Lazarus, R. S. Psychological stress and the coping pro-cess. New York: McGraw-Hill, 1966.

Lefcourt, H. M. Locus of control, current trends intheory and research. Hillsdale, N.J.: Erlbaum, 1976.

Pyszczynski, T. A., & Greenberg, J. Role of discon-tinued expectancies in the instigation of attributionalprocessing. Journal of Personality and Social Psy-chology, 1981, 40, 31-38.

Rosenbaum, R. M. A dimensional analysis of the per-ceived causes of success and failure. Unpublisheddoctoral dissertation, University of California, LosAngeles, 1972.

Ross, L. The intuitive psychologist and his shortcomings:Distortions in the attribution process. In L. Berkowitz(Ed.), Advances in experimental social psychology(Vol. 10). New York: Academic Press, 1977.

Rotter, J. B. Generalized expectancies for internal ver-

sus external control of reinforcement. PsychologicalMonographs, 1966, 80(\, Whole No. 609).

Seligman, M. E. P. Helplessness: On depression, de-velopment, and death. San Francisco: Freeman, 1975.

Simon, H. A., & Newell, A. Human problem solving:The state of the theory in 1970. American Psychol-ogist, 1971, 26, 145-159.

Weiner, B. A theory of motivation for some classroomexperiences. Journal of Educational Psychology,1979, 71, 3-25.

Wong, P. T. P. Frustration, exploration, and learning.Canadian Psychological Review, 1979, 20, 133-144.

Wong, P. T. P., Walters, D. A., & Sproule, C. F. Initialvalidity and reliability of the Trent Attribution Profile(TAP) as a measure of attribution schema and locusof control. Educational and Psychological Measure-ment, 1978,35, 1129-1134.

Wortman, C. B., & Dintzer, L. Is an attributional anal-ysis of learned helplessness phenomena viable? A cri-tique of the Abramson-Seligman-Teasdale reformu-lation. Journal of Abnormal Psychology, 1978, 87,75-90.

Zuckerman, M. Attribution of success and failure re-visited, or: The motivational bias is alive and well inattribution theory. Journal of Personality, 1979, 47,245-287.

Received May 2, 1980Revision received September 15, 1980

Documents

When People Ask Why Questions, and the Heuristics of ... · When People Ask "Why" Questions, and the Heuristics of Attributional Search ... Central to attribution theory is the as-