Social Explanation and Social Expectation - American Psychological ...

10 downloads 74 Views 1MB Size Report
In more general terms, Ross et al.'s (1975) contention is that ... marriage to a lawyer who was frequently unemploj ed and unable to ... subsequent term cf military service, involvement in a number of ... the Peace Corps, participation in a dangerous medical experiment ..... judges) of the subjects' written explanations in our.
Journal of Personality and Social Psycholocy 1977, Vol 35, No. 11, 817-829

Social Explanation and Social Expectation: Effects of Real and Hypothetical Explanations on Subjective Likelihood Lee Ross, Mark R. Lepper, Fritz Strack, and Julia Steinmetz Stanford University Subjects in three experiments were induced to explain particular events in the later lives of clinical patients whose previous case histories they had read, and they were then asked to estimate the likelihood of the events in question. Each experiment indicated that the task of identifying potential antecedents to explain an event increases that event's subjective likelihood This phenomenon was replicated across a variety of clinical case studies and predicted events and was evident both under conditions in which subjects initially believed the events they explained to be authentic, only to learn afterward that no information actually existed about the later life of the patient, and under conditions in which subjects knew from the outset that their explanations were merely hypothetical. Implications of these findings for previous investigations dealing with belief perseverance and the consequences of hindsight perspective are outlined, and potential boundary conditions of the observed effect are discussed.

Our perceptions of social phenomena—from automobile accidents to political upheavals, from academic failures to acts of altruism—are organized in considerable measure by the underlying causal connections we detect or postulate. Attribution theorists, among many others, have long emphasized the extent to which causal inferences add coherence to our perceptions of the entities and events in our social world. Indeed, several generations of such theorists have explored both the logical rules or schemata that may underlie judgments of social causality and the biases that may distort such judgments (e.g., deCharms, 1968; Two of the three experiments described in this article were previously reported in a paper by Fritz Strack, Mark R. Lepper, and Lee Ross entitled "Explaining is Believing: The Effects of Providing a Causal Explanation on the Perceived Probability of an Event's Occurrence " This paper was presented at the 56th Annual Meeting of the Western Psychological Association, Los Angeles, April 1976. The present research was supported, in part, by National Institute of Mental Health Research Grant MH-26736 to Lee Ross and Mark R. Lepper and by a Harkness Foundation Fellowship held by Fritz Strack. The authors gratefully acknowledge helpful comments on the manuscript and/or research by Robert Abelson, John Carroll, David Greene, and Richard Nisbett. Julia Steinmetz is now at the University of Oregon. Requests for reprints should be sent to Lee Ross, department of Psychology, Jordan Hall, Stanford University, Stanford, California 94305.

Heider, 1944, 1958, Jones & Davis, 1965; Jones etal., 1972; Kelley, 1967,1973; Kruglanski, 1975, Ross, 1977). Recently, however, Ross, Lepper, and Hubbard (1975) have suggested that an unwarranted perseverance of initial beliefs and impressions may be one cost of man's proclivity for causal explanation. Briefly, Ross et al. (1975) reported two studies in which false feedback concerning performance on a novel discrimination task (i.e., distinguishing authentic suicide notes from fictitious ones) continued to influence both the self-perceptions of the actors and the social perceptions of observers even after that initial feedback had been totally discredited through an extensive, standard debriefing procedure. This perseverance effect has since been replicated using a variety of experimental settings and procedures for discrediting the evidential basis for these impressions of information (cf. Lau, Lepper, & Ross, Note 1; Jennings, Lepper, Ross, & Steinmetz, Note 2). For instance, Lau et al. (Note 1) demonstrated that students' erroneous impressions concerning their logical problem-solving abilities (based on their performance in an initial test situation) persevered, even after the students learned that good or poor teaching performance had virtually guaranteed their original success or failure. Despite subsequent exposure to alter-

817

818

ROSS, LEPPER, STRACK, AND STEINMETZ

native teaching methods, these initial beliefs persisted and influenced students' curriculum preferences in the classroom several weeks later. In discussing the biased attributional processes that might underlie such results, Ross et al. (1975) speculated that perseverance of initial impressions and expectations might be due, in part, to subjects' attempts to explain the evidence they had encountered. Once an action, outcome, or personal disposition has been interpreted as the consequence of known or even postulated antecedents, Ross et al. contended, these antecedents may continue to imply the relevant consequence, even after the evidence that initially led to this attribution has been removed or undermined. Consider, for example, a subject in the Ross et al. study who attributed her apparent initial success in discriminating real from fictitious suicide notes to her empathetic personality and the insights she gained from the writings of a novelist who committed suicide. Or, consider a subject who attributed her apparent initial failure at this task to her lack of familiarity with people who might contemplate suicide. Even after debriefing, these subjects retain a plausible basis for inferring the likelihood of the relevant outcome and for drawing correspondent inferences concerning their actual ability in the discrimination task. Similarly, observers who identified or inferred personal characteristics to account for the apparent initial success or failure of their peers also retained a basis for perseverance in their social attributions following debriefing. In more general terms, Ross et al.'s (1975) contention is that even the removal of all evidence that some event ever occurred may leave a set of salient or highly available (cf. Tversky & Kahneman, 1973) antecedents that continue to imply the occurrence of that event. With such a causal schema intact, the perceiver is apt to continue to see the event as more likely to occur than would have appeared likely prior to presentation of the now disallowed information. As a result, any conditions that induce an individual to explain an event should lead the individual to regard that event as more probable. The present research was designed to test directly the contention that the process of

explaining an event increases its subjective likelihood for the perceiver. Three experiments are reported, each dealing with explanations and predictions of behavior in a clinical judgment context. In Experiment 1, experimental subjects attempted to explain events that they believed to be authentic at the time of their explanation task; however, before making predictions, subjects learned that the events were fictitious. To determine whether the effects obtained in the first study were dependent on subjects' initial beliefs in the authenticity of the event, in Experiments 2 and 3 hypothetical explanation conditions were added to the research design. Under these conditions subjects knew the events to be fictitious at the time of the explanation. In all three experiments subjects were provided with detailed case histories, and experimental subjects were asked to write down factors or reasons that might have led the patient to a particular course of action. It should be noted at the outset that these explanation instructions represented compound manipulations and that the present experiments provided limited opportunity for isolating the particular features of these manipulations that are necessary or sufficient to influence subjective likelihood. The subjects' final task in each of these studies involved estimating the likelihood of several possible events in the patient's later life, including, of course, the particular events experimental subjects had previously explained. Experiment 1 Method Subjects Forty-eight Stanford University undergraduates served as subjects, receiving for their participation either a $2 payment or credit toward a class requirement. Subjects participated in the study in small groups, but they were randomly assigned to experimental conditions on an individual basis, and they did not communicate with each other during the session.

Experimental Materials The experimental materials providing the basis for the explanation and prediction tasks in this and the succeeding experiments were two authentic clinical case histories of approximately 3,000 words each taken

SOCIAL EXPLANATION AND SOCIAL EXPECTATION

819

from a recent clinical casebook by Goldstein and judgments of clinicians and laypersons in order to Palmer (1975). These two cabe histories, the cases of isolate the effects of clinical training Subjects proceeded ''Shirie> K." and "George P ," involved descriptions of to read about the "important clinical tasks" of explainthe patient's presenting sj mptoms at the time she or he ing past events and correctly anticipating future sought professional help, together with an extensive developments while avoiding the pitfalls of either summary of possibly significant clinical information, overinterpreling or neglecting relevant case material obtained through interviews with the patient, concern- Finally, subjects were reminded that the\ would be ing the patient's background and formative experiences working with authentic case histories. Both cases provided a wealth of detailed and potentially Control condition —The 16 subjects in the control diagnostic material on which subjects could draw in condition proceeded to read one of the two clinical explaining and predicting subsequent events in each case histories described above, with attention to patient's life. information about the patient's background, formative In one case, a \ oung housewife, Shirley K , arrived experiences, and symptomatic problems that might at the clinic complaining of frequent headaches and help predict later events in the patient's life Subjects dizziness. She expressed great anxiety over her un- were then asked to rate the likelihood of occurrence of controllable thoughts of harming her 2-year-old son each of five possible events in the patient's later life and repugnance for her current husband Her history These events were suicide, financial contributions to included an early and unhappy marriage to escape a the Peace Corps, participation in a dangerous medical manipulating mother and restrictive father, a subse- experiment, alcoholism, and volunteer work in a quent liason with a musician who fathered her son and political campaign. eventually committed suicide, and a current, abhorrent Explanation conditions.—The 32 subjects in the marriage to a lawyer who was frequently unemploj ed explanation conditions also were asked to read one of and unable to provide adequate support for his famih. the two clinical case histories Hut in so doing, they The report further described Shirley's reactions to the were asked to put themselves in the role of a clinical suicide of her lover, the death of her father, and the psychologist who has just learned about a subsequent commitment of her mother to a mental institution. event in the patient's life For half of the subjects the In the other case, George P., a middle-aged bachelor, relevant event was a successful suicide attempt (suicide was seen upon his readmission to a Veterans Admin- condition), for the remainder the relevant event was a istration hospital suffering from stomach pains and a series of financial contributions to the Peace Corps generalized weakness and malaise. His history included (Peace Corps condition) The subjects were instructed, an early separation from his family due to disagree- furthermore, to explain the event in question as follows • ments with his father, a period spent as a hobo, a "We are interested in what evidence, if any, you can subsequent term cf military service, involvement in a find in the case study that might help us to explain number of unsuccessful business ventures, and his or might have allowed us to anticipate Shirk)' K.'s eventual return home to care for his mother until her (or George P.'s) suicide (or contributions to the death. The report also described George's previous Peace Corps) " Subjects were first required to write hospital admissions and health problems, his potential down the "factors or reasons" that would explain a difficulties with unadmitted alcoholism, and a recent particular course of action. Then the} were asked to dispute which led George to resign from hib last job summarize the "main reason for the patient's action" as a food machine serviceman. without further consulting the case history or their Clearly, these two cases differed considerably in the notes. Case history assignments and explanation diagnostic material provided for subjects. The use of manipulations were orthogonal. Eight subjects were two different cases, which were varied orthogonally assigned to each of the four possible pairings of cases to the events to be explained, was simply an attempt to and events to be explained. insure that any results obtained would not be reThe instruction booklet then proceeded to inform stricted to a particular explanation/prediction context subjects in the explanation conditions that the experiAccordingly, case was treated as a factor in all of the menter did not, in fact, know anything about the life principal experimental analyses in the following studies. of the relevant patient subsequent to the case history With a single exception (to be reported in Experiment period. In addition, subjects were specifically informed 2), however, this factor exerted no influence relevant to that "no information is available about whether he the experimental hypothesis; hence, the data were (she) committed suicide, made financial contributions collapsed across this dimension. to the Peace Corps, or followed any other particular course of behavior." The booklet avoided any suggestion that subjects had been deliberately misled; it simply Procedure clarified the fact that no information was actually available concerning the patient's later life. This Upon their arrival, subjects were seated and pre- procedure was designed to discredit thoroughly the sented with a brief oral introduction to a study "in- authenticity of subjects' prior impressions and make volving clinical judgments to be made on the basis of them realize that their explanations had merely been actual case histories" The experimenter then dis- hypothetical, without resorting to the debriefing tributed folders containing all relevant experimental paradigm used in previous research. Finally, after instructions, materials, and measures. The subjects' receiving this information, explanation subjects were written instructions first presented the purported asked to rate the likelihood of occurrence of the same rationale for the study—an effort to contrast the

820

ROSS, LEPPER, STRACK, AND STEINMETZ

five pobsible events in the patient's later life that had been rated by the control subjects, including the two critical events that subjects in the two experimental conditions had previously explained Dependent measures -The likelihood ratings providing the dependent measures in Experiment 1 were introduced to all subjects as the "second important task faced by clinicians, that of prediction." Subjects were asked, specifically, to "compare the target person with the average American male [female] of his [her] age and to rate the likelihood that the patient would manifest certain behaviors in the future " Each likelihood estimate was made using a 9-point Likert-t) pe scale from —4 (much less likely than the average American male [female]) to +4 (much more likely than the average American male [female]). A midpoint labeled "about as likely as the average American male [female]"' was also provided. Explanation-condition subjects first rated the likelihood of the event they had explained, their third rating (of the five ratings) dealt with the likelihood of the event explained by the other explanation-condition subjects. Half of the control subjects rated suicide first and Peace Corps contributions third, the remainder rated these events in reverse order1 The order of the other events rated was constant across conditions When subjects finished their assessments of subjective likelihood for all five events, the experimental session was concluded. The true purpose of the study and its potential implications and possible personal relevance were revealed to subjects in detail Subjects were thanked for their participation and urged not to discuss the study with their peers.

Results Table 1 summarizes the effects of explanation upon subjects' likelihood estimates for suicide and Peace Corps contributions and thus presents the data directly relevant to the experimental hypothesis. An inspection of Table 1 reveals that, as predicted, the task of initially explaining a particular event in the life of a patient increased subjects' subsequent estimates of the actual likelihood of that event's occurrence. A series of 2 (Event Explained) X 2 (Case) analyses of variance performed upon likelihood estimates for the two critical events and upon the difference between these estimates confirms the statistical significance of the relevant effects. First, a comparison of the two explanation conditions reveals that suicide by the patient was rated more likely by those subjects who

had initially explained suicide than by those who had initially explained financial contributions to the Peace Corps, P(\, 28) = 14.04, p < .001. Conversely, Peace Corps contribu-

Table 1 Effects of Subject Explanation on Estimated Likelihood of Events in Experiment 1 Estimated likelihood of Peace Difference Corps in hkelicontrihood Suicide butions estimates

Event explained

n

Suicide Peace Corps contributions No explanation (control)

16

+2.33

-.84

+3.17

16

+.17

+1.14

-.97

16

+.50

-1.83

+2.33

Note Likelihood estimates are assessed relative to the likelihood of the event occurring in the life of an average American of the same age and sex as the patient The more positive the number, the greater the likelihood (up to a value of +4.00); the more negative the number, the lesser the likelihood (up to a value of -4.00).

tions were rated as more likely by the subjects who had explained such contributions than b\ those who had explained suicide, F(l, 28) = 8.31, p < .01. The effect of the event explained on the difference between these two likelihood estimates was, of course, highly significant, ^ ( 1 , 28) = 36.17, p < .001. Furthermore, control subjects who had explained neither event rated both suicide and Peace Corps contributions as less likely than did those subjects who had explained each of the two relevant events, F(l, 28) = 6.98, p < .05, and ^ ( 1 , 28) = 22.61, p < .001, respectively. Significantly, subjects in the two explanation conditions did not differ in their likelihood estimates for any of the three events that had never been the object of explanation in any conditions (i.e., medical experiment participation, alcoholism, and volunteer political work). Nor were there any significant interaction effects between the events explained and the case histories examined by subjects. The effect of explanation, in other words, was specific to the particular events that had been 1 It should be noted that the possibility of an orderof-rating artifact influencing probability estimates is dealt with explicitly, and discounted, through the research design subsequently employed in Experiment i

SOCIAL EXPLANATION AND SOCIAL EXPECTATION

explained but did not depend upon the content of the particular case history used in the explanation task. Having explained a patient's purported later actions in terms of plausible antecedent conditions described in a case history, subjects rated that behavior (in the absence of relevant data) as a relatively more likely outcome.

821

found the effects of explanation or construction of plausible social schemata with the effects of simply forming a firm initial belief in the relevant event's occurrence. As a result, it seems possible that the effects of explanation upon subjects' subsequent beliefs concerning the likelihood of an event in Experiment 1 ma}- occur only as a function of the relative irreversibility of subjects' initial beliefs (Ross et al., 1975) or the seeming "certainty of Experiment 2 hindsight" enjoyed by subjects (Fischhoff, To extend and clarify these initial results, 1975, 1976, Fischhoff & Beyth, 1975). By the design of Experiment 2 added hypothetical examining the effects of explicitly hypothetical explanation conditions to the nonhypothetical explanations and by comparing the relative conditions employed in the previous experi- impact of hypothetical and nonhypothetical ment. Thus, half the subjects in Experiment 2 explanations, we therefore hoped to explore were again led to assume that an event had the extent to which our initial findings were actually occurred, were asked to explain the dependent upon such processes. event, and only then were made aware that no information actually existed about the Method occurrence or nonoccurrence of the event in question. For these subjects, therefore, the Subjects procedure was identical to that employed in Thirty-two undergraduates of both sexes were Experiment 1. For the remaining subjects, the randomly assigned to experimental conditions m procedure was somewhat different. These Experiment 2 These subjects received credit toward a subjects were informed from the outset that course requirement for their participation the event to be explained was merely hypothetical—that is, they knew that no informa- Design and Procedures tion existed about the actual occurrence or The two case studies, the two critical events to be nonoccurrence of the critical event before explained, and virtually all other instructions and engaging in the hypothetical explanation task. measures were identical to those employed m ExperiInvestigation of the effects of such hypo- ment 1. The only exceptions were the omission of the thetical explanations was undertaken to assess control (no explanation) condition and the addition of hypothetical explanation conditions. the relative plausibility of several alternative In these added conditions subjects were informed at accounts of the results of nonhypothetical the outset that the explanation task was merely explanation conditions employed in the first hypothetical, the following paragraph in the initial experiment. On a relatively trivial level, for task descriptions was presented to subjects • example, it seemed conceivable that some Now, of course, we don't actually know anything about the behavior or experiences of George P. subjects in Experiment 1 may have ignored [Shirley K.] after the writing of this case study, and or failed to attend to the final experimental we have no way of finding out. We merely want you disclaimer of knowledge concerning the to imagine that you, as a clinical psychologist, have patient's later life and, accordingly, may learned that George P. [Shirley K.] committed suicide [contributed to the Peace Corps]. simply have persisted in their prior beliefs about the patient. By making the hypothetical These subjects, like those in Experiment 1, also renature of the explanation task explicit from ceived the explicit statement after the explanation the outset, we sought to reduce the likelihood task that no information was available concerning the patient's later life. of such responses. In the nonhypothetical explanation conditions, of At the same time, data from the hypothetical course, this additional paragraph was omitted, and the explanation conditions introduced in Experi- subject searched the case history for antecedent causes, ment 2 promised to have relevance for more believing that the event in question had actually substantive theoretical issues as well. Obviously occurred. As in the experimental conditions of Experinonhypothetical explanation conditions con- ment 1, it was only after the explanation phase was

822

ROSS, LEPPER, STRACK, AND STEINMETZ

Table 2 Effects of Hypothetical and Nonhypothetical Subject Explanations on Estimated Likelihood of Events i?r Experiment 2

although in the predicted direction, was not individually significant, F(l, 24) = 2.41, p > .10. The effect of event explained on the difference between the two likelihood ratings, however, was highly significant, F(l, 24) Estimated likeli= 10.82, p < .01. Moreover, in none of these hood of analyses did the interaction-between event Peace Difference explained and the hypothetical /'nonhypothetiCorps in likelical factor approach significance (all Fs < 1.25). contrihood Finally, as in Experiment 1, explanation had n Suicide butions estimates Event explained similar effects in the context of both case histories but had no significant effect on likeliSuicide 8 +2.94 Hypothetical -.25 +3 19 hood estimates for any of the three noncritical 8 +2.88 Nonhypothetical -.31 +3.19 events. 16 +2.91 Combined - 28 +3.19 Comparing the relative impact of hypoPeace Corps thetical versus nonhypothetical explanations contributions 8 +2 13 Hypothetical on subsequent expectations, the data suggest + .56 + 1 56 8 + 100 + 1.09 Nonhypothetical -.09 that both 13^63 of explanation had a sub16 + 1.57 Combined + .83 +.74 stantial effect on subjective likelihood estiNote Likelihood estimates are assessed relative to mates. When 2 (Event Explained) X 2 (Case) the likelihood of the event occurring in the life of an analyses of variance were performed, sepaaverage American of the same age and se\ as the rately for hypothetical and for nonhypothetical patient. The more positive the number, the greater explanation subjects, the sensitive dijferencethe likelihood (up to a value of +4.00); the more negative the number, the lesser the likelihood (up to in-likeliltood-estimates measure yielded a significant effect of explanation for the nonhypoa value of —4.00) thetical explanation conditions alone, F(l, 12) = 6.75, p < .05, and a marginally significant completed and the prediction phase was about to begin effect for the hypothetical explanation condithat these subjects learned that no information actually existed about the later life of the patient. tions alone, F ( l , 12) = 4.21, p < .10, despite the small sample sizes. Nevertheless, the results provide some hints Results that the relative impact of hypothetical and The effects of hypothetical and nonhypo- nonhypothetical explanations may depend thetical explanations on subsequent likelihood upon the specific events explained and cases estimates for suicide and Peace Corps contri- involved. Among subjects who had explained butions obtained in this second experiment are a patient's suicide, for example, the pattern summarized in Table 2. Overall, these results of results appears virtually identical for clearly replicate and extend the phenomenon hypothetical and nonhypothetical explanation demonstrated in Experiment 1. conditions. Among subjects who had explained Separate 2 (Event Explained) X 2(Hypo- Peace Corps contributions, by contrast, the thetical/Nonhypothetical Explanation) X 2- effects of explanation seemed weaker for (Case) analyses of variance were performed on hypothetical than for nonhypothetical condithe likelihood estimates for each of the two tions, although the relevant interaction effect events and on the difference between these two did not approach statistical significance. estimates. For suicide ratings, the effect of Finally, on the difference-in-likelihood-estiexplanation was statistically significant, mates measure, there was a significant secondF(l, 24) = 5.08, p < .05; that is, subjects order (Event Explained X Hypothetical/Nonwho explained suicide (hypothetical and non- hypothetical Explanation X Case) interaction, hypothetical conditions combined) rated suicide F(l, 24) = 5.07, p < .05, suggesting that the as more likely than did subjects who had relative effects of the two types of explanation explained Peace Corps contributions. For may vary somewhat as a function of specific Peace Corps ratings the relevant main effect, contextual factors and circumstances. Further

SOCIAL EXPLANATION AND SOCIAL EXPECTATION

experimental evidence relevant to the issue of hypothetical versus nonhypothetical explanation, however, and a further test of the range of applicability of the more general experimental hypothesis, are presented in the third and final experiment. Experiment 3 Experiment 3 again compared the effects of hypothetical and nonhypothetical explanation upon estimates of the subjective likelihood that the explained event had indeed occurred. To clarify and extend our earlier findings, however, the design and procedures employed in this study differed from those employed in Experiment 2 in several important respects. First, a new pair of critical events—involvement in a hit-and-run accident and candidacy for a seat on the city council—were substituted for the suicide and Peace Corps contributions events that had been used in the two previous experiments. This substitution was largely an attempt to extend the generality of the basic demonstration concerning the effects of explanation. However, it also permitted a further examination of the relative impact of hypothetical versus nonhypothetical explanation tasks with a second set of events differing in their evaluative connotations and a priori probabilities. Second, to provide an appropriate baseline for assessing the specific effects of the explanation procedure on these two new events, a no-explanation (control) condition was reinstituted in Experiment 3. As in Experiment 1, likelihood estimates from these control subjects were obtained directly after exposure to the case studies. Finally, a third change in design was dictated by the possibility that a simple order artifact had played some role in the prior two experiments. In these studies, subjects had consistently first rated the likelihood of the event they had previously explained and only then rated the other four events. This confounding of order and explanation suggested some obvious alternative interpretations of results; hence, in Experiment 3, order of prediction was made orthogonal to the designation of the particular critical event to be explained.

823

Method Subjects A total of 80 undergraduates of both sexes served as subjects in Experiment 3. Subjects were paid $2 for their participation in the stud} and were randomly assigned to conditions.

Procedure With the exception of the changes noted above, the procedures and measures in Experiment 3 differed from those employed in the two previous studies only insofar as was necessary in terms of the changes in critical events. Thus, one new noncritical event (financial contributions to Amnesty International) was substituted for one noncritical event (volunteer participation in a political campaign) used in the previous studies. Furthermore, minor editing of one case study was undertaken, and dependent measures and details of the instruction booklet were appropriately revised Otherwise, the critical passages in the instruction booklets dealing with the explicitly hypothetical or implicitly authentic nature of the events to be explained in the respective experimental conditions and the procedure for obtaining subsequent subjective likelihood estimates remained virtually unchanged from the previous study In this study, 64 subjects in the explanation conditions explained one or the other of the two critical events before making likelihood estimates, 16 subjects in the control condition made these estimates without any intervening explanation task Half of the subjects first rated the likelihood of the patient subsequently becoming involved in a hit-and-run accident, while the other half first rated the likelihood of the patient's subsequent candidacy for city council The alternative critical event was always the third of the five events rated by subjects, the second, fourth, and fifth events rated, accordingly, were the noncritical events that no subjects were induced to explain Within each explanation condition, 4 subjects received each of the eight possible combinations of hypothetical versus nonhypothetical explanation, case, and order of events to be predicted. Within the control condition, 4 subjects received each of the four possible combinations of case and order of events.

Results Preliminary four-way analyses revealed no significant main effects or interactions involving the order-of-prediction variable. Subsequent analyses and presentations of data, accordingly, were collapsed across this factor. Table 3, then, presents the relevant mean likelihood estimates. Inspection of these data reveals that the primary experimental hypothesis has been strongly confirmed. The likelihood

824

ROSS, LEPPER, STRACK, AND STEINMETZ

Table 3 Effects of Hypothetical and Nonhypothehcal Subject Explanations on Estimated Likelihood of Events in Experiment 3 Estimated likelihood of Hitand-

Event explained

Difference Running in likelirun for city hood n accident council estimates

Hit-and-run accident 8 Hypothetical + .59 - 2 73 Nonhypothetical 8 +2 31 -2.20 16 + 1 45 -2.46 Combined Running for city council 8 Hypothetical - . 6 1 + 1 41 Nonhypothetical 8 +.11 -.09 16 Combined - 25 + 66 No explanation 32 (control) - . 0 8 -1.55

+ 3 33 +4 52 +3.92 -2.02

+ .20

- 91

+ 1.47

Note. Likelihood estimates are assessed relative to the likelihood of the event occurring in the life of an average American of the same age and bex as the patient. The more positive the number, the greater the likelihood (up to a value of +4.00); the more negative the number, the lesser the likelihood (up to a value of —4.00).

of the patient's subsequent involvement in a hit-and-run accident was rated far greater by subjects who previously explained such an event (hypothetical and nonhypothetical conditions combined) than by subjects who explained the alternative event of city council candidacy, F(l,56) = 9.86, p < .01. Conversely, the likelihood of city council candidacy was rated greater by subjects who previously explained that candidacy than by subjects who explained the hit-and-run accident, F ( l , 56) = 39.36, p < .001. An analysis performed on the difference between likelihood estimates for the two events provides additional dramatic evidence of the overall effects of explanation on the estimated likelihood of explained events, F ( l , 56) = 36.75, p < .001. Finally, as in Experiments 1 and 2, the relevant effects were consistent across the two cases presented to subjects, and none of the three noncritical events showed any significant effects of the subjects' explanation tasks. The baseline provided by the no-explanation

(control) condition illustrates again the specificity of the explanation effects. In each instance it is the rated likelihood of the explained event rather than the alternative event that is most removed from the baseline estimate of control subjects. Thus, control subjects regarded both the hit-and-run episode and the city council candidacy as significant!} less likely than did the relevant experimental subjects who had previously explained each of those events, F(\, 44) = 5.58, p < .05, and F{\, 44) = 10.83, p < .01, respectively. Table 3 also summarizes the data comparing the impact of hypothetical versus nonhypothetical explanation. As in the previous experiment, the effect of event explained on the critical difference-in-likelihood-estimates measure was apparent both for the hypothetical explanation conditions alone, ^ ( 1 , 24) = 24.3, p < .001, and for the nonhypothetical conditions alone, F(l, 24) = 18.0, p < .001. Once again, however, it is difficult to assess the relative magnitude of the effects of these two different t>pes of explanation conditions. For likelihood estimates concerning a hit-and-run accident, nonhypothetical explanation conditions seemed to have a slightly more powerful impact than hypothetical explanation conditions. For estimates concerning city council candidacy, this pattern was reversed ; that is, subjects who explained the event knowing it to be hypothetical showed a greater impact of such explanation than subjects who explained the event believing it to have actualh occurred. As in the previous experiment, however, none of the interaction effects involving the hypothetical versus nonhypothetical explanation and event-explained factors approached conventional significance levels. Content and Quality of Subjects' Explanations ' In each of three studies, subjects in our experimental conditions were asked to consider available case study materials and to construct from these materials a plausible explanation for the possible occurrence of a subsequent event in a patient's life. In each study, these explanation procedures significantly increased subjects' subsequent estimates of the likelihood that the event in question had actuall)

SOCIAL EXPLANATION AND SOCIAL EXPECTATION

occurred. Before proceeding to interpret and discuss this effect, it seems appropriate to describe briefly the nature of the actual explanations provided by subjects in these studies. An examination of subjects' written explanations reveals two general findings that, though difficult to express in quantitative terms, may be germane to the interpretation of the effects that we have described. First, it seems clear from subjects' explanations that both case studies were sufficiently rich in detail and complexity to allow virtually every subject to construct at least one plausible framework or script within which the critical event seemed to make sense. Typically, these explanations consisted of attempts to infer and document consistent dispositional tendencies that seemed to account for the patient's subsequent behavior (e.g., Shirley's inability to relate intimately to other people, or her chronic suppression of overt sexual impulses; or George's inability to accept responsibility, or his rebellious independence), coupled in many cases with additional attempts to specify particular early or critical experiences that may have given rise to these propensities (e.g., the overprotectiveness of George's mother or the suicide of Shirley's lover). In both Experiment 2 and Experiment 3, moreover, there were no detectable differences in the essays written by subjects in the hypothetical and nonhypothetical explanation conditions.2 Perhaps more striking than the general plausibility of these essays, however, is the extent to which subjects' explanations provided qualitative evidence of the sorts of biases and selective information-processing strategies that we have suggested may underlie the effects °f explanation on subsequent prediction. For example, subjects clearly tended to interpret the meaning or diagnostic significance of prior events in the patients' lives quite differently as a function of the event that they had been asked to explain. Thus, George's naval service was seen as evidence of his desire "to be with People" by a subject who felt George's city council candidacy was the result of his general affiliative tendencies, but it was seen as evidence of George's general "rebelliousness against his parents" by a subject who felt his suicide had been calculated to get back at

825

those who had made his youth unpleasant. Similarly, the suicide of Shirley's lover was seen by several subjects who had been asked to explain her own suicide as a "model that led her to take her own life," whereas other subjects asked to explain her city council candidacy or contributions to the Peace Corps saw the same suicide as a significant source of guilt which led Shirley to seek expiation through public service or altruistic concern for others. In these and numerous other examples, the potential relevance and possible predictive value of particular bits of information about the patient appear to have been colored by the explanation task subjects had been given. General Discussion Taken together, the results of these three experiments provide clear and consistent support for the hypothesis that providing an explanation for an event substantially increases the subjective likelihood of the occur- A series of analyses were performed concerning the length and persuasiveness (as rated by three "blind" judges) of the subjects' written explanations in our experimental conditions Xone of these analyses revealed any significant or consistent differences between h>pothetical and nonhypothetical conditions (all b\ < 1.0). Additional!}', correlations between subjects' likelihood estimates and the length and persuasiveness of these explanations were calculated (collapsing across the hypothetical/nonh\ pothetical variable). These analyses provided no evidence that either length or persuasiveness was consistently associated with heightened subjective likelihood. The relevant correlations involving length and estimated likelihood were generally positive, although they never reached accepted significance levels The correlations involving persuasiveness and estimated likelihood were inconsistent, ranging from moderately positive to moderately negative values. In fact, the only statistically significant correlation obtained indicated that those subjects in Experiment 3 who wrote the most persuasive explanations for a patient's city council candidacy were the least inclined to estimate such a candidacy as likely (r = .54, p < .01). Obviously, one should interpret a single unpredicted but significant correlation in the midst of a number of nonsignificant ones with considerable caution, recognizing that multiple interpretations (not excluding chance) could be offered. We can at least conclude, however, that no simple positive relationship existed between the length or quality of the subjects' written explanations for an event a,nd their beliefs concerning the likelihood of that event.

826

ROSS, LEPPER, STRACK, AND STEINMETZ

rence of the event. Having generated a rists have long contended (cf. Heider, 1944, plausible account for suggesting how a particu- Kelley, 1967) that the occurrenceofasignificant lar event might have been predicted from and/or unanticipated event is likely to evoke know ledge of a patient's prior history, subjects a search for some explanatory framework that appeared consistently willing to make the will allow the individual to make sense of the inferential leap from possibility to probability. event. The present research, moreover, suggests Furthermore, the effect of explanation upon that the search for possible links between estimated likelihood occurred, even when specified consequences and possible antesubjects were not initially led to believe that cedents may itself be sufficient for producing the events had actually occurred. Though the changes in likelihood estimates of the type data provided some intriguing hints that the observed in previous research. Thus, the relative impact of hypothetical and non- apparent inevitability that events seem to hypothetical explanation procedures may de- accrue when viewed with hindsight may pend upon specific characteristics of the case or result, in large part, from the explanatory event to be explained, it is clear that the task framework the individual has generated in of explanation enhanced subjective likelihood, reflecting upon that event. Similarly, the even when subjects recognized from the outset formation of perceived antecedent-consethat the event described (and, hence, the task quence linkages postulated by Ross et al. (1975) to underlie post-discrediting impression of explanation) was purely hypothetical. The theoretical significance of the results perseverance need not be the product of obtained in the hypothetical explanation initial certainty about the relevant consecondition should be quite clear. These data quences; any circumstances, in fact, that lead help to eliminate a number of possible alter- the subject to generate such explanations may native interpretations that otherwise might be foster subsequent impression perseverance. offered to account for the effects obtained in nonhypothetical explanation conditions alone. Implications, Alternative Interpretations, and First, the hypothetical condition procedures Boundary Conditions largely rule out explanations involving the simple failure of subjects to attend or take It is perhaps difficult to identify circumseriously the experimenter's eleventh-hour stances in everyday experience that are disclaimers concerning the availability of exactly paralleled by the present experimental knowledge about the patient's later life. In conditions—particularly those involving hypothe hypothetical conditions, unlike the non- thetical explanation. In a very general sense, hypothetical conditions, subjects were clearly however, there are many circumstances in informed from the outset that their explana- which social decision makers—and those tions and predictions would deal with purely experts who advise them—are forced both to hypothetical events. Perhaps more signifimake predictions and to explain or identif} cantly, the results of the hypothetical condition the antecedent factors that justify such suggest that the relevant distortions in subjecpredictions. Clinical, educational, and corrective likelihood seem to have been neither simple tional settings and many social and political instances of impression perseverance in the planning tasks characteristically demand comface of subsequent discrediting information (Ross et al., 1975) nor the product of specific plexly related inferences about the necessit} biases introduced when events are viewed and sufficiency for the link between existing with "the certainty of hindsight" (Fischhoff, antecedents and one or more possible future 1975, 1976, 1977; Fischhoff & Beyth, 1975; outcomes. Guidance counselors, parole board members, and military contingency analysts, Slovic & Fischhoff, 1977). for instance, frequently are required to consider Indeed, the results obtained in the present and assess the likelihood of various alternative hypothetical explanation condition suggest a outcomes that are merely hypothetical at the fruitful perspective for viewing the research time of analysis and prediction. Are judgments findings relevant to both hindsight and and actions in such circumstances distorted perseverance phenomena. Attribution theo- by biasing effects of explanation upon per-

SOCIAL EXPLANATION AND SOCIAL EXPECTATION

ceived likelihood? Does the act of justifying a prediction or explaining its basis (i.e., by identifying the antecedents that might lead to the predicted outcome or event) produce unwarranted subjective certainty about one's decisions? The present research raises, but obviously cannot answer, these important questions. In view of the complexities of these applied questions, it should also be recognized that the instruction to explain an event is itself a compound manipulation with several components and consequences. In the present experiments, subjects were provided with a great deal of raw material from which to construct an explanation; they were led to anticipate the subsequent explanation task as they read the relevant material; they were asked to write down antecedents; and they were asked to underscore the particular causes or explanations they deemed most critical. Obviously, at present we cannot specify which of these components were necessary and/or sufficient to increase subjective likelihood estimates. For instance, subjects' responses as they read the case history may well have been influenced by the anticipation of the explanation task they were to undertake. Was such anticipation necessary, or could an after-the-fact explanation manipulation have similarly enhanced subjective likelihood? Alternatively, was such anticipation sufficient, or was the subsequent production and underscoring of explanations required to produce such results? These questions furnish an obvious topic for future research. Perhaps a more serious challenge to the interpretation and implications of the present demonstrations involves the role of salience. Specifically, it is possible that the task of explaining a particular clinical outcome was likely to leave that outcome more salient than nonexplained outcomes, even after its authenticity was subsequently undermined. This heightened salience rather than the explanation task itself could have, in turn, distorted subjects' ultimate likelihood estimates. We cannot dismiss this alternative interpretation on the basis of the present evidence, although it is worth emphasizing that the explanation manipulation did seem to influence the subjects' interpretation of the case history

827

materials as well as their final likelihood estimates. One may be tempted to resolve the salience issue by simply manipulating outcome salience in the absence of an explicit explanation task. The difficulty with such a strategy, however, is obvious. A salience manipulation alone may lead subjects to read and review case history material in terms of its capacity to explain this salient outcome, even in the absence of any explicit experimental instruction to do so. Indeed, a manipulation of outcome salience is but one of many types of cognitive manipulations and factors relevant to social perception that may influence the likelihood of causal explanation or the nature of the explanations that are formed. The personal relevance of events in question, the extent to which such events seem to violate prior expectancies or theories, and a host of instructional or contextual variables that affect one's point of view (cf. Abelson, 1976) all seem likely to play a role in determining when and how individuals will undertake the cognitive work involved in causal attribution (Abelson, 1972; Bern, 1970). Certainly, painstaking research will be required to explicate the types of cognitive events that influence subjective likelihood and the precise role of causal explanation in such influence. For the present, however, our own suspicion is that the production of formal causal propositions may not be a necessary condition for, or even a frequent characteristic of, the process by which explanation affects subsequent expectations. That is, it seems probable that the mechanisms responsible for the present results, and more importantly the mechanisms operating in most familiar social prediction contexts, may be far less systematic and logical than those suggested by current models of the attribution process (e.g., Kelley, 1967, 1972). Rather, the individual's effort at interpretation may be more accurately characterized as an attempt to construct or imagine a scenario or causal script (Abelson, 1976) into which the event can be easily fit. In this attempt, material that is concrete, salient, or of high associative value may be used extensively, whereas logically significant but more abstract evidence may be largely neglected (cf. Nisbett & Borgida, 1975).

828

ROSS, LEPPER, STRACK, AND STEINMETZ

Correspondingly, individuals may be likely to evaluate the appropriateness and explanatory power of a possible script by "satisficing" rather than "optimizing" criteria (Simon, 1957). Hence conditions that promote the availability (Tversky & Kahneman, 1973) of even a single coherent and plausible script in which an event appears as a probable final episode may prove sufficient to heighten the perceived likelihood of that event's occurrence, even if the relevant script is not the most satisfying or plausible one that could be generated by the individual. Implicit in our discussion of the link between satisfactory explanations and heightened subjective likelihood, of course, is the requirement that subjects be both able and willing to construct plausible causal scenarios or scripts. In the present studies, as we have noted, the case study materials provided a rich source of information for the explanation task, and subjects had little difficulty in selectively sampling and interpreting details to achieve this end. However, one can readily imagine situations in which attempts at explanation or script construction will be apt to fail. Although the present studies provide no relevant evidence, it seems quite possible that such failures may lead to decreases, not increases, in subjective likelihood. If our speculation that informal scenarios and scripts influence social expectation is proven to be correct, however, the practical implications of tlie present research, as well as the theoretical ones, will be greatly increased. Indeed, any social decision in which one must use information about the past to guide and justify future courses of action would be potentially subject to such influences. Conclusions The present experiments demonstrate that the act of providing a causal explanation of an event will increase the subjective likelihood of that event's actual occurrence. This increase in subjective likelihood as a function of explanation was demonstrated across a variety of specific events and was shown to occur whether the explanation task involved an attempt to explain an event that the subject initially believed actually had occurred or an attempt

to provide an explanation for an event presented as purely hypothetical from the outset. These findings clarify previous investigations dealing with belief perseverance and with the consequences of hindsight. They also prompt more general discussion of the cognitive components and consequences of explanation manipulations, the relationship between explanation and prediction, and the boundary conditions that may govern this relationship. Reference Notes 1. Lau, R. R., Lepper, M. R., & Ross, L. Persistence of inaccurate and discredited personal impressions: A field demonstration of altributional perseverance. Paper presented at the 56th Annual Meeting of the Western Psychological Association, Los Angeles, April 1976. 2. Jennings, D., Lepper, M. R., Ross, L., & Steinmetz, J. Persistence in impressions of personal persuasiveness: The perseverant impact of discredited success and failure experiences. Unpublished manuscript, Stanford University, 1977.

References Abelson, R. P. Are attitudes necessary? In B. T. King & E McGuinnies (Eds.), Attitudes, conflict, and social change. New York. Academic Press, 1972. Abelson, R. P. Script processing in attitude formation and decision making. In J. S. Carroll & J. W. Payne (Eds.), Cognition and social beliavior. Hillsdale, N.J.. Erlbaum, 1976. Bern, D. J. Beliefs, altitudes and human affairs Belmont, Calif.: Brooks-Cole, 1970. deCharms, R. Personal causation. New York: Academic Press, 1968. Fischhoff, B. Hindsight ^ foresight: The effect of outcome knowledge on judgment under uncertainty. Journal of Experimental Psychology: Human Perception and Performance, 1975,1, 288-299. Fischhoff, B. Attribution theory and judgment under uncertainty. In J. H. Harvey, W. J. Ickes, & R F. Kidd (Eds.), New directions in attribution research Hillsdale, N J . : Erlbaum, 1976. Fischhoff, B. Perceived informativeness of facts Journal of Experimental Psychology: Human Perception and Performance, 1977, 3, 349-358. Fischhoff, B., & Beyth, R. " I knew it would happen"— Remembered probabilities of once-future things Organizational Behavior and Human Performance, 1975, 13, 1-16. Goldstein, M. J., & Palmer, J. O. The experience o) anxiety. New York: Oxford University Press, 1975. Heider, F. Social perception and phenomenal causalityPsyclwlogkal Review, 1944, 51, 358-374. Heider, F. The psychology of interpersonal relationsNew York: Wiley, 1958. Jones, E. E., & Davis, K. E. From the acts to disposi-

SOCIAL EXPLANATION AND SOCIAL EXPECTATION tions. The attribution process in person perception. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 2). New York- Academic Press, 1965. Jones, E. E., et al. (Eds.). Attribution- Perceiving the causes of behavior. Morristown, N.J.: General Learning Press, 1972. Kelley, H. H. Attribution theory in social psychology. In D Levine (Ed.), Nebraska Symposium on Motivation (Vol. 15). Lincoln: University of Nebraska Press, 1967. Kelley, H. H. Attribution in social interaction In E. E. Jones et al. (Eds.), Attribution: Perceiving lite causes of beltavior. Morristown, N.J. • General Learning Press, 1972. Kelley, H. H. The processes of causal attribution American Psychologist, 1973, 28,107-128. Kruglanski, A. W. The endogenous-exogenous partition in attribution theory. Psychological Review, 1975, 82, 387-406. .Visbett, R. E., & Borgida, E. Attribution and the

829

psychology of prediction. Journal of Personality and Social Psyclwlogy, 1975, 32, 932-943. 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. Ross, L., Lepper, M R , & Hubbard, M. Perseverance in self-perception and social perception: Biased attributional processes in the debriefing paradigm. Journal of Personality ar.d Social Psychology, 1975, 32, 880-892. Simon, H. S. Models of man. New York: Wiley, 1957. Slovic, P., & Fischhoff, B. On the psychology of experimental surprises. Journal of Experimental Psychology: Buman Perception and Performance, 1977, 3, 544r-551. Tversky, A., & Kahneman, D. Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 1973, 5, 207-232. Received March 14, 1977 •

Announcement for Student Subscribers Outside the United States Students in psychology programs outside the United States are eligible to subscribe to the Journal of Personality and Social Psychology at the rate of $20 for 1978. This is the same rate paid by members of the American Psychological Association. In order to obtain a Student in Psychology application and order form to subscribe to JPSP and all other APA journals, write to Box M, American Psychological Association, 1200 Seventeenth Street, N.W., Washington, D.C. 20036, U.S.A.