This article is for use only of students enrolled in GPS 520: Advanced Social Psychology for purposes associated with this course and may not be retained or further disseminated.

Leary, M. R., & Miller, R. S. (1986).  Clinical inference.  In M. R. Leary & R. S. Miller, Social Psychology and Dysfunctional Behavior: Origins, Diagnosis, and Treatment (pp. 122-146).  New York: Springer-Verlag Inc

Clinical Inference

Imagine that you learn of a 16-year-old girl who has begun inquiring "How many aspirin does it takes to kill somebody?" She is newly pregnant and has been deserted by her boyfriend, and is a member of a strict, authoritarian family in which the father drinks heavily. She is not an attractive girl, but she is a solid student. Is the girl really suicidal? Should she, or her entire family, be urged to seek psychological treatment? If so, what diagnosis applies and what treatment is appropriate? What criterion can gauge her progress, and when can therapy be stopped?

Clinicians and counselors are constantly asked to answer thorny questions like these, assessing the behavior and circumstances of other people and judging whether change is desirable or likely. In this, they are much like anyone else as they evaluate and interpret the behavior, intentions, and personalities of those they meet. The formal judgments of therapists usually have more impact than a layperson's casual observations, however, and errors are less allowable. It is thus particularly important to understand the person perception or interpersonal judgments of clinicians and counselors, and an extensive body of social psychological research is applicable here.

Unfortunately, the overriding lesson of studies of interpersonal perception is that we are much poorer judges of others than we believe ourselves to be. It has long been known, for instance, that. we quickly form impressions of other people from only limited information about them, and that, once formed, those impressions are rather resistant to change. Early studies of person perception in general (Asch, 1946; Bruner & Tagiuri, 1954) and of therapists' judgments of clients in particular (Meehl, 1960; Rubin & Shontz, 1960) showed that impressions were rapidly formed and that the presentation of additional, even contradictory information often did little to change them. In recent years, however, both social and clinical psychologists have come to realize that the potential problems with our social judgments go far beyond mere haste and resiliency. We seem to select, weigh, and interpret information in ways that sustain our existing beliefs, practically guaranteeing that, whatever we think, we will seem to be right. We see patterns in behavior where no patterns exist, and we are confident of factual beliefs that in truth are quite wrong. Moreover, our behavior toward people is influenced by our beliefs about them, and even our erroneous predictions about others can become self-fulfilling as we create in others the responses we expect to find. In short, recent studies of social cognition have compiled a catalogue of faults and biases in human judgment that seem to be common, if not unavoidable, components of social thought (e.g., Nisbett & Ross, 1980).  

For all of their expertise, cleverness, and wisdom, professional psychologists appear to be prone to some of these judgmental shortcomings, and in this chapter we examine those potential problems that seem to us to be most relevant to clinical practice. Our intent is not to belittle the processes of clinical inference, for its problems are pitfalls for us all, scientists and practitioners alike (Mahoney, 1976). As Wiggins (1981, p. 14) aptly noted, clinicians' errors are human errors, and we may all profit by understanding them better:

Are the judgmental shortcomings and biases of clinicians distinctively different from those of other professional decision makers? That is, have these shortcomings been demonstrated in groups of stockbrokers, physicians, intelligence analysts, electrical engineers, etc.? (Answer: definitely yes.) . . . Given that all of us-laypersons, clinicians and other professionals-are in the same boat, how would you evaluate our characteristic judgmental and inferential strategies with reference to the formal canons of scientific inference? (Answer: C-).

Our challenge, of course, is to improve social judgment, not simply to criticize it, and the final section of this chapter examines possible strategies for improving that C- grade, focusing on potential means of improving human (e.g., clinical) inference.

Statistical Versus Intuitive Decision Making

The first shortcoming we examine is also one of the best known. In deciding on diagnoses or courses of treatments, clinicians must integrate information from a wide variety of sources: psychological tests, assessment interviews, demographics, self-reports, etc. This is often done by subjectively, impressionistically, and intuitively combining the data to yield an overall judgment of (it is hoped) high accuracy. Unfortunately, Meehl (1954) and a legion of followers (e.g., Goldberg, 1970; Mischel, 1968; Sawyer, 1966; Szucko & Kleinmuntz, 1981) have persuasively argued that decision makers would do better by pulling out hand calculators and combining their data in a mechanical, statistical way. In fact, say these observers, clinicians' faith in their ability to accurately integrate diverse date is misplaced, whatever their experience and expertise; a clerk with a calculator can do better.

This assertion has elicited a number of defensive reactions from wounded clinicians (e.g., Holt, 1970), but it surprises us that anyone would be seriously insulted by the statistical advocates. No one has ever suggested that clinical expertise is not essential to the selection of the data to be considered and the validation of assessment techniques; the superiority of statistical techniques lies merely in the combination of disparate, incomparable data, and computers are unquestionably better at that than are people. The best way to integrate such data is to construct a linear model, as in regression analysis, in which the diverse sources of data are individually weighted so as to maximize the correlation between the overall equation and the criterion one is trying to predict. It sounds impersonal, but it works: "in the entire literature [of nearly 50 years' time] there is no study that has shown informal judgment procedures to be superior to predictions made from a simple linear statistical model" (Wills, 1982, p. 9).

The distinction between the expert selection of data and its statistical integration was clearly illustrated by Einhorn (1972). He asked physicians to predict an outcome more definitive than those usually faced by psychologists (i.e., death) by interpreting biopsies of patients with Hodgkin's disease. The experts were totally unable to accurately predict the survival time of the patients, but the information selected for analysis by the doctors did predict survival time when used in a linear regression model. Results like these, which also apply to psychologists (e.g., Dawes, 1979), are so consistent that Wiggins (1981) noted that "the original issue of clinical versus statistical prediction (should multivariate input data be combined by a person or a computer?) now seems to most of us a rather odd topic to occupy the attention of serious scientists for 25 years" (p. 14).

Still, many clinicians remain reluctant to use statistical decision-making procedures, and this hesitancy seems to be grounded in two separate concerns (cf. Dawes, 1979, 1982). First, statistical approaches appear to ignore the unique qualities of individual clients, the idiosyncrasies that can only be appreciated by a human judge. "Reducing people to numbers" seems distasteful, and clinicians may be reluctant to forgo the fine-tuning of a diagnosis that seems to require a professional's intuition. It is probably something of a professional conceit, however, to assume that one's own subjective impressions should replace the results of a standardized assessment of proven validity. Instead, one's intuitive judgments should be added to the linear model to be combined statistically with other, more impersonal data. Professional judgments can and should be made, but the integration of those judgments with other disparate data should still be done through some formal procedure.

This all sounds quite complicated, and the time, equipment, and sophistication needed to do a regression analysis have also been imposing barriers to the use of statistical techniques. Although it is less exact, intuitive decision-making is certainly simpler and quicker (Cantor, 1982). Remarkably, however, one need not do anything too complicated to improve on mere intuition. Simply guessing at how much each piece of data should be weighted does better than clinical intuition, provided that each weight is in the right (positive or negative) direction (Dawes, 1979). In fact, even treating all sources of data as equally valuable-that is, not differentially weighting them at all- standardizing them, and just adding them together works appreciably well (Dawes, 1982; Wainer, 1976). Sophistication is required in choosing the information one needs to know to make a diagnosis or select a treatment, but once the data are in hand their integration should be a simple mechanical task, not a test of professional wisdom. As Dawes and Corrigan (1974, p. 105) suggested, "the whole trick is to know what variables to look at and then to know how to add. 

Caution in trusting our intuitive decisions is also advisable in that we are often not consciously aware of how we have been influenced by particular data. Information that we think important may not have influenced us at all, whereas data we have dismissed as trivial may have had a major impact.

The Limits of Introspection

It seems plausible to assume that we are usually aware of the manner in which certain information about a person affects our subsequent judgments about him or her. Studies of judgmental processes suggest, however, that we actually have quite poor insight into the origins of our judgments and that we often do not know why we think the way we do. For instance, Nisbett and his colleagues (Nisbett & Bellows, 1977; Nisbett & Wilson, 1977a, 1977b) have repeatedly demonstrated that college students often overlook stimuli that are influencing them, often deny that genuinely important stimuli had any impact at all, and often identify as critical stimuli that did not influence them a bit.

In studies of this sort, subjects are usually asked to form an impression of some target stimulus as the available information about the target is systematically manipulated; as they form their judgments, the subjects are asked to indicate how influential various pieces of information are. For example, Nisbett and Bellows (1977) asked their subjects to evaluate a lengthy job application that either described or did not mention the applicant's attractiveness, intelligence, recent clumsiness, car accident, and so forth. The subjects' perceptions of how these data affected their judgments of the applicant were in most cases quite unrelated to each datum's actual effects. The subjects simply did not know whence their judgments came.

In another study, Nisbett and Wilson (1977a) asked subjects to evaluate via videotaped interview a foreign professor who was introduced either as a warm, likable fellow or as a cold, autocratic martinet. The subjects reported their global liking for the professor and specifically evaluated his appearance, mannerisms, and accent (which, of course, were the same for all viewers). As one would expect, they liked the "warm" professor much better than the "cold" one, and those global evaluations affected their judgments of the target's characteristics, which were seen as endearing with the warm version, irritating with the cold. The subjects categorically denied that their (supposedly objective?) ratings of the three attributes had been influenced by their liking for the target, however, and some of them actually insisted that their disapproval of his mannerisms had led to their disliking him-a complete reversal of the true causal relationship! The subjects were apparently unable to disentangle their specific ratings of the target from their global like or dislike for him, but they nonetheless were completely certain that their judgments were unbiased, impartial, and objective.

Nisbett and Wilson (1977b) provided another particularly illuminating example of the limits of introspection. One of their studies asked shoppers to evaluate four identical pairs of nylon stockings and to pick the pair of the best quality. A serial position effect emerged, with the stockings on the far right being preferred over those on the far left by nearly a 4: 1 ratio. When asked to explain their choice among equals, however, the shoppers conjured up imaginary differences in materials and manufacture as the reasons for their right-sided preference. Moreover, when specifically asked about a possible position effect, nearly all of them completely dismissed it, thinking it a silly idea. These good people did not know what had influenced their judgments and scoffed at one influential factor when it was presented to them.

Clinicians sometimes act the same way. Gauron and Dickinson (1966) asked psychiatrists to make tentative diagnoses and to rate their confidence in those diagnoses as a series of data were obtained about a client. A piece of information was designated as important if it substantially influenced a subject's final diagnosis or enhanced his confidence in his final judgment. The subjects' own estimates of the importance of each datum were totally unrelated to its true importance, however, and it again appeared that the subjects were largely unaware of the factors that had actually caused them to make the decisions they had made.

Findings like these unfortunately suggest that clinicians' inferences may be routinely affected by variables of which they are unaware, or worse, whose influence they consciously deny. Like the participants in Nisbett and Wilson's (1977a) study, for instance, clinicians seem to be influenced by their liking for a client. Attractive, likable clients may receive more help and greater effort (Doherty, 1971; Fehrenbach & O'Leary, 1982) and be judged as progressing more satisfactorily than less attractive clients (Brown, 1970; Shapiro, Struening, Shapiro, & Barten, 1976). As another example, clinicians may be led by clients' social class to interpret the same objective symptoms differently for clients of different classes (Abramowitz & Dokecki, 1977; Routh & King, 1972). Similarly, it may be hard for clinicians to make judgments that do not take a client's sex or race into account (cf. Zeldow, 1984).

In short, the certainty with which we identify the sources of our interpersonal judgments is often misplaced. We may feel objective and not be, and the reasons we announce may not be real. Feeling deliberate and impartial does not make us so.

Personalistic Biases in Judgment

A broad source of data that usually has less impact on our judgments than it should is the situational context in which a person behaves. We seem to be oriented toward assessing and understanding others' unique personalities and that orientation leads us to underestimate the extent to which others' behavior is influenced by the constraints and demands of the surrounding situation. In fact, our judgments are biased by a tendency to overestimate how much others' dispositions routinely influence their behavior. Indeed, this personalistic bias is so pervasive and robust that social psychologists have termed it the "fundamental attribution error" (Ross, 1977).

Some examples are in order here. In an early study of personalistic biases, Jones and Harris (1967) presented undergraduates with debaters' speeches supporting or attacking Fidel Castro. When the speeches were said to have been freely chosen by the debaters, the students reasonably assumed that the advocated positions reflected the speakers' private opinions. In another condition of the study, however, Jones and Harris forewarned subjects the positions had been arbitrarily assigned by the debate coach. Subjects in that condition generally ignored the fact that the speakers had had no choice of what to say and continued to believe that the advocated positions indicated the speakers' true beliefs.

In a similar study, Napolitan & Goethals (1979) had undergraduates talk with a counselor who behaved in a manner that was either warm and friendly or cool and aloof. The subjects were informed that the counselor's style was either spontaneous and genuine or prearranged and feigned for the purposes of the study. The subjects apparently found it difficult to believe that a person acting friendly (or unfriendly) wasn't really friendly (or unfriendly), however. They disregarded the possibility that the counselor's behavior had been influenced by the role she had been assigned and saw her behavior as a reflection of her personality in every case.

These and other studies (e.g., Ross, Amabile, & Steinmetz, 1977) demonstrate that we are likely to overlook or discount even obvious situational explanations for others' actions. "Behavior" seems to be synonymous with "personality" whatever the circumstances. Still, studies with naive undergraduates do not necessarily suggest that professional psychologists are also prone to a personalistic bias, and one may wonder whether these findings are generalizable to highly trained scientists.

Indeed they are. In a content analysis of the first 6 months of the 1970 Psychological Abstracts, Caplan and Nelson (1973) found that 82% of the research dealing with black Americans implicitly interpreted their difficulties in terms of "personal shortcomings," blaming blacks for their problems. Moreover, Batson, O'Quin, and Pych (1982) have argued that trained helpers (e.g., clinicians and counselors) are especially likely to infer that "a client's problem lies with the client as a person even when it is really due to some aspect of the client's situation" (p. 60). Not only do trained helpers clearly employ more dispositional explanations for clients' problems than untrained observers do (Batson, 1975; Batson & Marz, 1979; Pelton, 1982), but Batson et al. also suggested that both the information available to helpers and their role as helpers virtually guarantee that personalistic biases will occur.

As observers, for instance, helpers' attention is focused on the client rather than the situation; moreover, the client may be seen in a therapeutic setting far removed from his or her normal environment. The client's dispositions are thus much more salient than the situational context, and because we tend to attribute causation to whatever seems salient (Storms, 1973; Taylor & Fiske, 1978), the person gets blamed. As helpers, clinicians and counselors are expected to help, and most of their resources are oriented toward changing the client, not the situation (cf. Batson, Jones, & Cochran, 1979). In addition, the training they receive may set the expectation that the person is the problem (Snyder, 1977). In short, it seems that merely being a helpful observer of other people's problems fosters a personalistic bias; undergraduates asked simply to play the role of peer counselor in mock counseling sessions adopt a more dispositional point of view than their "clients" do (Snyder, Shenkel, & Schmidt, 1976).

A tendency to discount situational determinants of clients' behavior may not always mean that clinicians' judgments are actually wrong (Batson et al., 1982; Harvey, Town, & Yarkin, 1981). However, there are two other potentially troubling sequelae of personalistic biases that should not be overlooked. First, by focusing on clients' personalities, clinicians adopt a perspective that differs from that of their clients, who are likely to attend to the situation surrounding them (Jones & Nisbett, 1971; Storms, 1973). This actor-observer discrepancy is predictable, but it may occasionally engender some dispute (see chapter 7).

More treacherous is the possibility that personalistic biases lead to undeservedly negative perceptions of clients. To the extent that clients are held personally, even solely, responsible for their various dysfunctions while the contributions of adverse situations are overlooked, they appear to be even more incapable than they really are. Wills (1978) reviewed several studies that found that trained helpers' perceptions of their clients are more negative and damning than those of lay observers or the clients themselves (even when the clients were normal!). Moreover, the more experienced a therapist is, the more negative his or her judgments are likely to be (Fehrenbach & O'Leary, 1982; Wills, 1978). Other factors probably contribute to these harsh perceptions-for instance, dissimilarities between therapist and client and exposure to the worst of a client's behavior-but personalistic biases are almost certainly involved (Wills, 1978).

It is thus often disadvantageous to underestimate, however unwittingly, situational determinants of a client's distress. Not only may important influences be overlooked, but unjustified blaming of the client may result. Personalistic biases are just one example, however, of the manner in which unspoken assumptions, perceptual perspectives, and preconceptions about the nature of a client's problems influence and direct a helper's interpretation of events.

The Constraints of Preconceptions

Thus far we have asserted that we should not trust our intuitions when integrating diverse data, that our decisions are often influenced by factors of which we are unaware, and that our person-centered perspectives often ignore how much others' behavior is influenced by situational pressure. Each of these concerns is important, but they all speak to the manner in which we form our tentative judgments and hypotheses. What happens once we have some initial judgments in mind?

This is a vital question that will occupy us for much of the remainder of this chapter, As we will see, the manner in which we are influenced by preconceptions and tentative hypotheses is a key concern because we interpret events in ways that are likely to support and sustain our preconceptions, whatever they are. Once we form a judgment, or adopt a particular theoretical perspective, our evaluation of subsequent data no longer seems to be impartial and detached. Instead, our perceptions of the world are molded and shaped to fit our existing beliefs (instead of the other way around).

In a striking demonstration of this process, Lord, Ross, and Lepper (1979) found groups of Stanford undergraduates who were either in favor of the death penalty (believing that it deterred crime) or opposed to it (thinking it ineffective). Lord et al. presented both groups with two research studies, one that suggested that capital punishment was a deterrent and another that showed it was not. In addition, the two studies employed different methodologies: Each subject found that one study used a cross-sectional design, the other a longitudinal technique, but each design supported the death penalty for half the subjects and opposed it for the other half. Subjects were thus confronted with mixed evidence that only partly supported their existing beliefs and logically cast doubts on the certainty of their positions. Did they accept the mixed data and moderate their stances? Hardly. They praised and accepted the study that supported their beliefs, but criticized and rejected the study that opposed their positions. This meant that when the crosssectional design confirmed their beliefs it was judged to be the only reasonable way to study the issue, and when the longitudinal design supported them, it seemed to be the only sensible approach. Evidence with which they agreed was accepted at face value, but disconfirming data were severely criticized. The remarkable end result was that when the two opposing groups were given identical mixed information they each became more certain that they were right.

Our preconceptions clearly control our interpretations of incoming information, and other fascinating examples abound. Carreta and Moreland (1982) reminded us that during the U.S. Senate Watergate hearings in 1973, supporters of President Nixon were confronted with daily headlines that increasingly implicated Nixon in criminal activities. Most people who had voted for Nixon were largely unaffected by the bad news, however. Unlike McGovern supporters, whose attitudes toward Nixon became less and less favorable, believers in Nixon maintained their liking for him; they dismissed the headlines as liberal slander and shrugged off the burglaries as "politics as usual." Information that contradicted their beliefs was not believed.  

One need not oppose a person's deeply held beliefs to show that preconceptions control interpretations, however. Even simple presuppositions structure our perceptions and judgments. For instance, Snyder and Frankel (1976) showed Dartmouth undergraduates a silent videotape of a woman being interviewed and told some of them that the conversation concerned sex, others that it pertained to politics. Those watching the "sex" interview perceived the woman to be considerably more anxious and ill at ease than those watching the "politics" interview, although everyone, of course, saw the same tape.

Finally, social psychological research suggests that preconceptions control interpretations in subtle ways; we don't make egregious errors that are easily noticed, but we make errors nonetheless. For instance, Darley and Gross (1983) showed Princeton undergraduates one of two videotapes that provided information about the social class of a young girl named "Hannah. " Some subjects found that Hannah was rather poor, playing in a paved, deteriorating schoolyard and returning to a dingy, small home, whereas others found Hannah to be fairly well off, playing in expansive, grassy fields and living in a lovely house. On the basis of these demographic data alone, the subjects did not blindly assume that the upper-class target was doing better in school; they avoided such blatant stereotyping and guessed Hannah's academic achievement to be about average, regardless of her social class. However, when Darley and Gross then showed subjects an ambiguous tape of Hannah taking an aptitude test, her social status clearly affected their judgments (and in a manner of which they seemed to be totally unaware). The tape showed Hannah performing inconsistently, correctly answering some difficult questions but blowing some easy ones. All subjects saw the same tape, but they interpreted it very differently depending upon their beliefs about her social class. Subjects who thought Hannah was poor cited her many mistakes and judged her as performing below her average fourthgrade level; subjects who considered her well-to-do noted her many successes and rated her as better than average. These perceivers, then, did not leap to judgments about Hannah on the basis of stereotypes alone (thus making errors that might be easily noticed). They reserved judgment until they had more data but then interpreted her actions in a biased manner that was determined by their stereotypes. We can imagine the confidence they felt in their judgments, never realizing that other people with different preconceptions were witnessing the same test results and reaching completely contradictory conclusions (cf. Vallone, Ross, & Lepper, 1985).

Professionals' Preconceptions

As unlikely or threatening as it may seem, the preconceptions of scientists and therapists bias their judgments in just the same way (Mahoney, 1977). In particular, several studies show that clinicians' or counselors' theoretical orientations, their advance knowledge about a client, or even their knowledge of others' judgments influence the interpretations they form. In a renowned study, for example, Temerlin (1968) asked psychiatrists, clinical psychologists, and clinical graduate students to listen to a taped interview with a "prospective patient." Just before the tape was played, a prestigious, well-respected colleague mentioned either that the "patient" was "a very rare person, a perfectly healthy man," or interesting because he "looks neurotic, but actually is quite psychotic." When "healthy" was suggested to the therapists, they unanimously agreed that the target showed no signs of disturbance. By contrast, when disorder was suggested, 92% of the subjects diagnosed some dysfunction-60% of the psychiatrists considered him psychotic-although, again, all subjects had heard the same interview.

Clinical influence appears to be so complex a process (Cantor, 1982) that it is possible for professionals to disagree like this and never realize it. Indeed, the professional training a therapist receives may be a particularly important 11 preconception" that leads him or her to recognize different symptoms and to make different diagnoses (of the same behavior) than a colleague with another theoretical bent. For instance, Langer and Abelson (1974) asked both behaviorally oriented and psychodynamically oriented clinicians to evaluate a videotaped interview of a man who was described as either a "patient" or a "job applicant." Regardless of his label, the behaviorists found him to be fairly well adjusted, but the psychodynamicists considered the "patient" more disturbed than the "job seeker." Although reacting to the "patient" label this way is not necessarily unreasonable (Davis, 1979), a further study found that the psychodynamicists also believed the "patient's" problems to be more dispositional than the behaviorists did (Snyder, 1977). The biasing effects of professional orientation were also demonstrated by Allyon, Haughton, and Hughes (1965), who taught a schizophrenic, institutionalized for 20 years, to carry a broom by using cigarettes as reinforcements. They asked a psychoanalyst to evaluate her and were told that the broom was "(1) a child that gives her love and she gives him in return her devotion; (2) a phallic symbol, [or] (3) the sceptre of an omnipotent queen" (p. 3).

Thus, there seems to be little doubt that a therapist's professional perspective influences his or her interpretations of behavior (see also Bishop & Richards, 1984). Indeed, one reading of this literature suggests that the variability in clinical judgment from one observer to the next implies that "such judgments may be less informative about the patients they are meant to describe than about the clinician who makes them" (Grosz & Grossman, 1964, p. 112). Whatever the case, there are two further examples of the judgmental effects of preconceptions that deserve attention.

Anchoring

Preconceptions can function as cognitive "anchors," initial estimates that exert a disproportionate influence on final judgments. Anchoring effects occur when initial impressions are not thoroughly revised to accommodate new information; the result is that different preconceptions yield different judgments that remain biased toward the original estimates (Tversky & Kahneman, 1974). In one demonstration of this effect, subjects were provided a random estimate of some quantity (e.g., the percentage of African countries in the United Nations) by spinning a roulette wheel. When they were then asked to revise that "estimate" upward or downward to reflect the true value, subjects seemed reluctant to change that wholly arbitrary starting point, and their final estimates reflected the impact of the anchor. For instance, subjects starting with an "estimate" of 10% would revise upward to only 25%, whereas those starting with 65% would revise downward to only 45%; the difference in the two final judgments reflects an anchoring effect (Tversky & Kahneman, 1974).  

Similar results have been obtained with psychologists, psychiatrists, and social workers by Friedlander and Stockman (1983). They asked their subjects to evaluate a case of anorexia nervosa by successively reading five detailed interviews, which varied according to when indications of pathology appeared in the sequence of information. Substantial anchoring occurred in that late-appearing pathology had less impact on clinical judgment; when the anorexia was mentioned early in the sequence of information the case was viewed as significantly more serious, although by the end of the study all subjects had the same body of information in hand (cf. Jones, Rock, Shaver, Goethals, & Ward, 1968). In fact, experienced counselors and clinicians may be especially likely to settle on a diagnosis prematurely (Friedlander & Phillips, 1984; Hirsch & Stone, 1983; Houts & Galante, 1985), although all of us, once we have made a diagnosis, tend not to notice new symptoms that are inconsistent with that diagnosis (Arkes & Harkness, 1980).

In short, even if we have no biasing preconceptions about a case, the initial impressions we form as information is received often carry considerably more weight than they deserve. Whether they are preconceptions or first impressions, we are reluctant to revise our opinions when new data become available. Indeed, as we will soon discover, we may even cling to beliefs that can be shown to have no basis whatsoever. Before we consider belief perseverance, however, there is one more important example of the constraints of preconceptions to address.

Labeling

When a diagnosis or any other widely understood label is attached to dysfunctional behavior, it can become a public preconception that influences the subsequent judgments of an entire professional community. Labels can channel the perceptions and interpretations of observers just as private preconceptions do, but they can have even greater impact: Because they are consensually shared, many observers are affected. Indeed, some sociologists feel that the societal stigmatization that accompanies the labeling of abnormal behavior perpetuates deviancy that would otherwise be transitory (e.g., Scheff, 1975). This "labeling theory of mental illness" suggests that labels often set in motion events that exacerbate, not minimize, deviancy and that "helpers who label people may often create as much harm as good by the very process of practicing their trade" (Rappaport & Cleary, 1980, p. 77).

The concept that labeling creates deviancy has been roundly criticized (e.g., Gove, 1975), but it does have the value of alerting us to the deleterious effects that labels can have. Labels are particularly consequential in our educational systems, for example, where they are widely used. Fogel and Nelson (1983) provided teachers with a diagnostic label for a special education student and then showed them a videotape-of the child's behavior. The teachers' behavioral observations were not differentially affected by different labels, but their subjective evaluations of those behaviors were; the more severe the label, the more problematic the behavior seemed to be. Similarly, when Burdg and Graham (1984) informed some of their subjects that preschoolers whose intelligence they would test were "developmentally delayed, " the children received lower ratings and actually got lower test scores than other children who had been labeled as "normal." The "delayed" label affected not only the examiners' judgments but, apparently, their behavior as well, since they failed to elicit optimal performance from the children who had been randomly assigned the pejorative label. Perhaps labels can create ostensible dysfunction that would otherwise not exist.

Clinicians and counselors use labels, too, and in a famous (or infamous) study, Rosenhan (1973) examined the results of labeling sane people as schizophrenic. He encouraged eight normal people to report to mental institutions on both the East and West Coasts complaining of vague auditory hallucinations. All but one of them were diagnosed as schizophrenic and admitted to psychiatric wards, but once this was done they dropped all pretense of abnormality and tried to convince the staff that their sanity had returned. This proved hard to do. With a variety of anecdotal observations Rosenhan argued that "a psychiatric label has a life and influence of its own" (p. 253); once the pseudopatients were labeled schizophrenic, much of what they did, however normal, seemed indicative of schizophrenia. Average life histories were seen as pathogenic, ordinary behavior appeared maladjusted, and the only observers who seemed able to see through the ruse were other (real) patients. The pseudopatients were eventually released after stays that averaged 19 days, but their labels pursued them home; with the one exception, all of them were still presumed to have schizophrenia "in remission."

Rosenhan's suggestion that modern psychiatry is unable to distinguish sanity from insanity has been rejected by observers who argue that he tested no such thing (Farber, 1975; Millon, 1975; Spitzer, 1975; Weiner, 1975). Still, his investigation stands as a provocative reminder that we often see what we expect to see. Expectations, whether they be "preconceptions," "first impressions, " or "labels," exert powerful channeling effects on our interpretation and judgment of others' behavior. Moreover, such preconceptions are not easily changed. Even when we are shown that the evidence we used to establish a belief is completely false, the groundless belief may still persist.  

Belief Perseverance

We have already seen that once we form initial judgments, evidence that supports those beliefs is readily accepted whereas evidence that opposes those beliefs is denigrated (e.g., Lord et al., 1979). Another remarkable illustration of this tendency emerges from studies of belief perseverance that examine the manner in which people sometimes cling to an impression even "when the evidential basis for such a position is completely invalidated" (Jelalian & Miller, 1984, p. 29). In studies of this sort, subjects are usually provided information that straightforwardly leads to a particular judgment about themselves or others; they are allowed to formulate those judgments but are then shown that the original information was bogus and fictitious and utterly without diagnostic value. The common result is that the judgments persist although they are now without foundation.

The first such study was particularly relevant to social psychologists who deceive research participants and depend on a postexperimental debriefing to erase the subjects' misconceptions. Ross, Lepper, and Hubbard (1975) asked their undergraduate subjects to distinguish between authentic and fake suicide notes and provided them false feedback indicating that they had done very well, very poorly, or about average. Once this was done, however, the subjects were informed that the feedback was fake; they were assured it had nothing to do with them and were shown the experimenter's instruction sheet that had randomly assigned them to one of the three groups. The subjects were then asked to estimate how well they had actually done on the task and to predict their future performance on similar tasks.

Although the subjects understood that their feedback was randomly assigned to them, the judgments they had formed with that feedback tended to persist. Logically, none of the subjects had any valid knowledge whatsoever about their actual abilities, but the "success" subjects continued to believe that they were pretty good, the "average" group considered themselves mediocre, and the "failure" subjects felt that they were not particularly good at the task. Moreover, a second study by Ross et al. showed that belief perseverance also influenced observers who watched the procedure, hearing the false feedback and subsequent debriefing; they, too, continued to judge "successful" actors as much better at the task than their "failing" counterparts. Indeed, it was even harder to disabuse observers of their groundless judgments. Ross et al. found that it was possible to correct most of the belief perseverance of the actors by engaging them in an extensive discussion of the perseverance phenomenon and the harm it can do, but observers were largely unaffected by even this elaborate exposition. Once they had formed an impression of the actors-even one based on demonstrably fictitious information-it was remarkably resistant to change.

These findings are pertinent to cases like one recently in the news in which a schoolteacher was falsely accused (as it turned out) of child sexual abuse by a student bearing a grudge. We now know that the teacher is totally innocent, but having once associated him with a heinous crime, do we really like and trust him as much as we did before the incident? Erroneous impressions can persevere, and Wegner, Wenzlaff, Kerker, and Beattie (198 1) have shown that innuendo can have the same lasting impact on our judgments of others that factual accusations do. Even headlines that exonerate the innocent (e.g., "Andrew Winters Not Connected to Bank Embezzlement") produce negative impressions of those involved, and the perseverance phenomenon suggests that those impressions, although inaccurate and unjustified, change only with difficulty.

Why do beliefs persevere when the evidence that supports them is discredited? Ross et al. (1975) speculated that, once a person creates a plausible rationale for why he or she is good at evaluating suicide notes, for instance, the rationale may still seem reasonable and likely even when the information on which it was based is overturned. Ross, Lepper, Strack, and Steinmetz (1977) found that providing explanations for hypothetical events made them seem much more likely to occur, and, using the belief perseverance paradigm, Anderson, Lepper, and Ross (1980) showed that inventing explanations for one's beliefs made them much more resistant to evidential discrediting. Anderson et al. provided subjects case histories that suggested that risk taking made one either a good or bad firefighter, and then asked half of the subjects to explain why this should be so. When they then learned that the case histories were totally fictitious, subjects who had engaged in the causal processing exhibited considerably more belief perseverance than those who had not; indeed, those who had invented a theory to explain their beliefs were virtually unaffected by learning that the data base for their theory was worthless. To some degree, then, belief perseverance is based on the ease with which causal explanations spring to mind (Anderson, 1983b; Anderson, New, & Speer, 1985). Whatever explanations are most salient or available in our memories are likely to seem the most plausible, even when there is no valid basis for them.

Studies of belief perseverance do not deny that beliefs often change in response to new evidence. These studies do indicate, however, that change occurs grudgingly and that overwhelming evidence will often be required to change beliefs that were quickly formed (Jelalian & Miller, 1984). In clinical inference, data about clients are received over a period of time, the validity of a client's self-report is sometimes uncertain, and judgment involves constant causal analyses and explanations. As a result, the unwarranted perseverance of outdated hypotheses is always a possibility. When new, reliable evidence contradicts a clinician's working hypothesis and the hypothesis should be revised, it may still be hard to let go of the idea. Coles (1973) described Sigmund Freud's analysis of Leonardo da Vinci's life and art based on Leonardo's early memory of a "vulture" coming down to him in his cradle and opening his mouth with its tail. Vultures were symbols for mothers in ancient Egypt and, with this revelation in hand, Freud showed how only a man with Leonardo's special relationship with his mother could have created the Mona Lisa. Unfortunately, Freud was misled by a bad translation of Leonardo's notes; what Freud assumed was a vulture was actually a kite. With the basis for Freud's analysis discredited, did he revise his explanation of Leonardo's behavior? Of course not.

Perseverance phenomena are thus found outside as well as within the laboratory (Jelalian & Miller, 1984; Jennings, Lepper, & Ross, 1981), and those whose profession it is to explain and judge others' behavior should beware. This is doubly true, in fact, when the judge is able to select what data are to be considered; there is always the possibility that we not only value, but preferentially seek out, information that supports our existing beliefs.  

The Confirmatory Bias  

As they engage in counseling and psychotherapy, clinicians and counselors are not simply the passive recipients of whatever information their clients wish to divulge. Rather, they actively seek certain information, integrate that information to form ideas about the client and his or her dysfunction, and seek to test their hunches by gathering yet more information. Unfortunately, research on social inference suggests that, once initial impressions are formed, it is difficult for people to test the accuracy of those inferences in an unbiased fashion. Clinicians' and counselors' efforts to test their clinical intuitions can lead to erroneous conclusions. For instance, Snyder (1981) and his colleagues have repeatedly shown that undergraduate subjects, when asked to determine whether a belief about another person is true or false, are more likely to pursue information that will confirm the belief than to inquire after data that will prove it wrong. They seem to employ a one-sided strategy of seeking only instances that support their hypotheses instead of evenhandedly seeking instances that both do and do not fit them. Thus, Snyder found that people are biased toward confirming their assumptions about others and, as a consequence, rarely obtain impartial, representative information about them.

For instance, Snyder and Swann (1978a) invited half of their subjects to determine whether a person they would soon interview was an introvert; other subjects were asked to assess the person's extraversion. All the subjects were then provided a list of "interview topic" questions with which to conduct the interview. The questions were either neutral (e.g., "What are the good and bad points of acting friendly and open?") or biased toward eliciting introverted (e.g., "What do you dislike about loud parties?") or extraverted (e.g., "What do you do when you want to liven things up at a party?") responses. In general, the subjects selected questions that were likely to elicit evidence in support of their preconceptions about the person. The two groups of subjects thus adopted two very different lines of investigation, with each group ensuring that their targets would report many of the behaviors they expected to find. Indeed, Snyder and Swann found that the interviews were so biased that judges listening to audiotapes of the interactions actually believed the targets to be fairly introverted or extraverted, depending on the interviewers' respective preconceptions.  

An impersonal example may help the reader appreciate how biased such confirmatory -hypothesis testing can be. Suppose you are asked to determine the numerical rule we have in mind that explains this sequence of numbers: 2, 4, 6 (Wason, 1960). If you are allowed to generate as many series of three numbers as you like to test your understanding of our rule, being told each time whether your examples do or do not fit the rule, how would you proceed? In our experience with hundreds of students, we have found, like Wason (1960), that nearly everyone tests only numbers that fit their hypotheses; virtually no one tests possibilities that would prove them wrong. If they guess the rule is "all even numbers" they try 8, 10, 12 (which fits our rule); if they believe the rule is "increments of two" they try 1, 3, 5 (which also fits our rule). Very few explore a possibility that would negate their hypotheses, like -3, 0, 147, (it, too, fits our rule!), and by seeking only confirming instances the vast majority convince themselves of an incorrect belief. Few persist long enough to uncover the actual rule, "three ascending numbers " "  

Importantly, Snyder and White (1981) found that people recognize the value of disconfirming evidence but are generally just unwilling to pursue it. Snyder and White showed that when subjects were asked to determine if a person was not an introvert (or extravert), they sought disconfirming data with the same single-mindedness with which they usually pursue confirming instances. Still, confirmatory hypothesis testing appears to be the norm. Snyder and Swann (1978a) found that subjects continued to use confirmatory strategies even when they knew the hypothesis they were testing was rather unlikely to be true, and again when they were given a $25 incentive for accuracy. And even when subjects were given competing hypotheses, being asked to determine whether a person was "more like an extravert or more like an introvert," they chose one possibility and tried to confirm it instead of adequately testing both hypotheses (Snyder & Swann, 1978b).

Do similar biases affect clinicians and counselors? To the extent that they do, psychological professionals may occasionally convince themselves of hypotheses about clients that are simply untrue. Snyder (1981) suggested, for example, that:

the psychiatrist who believes (erroneously) that adult gay males had bad childhood relationships with their mothers may meticulously probe for recalled (or fabricated) signs of tensions between their gay clients and their mothers, but neglect to so carefully interrogate their heterosexual clients about their maternal relationships. (p. 294)  

Such a practice would no doubt confirm the clinician's expectation, since nearly everyone could report significant conflict with his or her parents (and spouse, and children, and colleagues) if asked (Renaud & Estess, 1961).  

In fact, several studies have examined clinicians' hypothesis-testing strategies, and when Snyder's (1981) method of asking subjects to choose their questions from a preset list is employed, even experienced professionals often lapse into a confirmatory approach (e.g., Dallas & Baron, 1985). When they are allowed to construct their own questions, however, the clinicians usually adopt a more evenhanded strategy, sampling both confirming and disconfirming data (Dallas & Baron, 1985; Strohmer & Chiodo, 1984; Strohmer & Newman, 1983). Overall, the less structured an interview situation is, the less likely confirmatory bias appears to be (Clark & Taylor, 1983; Trope & Bassok, 1982).

Thus, the confirmatory bias may play only a minimal role in clinical inference, but no one suggests that it should be casually dismissed; observers unanimously warn that, without a conscious effort to test one's assumptions in an unbiased manner, confirmatory tendencies can emerge (Dallas & Baron, 1985; Strohmer & Chiodo, 1984).

Illusory Correlations  

We have seen that our preconceptions lead us to preferentially accept and, occasionally, seek out data that support our assumptions. A final example of the impact of preconceptions on judgment involves the tendency to perceive plausible associations between events that are in fact unrelated, or are related in the direction opposite to that we detect. These fictional associations are "illusory correlations" (Chapman & Chapman, 1967), and they have been widely observed in clinical judgment.

-In a famous series of studies, Chapman and Chapman (1967, 1969) demonstrated that practicing clinicians often report noticing (and basing judgments on) associations between clients' responses to projective tests and their symptoms that have no empirical basis whatsoever. On the Draw-A-Person test, for instance, clinicians report that muscular drawings are associated with concern over one's masculinity, and unusual eyes are linked to suspicion of others; on the Rorschach, they note that various anal and feminine responses are associated with homosexuality. In reality, such associations do not exist, and the Chapmans were able to show that naive undergraduates, given projective data in which these various responses were randomly paired with the various symptoms, nevertheless reported the same plausible but fictitious correlations the clinicians say they see. Moreover, the Chapmans (1969) found that both clinicians and naive subjects generally failed to notice the real patterns in the Rorschach that link homosexuality to less stereotypical responses (e.g., seeing monsters on Card IV).

Not only are actual correlations overlooked, but subjects' erroneous perceptions of illusory correlations are often remarkably persistent. Chapman and Chapman (1967) built a huge negative correlation between symptoms and stereotypical projective responses into the data subjects were given and found that the subjects still insisted that the symptoms and responses were positively related (although the association between them did not seem as strong). Golding and Rorer (1972) repeatedly gave subjects an individual Rorschach response, asked them to predict what symptom would be associated with it, and then revealed the actual symptom, so that subjects received immediate feedback about the accuracy of their assumptions as they studied the data; strong illusory correlations were still obtained. Waller and Keeley (1978) went so far as to provide subjects with elaborate explanations of illusory correlation, complete with practice sessions in which true positive and negative correlations were demonstrated, but the subjects still reported patterns in Draw-A-Person data that were illusory (cf. Mowrey, Doherty, & Keeley, 1979; Starr & Katkin, 1969).

Results like these are certainly not unique to clinical judgments (e.g., Jennings, Amabile, & Ross, 1982). Hamilton (1981) has persuasively argued, for instance, that illusory correlations help perpetuate erroneous cultural stereotypes (e.g., "All those minorities do is collect welfare"). It simply appears that accurate assessment of covariation between events is much more complex than most people realize, necessitating consideration of far more than just the frequency with which the events jointly occur (Crocker, 1981; Nisbett & Ross, 1980).

The difficulty of these judgments allows peoples' expectations and preconceptions to have as much, if not more, influence on their judgments as do the objective data they face (Alloy & Tabachnik, 1984; Kayne & Alloy, in press). People's preconceptions can influence what they think they see in a set of data, and those preconceptions often "flourish in the face of evidence that would create grave doubts in any unbiased observer-certainly in any unbiased observer who owned a calculator, an introductory statistics text, and some conventional knowledge about how to use them" (Jennings et al., 1982, p. 227).  

Overconfidence

As clinicians and counselors form their complex judgments, their tentative evaluations gradually become more certain, and they feel more sure those evaluations are correct. However, like most people, they are probably more sure of themselves than they should be. People are generally overconfident that their beliefs are correct, thinking themselves right more often than they really are. Perhaps this is not surprising, given that biased interpretations, confirmatory hypothesis testing, and illusory correlations all combine to shield them from facts that would disconfirm their beliefs. There are, however, a handful of processes that we have not yet mentioned that contribute further to people's misplaced certainty, and in this section we consider additional reasons why we are often more sure of ourselves than is warranted.

First, some examples of overconfidence: Fischhoff, Slovic, and Lichtenstein (1977) have shown that when answering factual general-knowledge questions college students are wrong more often than they think. Fischhoff et al. confronted subjects with a wide variety of questions such as "Absinthe is (a) a liqueur or (b) a precious stone," and "Which magazine had the largest circulation in 1970, Playboy or Time?" and asked them to select an answer and to indicate how certain they were of their answers. The results showed that the subjects were consistently overconfident, choosing fewer correct answers than they had estimated they would. Moreover, they were sure enough of many of their wrong answers to wager money on them; they were poor bettors indeed, because they missed about I of every 8 questions they gave 50: 1 odds were correct!

Oskamp (1965) demonstrated the same phenomenon among clinical psychologists: He provided his professional subjects with a case study broken into four segments and asked them to predict the characteristics and behavior of the client, and to rate their confidence in their judgments, after each segment. The clinicians' predictive accuracy quickly reached a maximum, but as they read more of the case study their confidence in their judgments continued to increase. Ultimately, with the entire case study in hand, the subjects felt that 53% of their judgments were likely to be correct, but only 28% actually were. The clinicians were thus overconfident, and Oskamp concluded that "a psychologist's increasing feelings of confidence as he works through a case are not a sure sign of increasing accuracy for his conclusions" (p. 265).

Along with the influence of our preconceptions on our selection and interpretation of data, what can explain such overconfidence? We can identify four factors.

Hindsight and the "I-Knew-It-All-Along" Effect

One way we can feel as if we know more than we do is to overestimate how smart we used to be. It is difficult for people not to let their current knowledge of events contaminate their recollections of their past knowledge, so that they consistently remember themselves as smarter and less fallible than they really were. Knowing how things have turned out leads most of us to ,,remember" that the eventual outcome seemed obvious to us all along; although we could not have predicted the outcome we nevertheless feel that we "knew it all along" (Wood, 1978).

Many readers are familiar with this tendency in players of Trivial Pursuit; someone may come up with several answers to a question that are nowhere near correct, but still exclaim when the right answer is announced, "Oh, I knew that!" Several more formal demonstrations of this phenomenon have been conducted by Fischhoff (1975, 1977) and his colleagues. For instance, Fischhoff and Beyth (1975) asked subjects on the eve of President Nixon's 1972 trip to China to estimate how likely various results of the trip were. After the trip, the subjects were asked to recall their predictions, but they did not do particularly well. They remembered forecasting those events that did occur as being more probable than they had really seemed to them at the time, and similarly they believed that those events that did not transpire had never seemed very likely.

Hindsight, then, clearly distorts our judgments of what we used to think and know, generally leaving us with the impression that we were (and are) more knowledgeable than we really were. This hindsight bias, or the "I-knew-it-all-along" effect, has now been demonstrated in several studies (Goggin & Range, 1985; Leary, 1981, 1982; Slovic & Fischhoff, 1977; Wood, 1978), and it apparently is not easy to avoid; Fischhoff (1977) found that forewarning subjects by telling them about the bias did not help them correct their judgments.

Importantly, the bias affects clinicians. Using a sample of clinical psychology interns and psychiatrists, Hood (1970) found that knowledge of a client's later suicide distorted the clinicians' judgments of the suicidal intent evident in a description of the case. Knowing how things turned out, they were sure they could have predicted them all along. Thus, with the hindsight bias in place, clinical inference may seem much less difficult than it really is; no matter what people do, we may feel that their actions are not surprising and that our ability to judge others is sound.

Reconstructive Memory

Hindsight studies suggest that our memories of what we used to know are often faulty. Indeed, rather than being faithful reproductions of actual past events, many of our memories are "reconstructive." In other words, we often selectively misremember certain events, invent details, and fill in gaps, revising our recollections to produce seamless memories that fit our current circumstances (Loftus & Loftus, 1980). To the extent that we constantly edit our pasts to fit our presents, we may produce selective histories that make us seem much more clever, intuitive, and knowledgeable than we really are.

For example, past complexities may be remembered as obvious simplicities. Snyder and Uranowitz (1978) provided their subjects an extensive biography of a woman named "Betty K. " and then told them a week later either that Betty was a lesbian or that she was a heterosexual (or her sexual preference was not mentioned). The subjects' memories for the events of Betty's life were demonstrably affected by their current beliefs about her sexual orientation: Those who thought she was heterosexual remembered that she had dated often in high school, for instance, whereas those who considered her homosexual remembered that she had not had a steady boyfriend (both of which were true). The subjects literally found it hard to remember those facts that did not fit their current impressions.

Our selective memories also lead us to test our current beliefs in a biased fashion so that, again, we find support for them. Two days after reading about "Jane," Snyder and Cantor's (1979) subjects were asked to evaluate her for a job either as a librarian or as a salesperson. Remembering her introverted behaviors, half the subjects thought she would make a great librarian, whereas the other subjects, recalling her extraverted actions. thought she would be good in sales. In addition, each group of subjects considered her rather unsuited for the job for which the other group was recommending her.

Such reconstructive memory can lead to revisions of our personal histories as well. Conway and Ross (1984) showed that, in evaluating the success of a study-skills program, subjects remembered themselves as having started the program with poorer skills than they had really had. They also reported greater improvement in their skills than did a waiting-list control group, but their actual performances did not differ. In short, the subjects believed that the skills program had been quite effective when it had actually done very little. By misremembering the extent of their past problems, the subjects were Able to avoid acknowledging that they had wasted their time. Moreover, had their therapist requested an evaluation of the program, they would have no doubt praised it highly. Clients are not always the most objective judges of a therapist's efforts.  

Feedback From Clients

A great many studies have now shown that people usually accept uncritically groundless generalized descriptions of their personalities. They are particularly likely to accept vague personal evaluations that seem to be based on psychological tests and are delivered by a professional clinician (for a review, see Snyder, Shenkel, & Lowery, 1977). This gullible lack of discernment, known as the "Barnum effect," ensures that those of us in the business of giving psychological evaluations are likely to receive lots of acclaim for our prowess that, in truth, is substantially undeserved. We may receive praise from clients even when very little is accomplished. As Snyder et al. warned, "in no sense . . . can such praise be interpreted as 'validation' of either the clinician's skill or assessment procedures" (p. 113).

  Respecting Uniqueness: Making Exceptions

A final way in which we can become overconfident is to make exceptions for ourselves and for the decisions we make. Meehl (1973) described clinicians' occasional reluctance to apply actuarial data to the individual case; "After all," the argument goes, "we're concerned with the unique individual, not the group." The great fallacy of that argument, as Meehl observed, is that, although departures from normative decision rules are sometimes appropriate, a policy that allows frequent exceptions will ultimately allow many errors and is thus indefensible. A clinician who trusts his or her intuition over statistical probability, who too often believes "This is a unique case," will make too many mistakes.

In terms of overconfidence, clinicians and counselors who see themselves as unique cases to whom this chapter's inferential cautions do not apply are probably much too sure of themselves. Indeed, it is probably a further symptom of characteristic overconfidence to assume that one is personally immune to the various biases we have discussed. Unwarranted certainty and misplaced faith in our own veracity appear to be hard to avoid (Einhorn & Hogarth, 1978) and are unrelated to one's intelligence (Lichtenstein & Fischhoff, 1977); however, we might argue, as social philosophers do, that there is something intelligent in knowing what it is we do not know, and admitting our faults.  

The Behavioral Confirmation of Erroneous Inferences

None of the inferential errors we have considered would be so harmful if, eventually, truth won out and our various errors were corrected. Unfortunately, some of our most egregious errors are probably never discovered. By acting on mistaken judgments of others, we can elicit from them the behavior we expect to find, behavior that would not have occurred without our prompting. Interpersonal judgments can thus be self-fulfilling prophecies in which we get what we expect from others (which, in turn, further convinces us of our inferential skill). The manner in which our beliefs create their own realities, and the potentially therapeutic impact of this phenomenon, will be detailed in chapter 11; here we simply wish to illustrate how consequential errors in inference can be.

Snyder, Tanke, and Berscheid (1977) provided an elegant example. They recorded college males' first phone conversations with women whom they believed, on the basis of randomly manipulated photographs, were either physically attractive or unattractive. Judges' ratings of the conversations indicated that the men were much more interesting social partners when they thought they were talking to good-looking women; they were rated, for instance, as more sociable, warm, outgoing, interesting, bold, and socially adept. The men's (often erroneous) judgments of the women were clearly reflected in their behavior toward them. How did the women respond? Those who were presumed to be attractive really did sound more alluring, reacting to their obviously interested partners by sounding more appealing themselves. By comparison, the women who talked with the relatively detached men who thought they were unattractive sounded pretty drab. In both cases, the men got out of the women the behavior they expected, whether or not their judgments were sound.

Even more remarkable is the finding that the targets of our judgments may gradually come to see themselves as we do. Fazio, Effrein, and Falender (1981) examined what it is like to be the target of confirmatory hypothesis testing. As you may recall, when subjects test another's extraversion they often ask a very biased set of extraversion-eliciting questions (Snyder & Swann, 1978a). Fazio et al. found that, following such treatment, those asked the biased questions described themselves as more extraverted than they previously had. What is more, they behaved differently in subsequent interactions. When meeting a stranger, they were more likely to initiate a conversation, sit closer, talk more, and sound more extraverted on tape than those previously asked a set of "introversion" questions. Thus, we may not only get people to behave the way we think they will, but we may also slowly mold them into the people we think they are. Such is the power, and the peril, of clinical inference.

.Improving Clinical Inference

Again, no fingers are being pointed here. The inferential problems we have discussed are problems in human judgment and are not peculiar to clinical or counseling psychology. (Indeed, researchers also commit such errors.) In fact, clinical judgment may not be as treacherous as this chapter suggests; Hogarth (198 1) argued that most studies of decision making overestimate our problems by ignoring the corrective role of feedback in normal social judgments. On the other hand, the studies we have described may underestimate our problems, since they usually provided subjects with unambiguous, prepackaged data and implicitly encouraged participants to do their best (Nisbett & Ross, 1980). In any case, it seems obvious that the potential for grave errors exists, and many clinicians join us in urging caution (e.g., Turk & Salovey, 1985, in press). It is important to do more than just point out the pitfalls, however; our goal should be to improve social judgment, not merely to derogate it (Wiggins, 1982). What, then, can be done?

Education

Because there are limits to our introspection and we do tend to be overconfident, a first hurdle is convincing decision makers that potential problems even exist (Nisbett & Ross, 1980). Admitting our fallibility is far from enough, however, because recognizing inferential errors does not usually make them go away. Recall that warning people about hindsight biases (Fischhoff, 1977) and illusory correlations (Waller & Keeley, 1978), for instance, did not help correct them. Extensive training may be required to allow us to fully appreciate the extent of our judgmental biases (Fischhoff, 1982).

Turk and Salovey (in press) suggested a "bias inoculation" procedure in which people are allowed to experience judgmental errors with subsequent opportunities for analysis and feedback. The point would be to actually make these various mistakes so that they could be thoroughly understood and better detected later on. Nisbett and Ross (1980) would agree that any such training should be as vivid and concrete as possible, and they advocated the use of memorable slogans to keep corrective principles in mind: "It's an empirical question," so use the facts, not your intuition; "Beware the fundamental attribution error," and consider situations before labeling dispositions. Educational approaches can be profitable; the "process debriefing" of Ross et al. (1975), in which subjects were invited to understand belief perseverance, speculate as to its origins, and list its potential costs, was fairly effective in reducing perseverance.

Using Statistics

For complex judgments such as assessments of covariation, experience with the problem may not be enough (Waller & Keeley, 1978); statistical calculations are often required. For some judgments (detecting covariation with a four-cell contingency table, for instance) the needed calculations are simple (Nisbett & Ross, 1980), whereas for others (formulating Bayesian probabilities, for example) they are rather complex (Arkes, 1981). Nevertheless, the use of statistics and the reasoning that follows from them can cure a great many ills (Dawes, 1979; Nisbett, Krantz, Jepson, & Fong, 1982).

Arkes (1981) provides a fascinating example of how an explicitly statistical approach can salvage wrongful intuition. Assume that you are tempted to make a diagnosis of multiple personality because the chances that a person who was not a multiple personality would present you with the symptoms you have observed are, say, I in 100. Such odds do not mean your diagnosis has a 99% chance of being correct. Because multiple personalities are exceedingly rare (let's assume, generously, that 1 in 100,000 persons qualifies), your chances of encountering one are quite remote, and the actual probability that your diagnosis is correct is a mere .1%. If that seems astonishing, you can appreciate the value of statistics in setting you straight.

Considering Alternatives

Several types of judgments appear to be improved by careful consideration of why we might be wrong. More specifically, inventing plausible explanations for opposing points of view seems to help reduce belief perseverance, overconfidence, and biased interpretation of data. For instance, in a hindsight study, Slovic and Fischhoff (1977) found that asking subjects to explain explicitly how alternative outcomes could have occurred reduced the hindsight tendency to assume a particular outcome was inevitable. Similarly, getting subjects to list specific reasons why they might be right and why they might be wrong in an overconfidence study reduced their tendency to overestimate their knowledge (Koriat, Lichtenstein, & Fischhoff, 1980). Misplaced beliefs are also less likely to persevere when subjects imagine contradictory causal possibilities (Anderson, 1982). In general, it appears that the more salient or cognitively "available" a given belief is, the more plausible it is (Anderson et al., 1985); intentionally creating explanations for alternative beliefs can thus help us evaluate a given belief more impartially.

Being Explicit

The more information we have to process (Lueger & Petzel, 1979) and the more we have to rely on our faulty memories (Arkes, 1981), the less accurate our judgments will be. Dawes (1982) encouraged being explicit in our decision making, listing all logical possibilities and systematically estimating as carefully as possible the likelihood of each. Turk and Salovey (in press) echoed that suggestion, urging us to "de-automize" our judgmental processes so that we become more aware of our own thinking. Critically evaluating our judgments in a step-by-step fashion through self-interrogation and -thinKing aloud may help block those fuzzy, intuitive leaps where biases lead us astray. In addition, accountability, or the feeling that we will have to justify our decisions to others, may be valuable; Tetlock (1983, 1985) has shown that warning subjects of their accountability reduces anchoring effects and personalistic biases (cf. Harkness, DeBono, & Borgida, 1985).

Delaying Judgment

Finally, there may be some wisdom in putting off our judgments as long as possible. We have seen that, as soon as first impressions are formed, they begin to influence our interpretation of subsequent data. It may improve our judgments, then, to resist the tendency to arrive at premature conclusions. In fact, Dailey (1952) showed that the final judgments of subjects who began to predict another person's behavior halfway through his self-description were less accurate than those of subjects who made no predictions until the end. Delaying judgment as long as possible may be a desirable strategy.

Withholding judgment is probably easier said than done, however, and is likely to require conscious effort. Indeed, all of the strategies considered in this section ask us to actively attend to the manner in which we arrive at our judgments of others. Our inferential habits are frequently faulty, and improving them will no doubt require careful vigilance. Acknowledging those faults is the place to begin.

Conclusions

Clinical inference is open to the same biases, distortions, and errors that cloud all social judgment, likely leading to a misplaced faith in clinicians' judgmental abilities. We are prone to ignore how others' circumstances influence their behavior, and we tend to seek out and preferentially value information that supports our existing beliefs. If no such information exists, we may perceive it anyway, and we may even cling to beliefs for which there is no longer any support whatsoever. We are generally unaware of these inferential pitfalls because it is hard for us to recognize how data influence our judgments. The net result is that we are more sure of ourselves than we should be, often trusting our intuitions over statistical probabilities. We are probably not unable to change most of these bad habits, but we need first to acknowledge our fallibility and then be ready to go to work.