2.1. The early research on stereotypes

One of the first arguments that our perceptions are not necessarily strongly linked to objective reality came from a journalist. In a broad-ranging book called Public Opinion, Walter Lippmann (Reference Lippmann1922/1991) touched on stereotypes – and defined them in such a way as to color generations of social scientists’ views of stereotypes. Lippmann suggested that to understand the world in its full complexity is an impossible task. So people simplify and reduce the overwhelming amount of information they receive. Stereotypes, for Lippmann, arose out of this need for simplicity. He believed that people's beliefs about groups were essentially “pictures in the head.”

A “picture in the head” is a static, two-dimensional representation of a four-dimensional stimulus (most real-world stimuli have width, length, and depth, and also change over time). A picture is rigid, fixed, and unchanging. It is over-simplified and can never capture the full complexity of life for even one member of any group. This should sound familiar – it constitutes the working definition of stereotypes that many people, including many social scientists, still hold today. Thus, it constitutes one of the earliest perspectives suggesting that people's social beliefs may not be fully in touch with social reality.

Social psychologists ran with these ideas. Katz and Braly (Reference Katz and Braly1933) concluded that the high levels of agreement they observed regarding national, racial, and ethnic groups could not possibly reflect personal experience and instead most likely reflected the shared expectations and biases of the perceiver. This analysis was flawed because agreement per se is not evidence of inaccuracy (often, though not always, it reflects accuracy – e.g., Funder Reference Funder1987). In a similarly flawed manner, LaPiere (Reference LaPiere1936) interpreted his empirical results as demonstrating that stereotypes were inaccurate rationalizations of antipathy towards outgroups, even though (except for some anecdotes) he did not assess people's stereotypes.

Gordon W. Allport (Reference Allport1954b), in perhaps the most influential social psychological book written about stereotypes and prejudice, distinguished between, on the one hand, rational and flexible beliefs about groups, and on the other, stereotypes. Long ignored in many citations to G. W. Allport (Reference Allport1954/1979) is the fact that he clearly acknowledged the existence of rational and flexible beliefs about groups. He merely did not consider such beliefs to be stereotypes. For G. W. Allport, stereotypes are faulty exaggerations. All-or-none beliefs, such as “all Turks are cruel,” are stereotypes that are clearly inaccurate, overgeneralized, and irrational, because there are virtually no social groups whose individual members universally share some set of attributes. G. W. Allport also characterized stereotypes as unjustifiably resistant to change, steeped in prejudice, and leading to all sorts of errors and biases in social perception, and concluded they were a major contributor to social injustice. Overall, therefore, the early research on stereotypes helped set the stage for social psychology's later emphasis on error and bias.

2.2. Early social perception research

4.1. Social psychology falls in love with self-fulfilling prophecies

Many social psychologists were able to tell compelling stories about the results of Pygmalion in particular, and the power of self-fulfilling prophecies more generally (e.g., Darley & Fazio Reference Darley and Fazio1980; Gilbert Reference Gilbert and Tesser1995; Jones Reference Jones1986; Jost & Kruglanski Reference Jost and Kruglanski2002). Many studies yielded results seeming to support this perspective. Self-fulfilling prophecies occur, in part, because expectations lead perceivers to treat high expectancy targets differently than they treat low expectancy targets, and this differential treatment evokes expectancy-confirming target behavior. One classic pair of studies demonstrated this process: White interviewers’ nonverbal behavior discriminated against Black interviewees, and when White interviewees were subjected to the same behavior, their interview performance declined (Word et al. Reference Word, Zanna and Cooper1974). Similarly, teachers were at least sometimes more supportive of White students than of Black students (Rubovitz & Maehr Reference Rubovitz and Maehr1973; Taylor Reference Taylor1979). When women believed an attractive male interviewer was sexist, they presented themselves as more traditional, scored lower on an anagrams test, wore more makeup and accessories, and talked less (von Baeyer et al. Reference von Baeyer, Sherk and Zanna1981; Zanna & Pack Reference Zanna and Pack1975). An observational study of children in kindergarten through second grade concluded that teachers’ social class-based expectations created a “caste system” advantaging middle class students over lower class students (Rist Reference Rist1970).

One of the most influential and highly-cited classics of this era demonstrated the self-fulfilling effects of the physical attractiveness stereotype (Snyder et al. Reference Snyder, Tanke and Berscheid1977). Men were misled (through photographs) to believe a woman in another room was either attractive or unattractive. Not only did they behave in a friendlier and warmer manner to the women believed to be attractive, those women reciprocated with warmer and friendlier behavior themselves. Thus, originally false beliefs about the social skill of the attractive became (self-)fulfilled.

Self-fulfilling prophecies were not restricted to stereotypes. Competitive people saw the world as competitive and evoked competitive behavior even from people predisposed to be cooperative (Kelley & Stahelski Reference Kelley and Stahelski1970). People who falsely believed others are hostile evoked hostile behavior (Snyder & Swann Reference Snyder and Swann1978a). Israeli military instructors evoked expectancy-confirming performance from military trainees (Eden & Shani Reference Eden and Shani1982). Self-fulfilling prophecies seemed to be everywhere psychologists turned.

4.2. Expectancy-confirming biases

Self-fulfilling prophecies are not the only effect of expectations. Interpersonal expectancies also bias judgments of social reality. The extraordinary power of stereotypes regarding demographic categories, occupation, roles, mental diagnoses and many other social categories to bias judgments is a common theme in social psychological scholarship. For example, in one classic study, after viewing a fourth grade girl take a test, perceivers judged her to have performed more highly and to be smarter if they believed she was from a higher rather than lower social class background (Darley & Gross Reference Darley and Gross1983). Yet another concluded that mental illness labels (e.g., “schizophrenia”) led to such powerful expectancy biases that it became impossible to distinguish the sane from the insane (Rosenhan Reference Rosenhan1973). People constructed false “memories” about the supposed facts of a woman's life based on their stereotypes of whether she was lesbian or heterosexual (Snyder & Uranowitz Reference Snyder and Uranowitz1978). Similar findings obtained for stereotypes based on race, gender, and many other categories. In this context, it is perhaps unsurprising that one major review declared stereotypes to be the “default” basis of person perception (Fiske & Neuberg Reference Fiske and Neuberg1990).

Such biases were not restricted to stereotypes, and occurred for expectations regarding intro/extraversion, friendliness, and intelligence (e.g., Kulik Reference Kulik1983; Rothbart et al. Reference Rothbart, Evans and Fulero1979; Williams Reference Williams1976). Furthermore, such biases also infected social information-seeking. In an influential series of studies, Snyder and Swann (Reference Snyder and Swann1978b) found that not only do people systematically seek information that confirms their hypotheses, they constrain targets’ ability to do much other than confirm the initially erroneous expectation.

The extent to which expectations influence, change, and color (or, for stereotypes, taint) our interactions with and perceptions of other people seemed to be nothing short of stunning. The social psychological enthusiasm for expectancy-induced biases was at least comparable to that expressed for self-fulfilling prophecies. Here are some quotes representative of a widespread consensus in social psychology:

• Owing to a variety of cognitive biases, a perceiver's initial expectancies for a target are apt to be maintained, regardless of whether the target's behavior confirms, disconfirms, or is ambiguous with respect to the perceiver's expectancy (cited in Deaux & Major Reference Deaux and Major1987, p. 381)

• Specifically, all of these processes are biased in the direction of maintaining the preexisting belief system, that is, the very stereotype that initiated these biasing mechanisms. (Hamilton et al. Reference Hamilton, Sherman and Ruvolo1990, p. 39)

• The thrust of dozens of experiments on the self-fulfilling prophecy and expectancy-confirmation processes, for example, is that erroneous impressions tend to be perpetuated rather than supplanted, because of the impressive extent to which people see what they want to see and act as others want them to act. (Jost & Kruglanski Reference Jost and Kruglanski2002, pp. 172–73)

• A particularly pernicious example of self-fulfilling beliefs and expectations, and the one most studied by social psychologists, is that of stereotypes and other negative beliefs about particular groups of people. Some of these effects are obvious, although no less important for their obviousness. If it is widely believed that the members of some group disproportionately possess some virtue or vice relevant to academic or on-the-job performance, one is likely (in the absence of specific legal or social sanctions) to make school admission or hiring decisions accordingly – and in so doing to deprive or privilege group members in terms of opportunities to nurture their talents, acquire credentials, or otherwise succeed or fail in accord with the beliefs and expectations that dictated their life chances. (Ross et al. Reference Ross, Lepper, Ward, Fiske, Gilbert and Lindzey2010, p. 30, emphasis mine).

5.1. Rist (Reference Rist1970)

Rist (Reference Rist1970) conducted an observational study of kindergarten through second grade, and concluded that teachers’ social-class–based expectations were so powerfully self-fulfilling that they created a “caste system” serving to maintain the advantages of middle-class students. According to Google Scholar, this study has been cited over 1600 times. It is quite striking, therefore, to discover that it actually provided no evidence of self-fulfilling prophecies whatsoever. Rist (Reference Rist1970) reported only a single piece of evidence regarding student achievement, and that was in a footnote (Note 5, p. 443). That footnote reported that, at the end of the year, there were no differences in the IQ scores among the kindergarten students who were targets of high or low social class teacher expectations. In other words, his only quantitative assessment of achievement provided no evidence that teacher expectations produced changes in student achievement.

Rist (Reference Rist1970) did provide a wealth of information about teacher treatment of students. In short, the teacher assigned the students to tables based on their social class, and proceeded to direct most of her attention to the middle-class students. Rist's (Reference Rist1970) “caste system” conclusion was based on his observation that this table assignment pattern continued partially intact through second grade. However, it was only partially intact, and, indeed, there was actually considerable movement among students from kindergarten to first grade and again from first grade to second grade. If there was a “caste system,” it was a strikingly fluid one that produced no observed impact on students’ achievement by the only measure of such impact reported.

5.2. Rosenhan (Reference Rosenhan1973)

Rosenhan (Reference Rosenhan1973, cited over 2,000 times) tested – and claimed to confirm – one of the most audacious hypotheses in all of psychology: that the insane are indistinguishable from the sane. This is so extreme that readers might naturally wonder if I am setting up some sort of straw argument by overstating Rosenhan's claims. Here is what Rosenhan (Reference Rosenhan1973) himself wrote in his paper:

• If sanity and insanity exist, how shall we know them? The question is neither capricious nor itself insane. However much we may be personally convinced that we can tell the normal from the abnormal, the evidence is simply not compelling. (opening sentences, p. 250).

• Based in part on theoretical and anthropological considerations, but also on philosophical, legal, and therapeutic ones, the view has grown that psychological categorization of mental illness is useless at best and downright harmful, misleading, and pejorative at worst. (p. 251)

• Psychiatric diagnoses, in this view, are in the minds of the observers and are not valid summaries of characteristics displayed by the observed. (p. 251)

• We now know we cannot distinguish insanity from sanity. (p. 257)

I have not overstated Rosenhan's claims; instead, his claims themselves are vast overstatements. To understand how and why, it is necessary to first summarize his report. He had eight people (“pseudopatients”) with no prior histories of mental illness admitted to psychiatric hospitals in order to see if the professional staff could identify them as sane. To get admitted, all eight complained that they had been hearing voices. Upon admission, they ceased displaying all intentionally false expressions of disturbed behavior and they did not intentionally alter any other aspect of their life history.

They were kept institutionalized for an average of 19 days. When they were released, none were identified as sane; all were released with a diagnosis of “schizophrenia in remission.” Rosenhan (Reference Rosenhan1973) also provided qualitative examples of staff interpreting normal behavior as evidence of pathology (e.g., pacing halls out of boredom was interpreted as nervousness). Thus, Rosenhan concluded that the sane were indistinguishable from the insane because diagnosis pervasively colored the institutional staff members’ interpretations of the pseudo-patients’ behavior and life histories.

However, there is actually far more evidence of reasonable, rational, and valid judgment on the part of the doctors and staff than first appears. How the pseudopatients initially got themselves admitted should give some reason for pause. They were admitted complaining of auditory hallucinations. Regularly hearing voices saying things like “thud,” “empty,” and “hollow” (what they claimed to be hearing) is not remotely normal. Therefore, an initial diagnosis of some form of psychosis does not seem to reflect gross distortion on the part of the psychiatric staff.

How rigidly resistant to change were the doctors’ and staffs’ expectations? Rosenhan's (Reference Rosenhan1973) interpretation was that they were highly rigid. After all, none were diagnosed as sane. But let's focus on Rosenhan's actual results, rather than his interpretations. First, the average hospital stay was 19 days, and most were kept under two weeks. How this reflects rigidity was never articulated.

How about the diagnosis of “schizophrenia in remission”? Rosenhan argued that it showed that there was nothing these completely sane pseudopatients could do to convince the doctors that they were really sane. However, “schizophrenia in remission,” at that time, meant “the patient is showing no current signs of schizophrenia” (Spitzer Reference Spitzer1975; Spitzer et al. Reference Spitzer, Andreasen and Endicott1978). Thus, in Rosenhan's own data, and in contrast to his conclusions, the staff did indeed recognize that the pseudopatients were behaving in a manner devoid of evidence of psychosis.

Rosenhan (Reference Rosenhan1973) also reported a follow-up study in which staff at institutions were informed to be on the lookout for pseudopatients. Because none were actually sent, any identification of a person as a pseudopatient is an error, and all such errors were interpreted by Rosenhan as supporting his extraordinary “the sane are indistinguishable from the insane” hypothesis. How many such errors did the psychiatrists make? Although Rosenhan (Reference Rosenhan1973) did not report the data necessary to compute this figure exactly, it can be plausibly estimated as no higher than 6%, and probably considerably lower.

To keep the math simple, let's assume there were only two psychiatrists and we interpret “at least one” to mean “half” (the result is the same if we take half of two, or half of 100). If it was more than half, Rosenhan (Reference Rosenhan1973) probably would have stated so. Two psychiatrists by 193 patients is 386 judgments. 21 (judged fakers)/386 = 6%. 6% errors is the same as 94% accuracy.

Given the possibility that 6% of those admitted were, in fact, not suffering from psychopathology, even 6% may overstate the actual error rate. Any error is, well, an error – but these results are not exactly a testament to the extraordinary biasing power of psychiatric diagnoses and expectations. Indeed, the entire study – its results demonstrating high accuracy and small but real bias, and the manner in which its evidence of bias was so greatly overstated – is consistent with the tripartite pattern I first used to describe Hastorf and Cantril (Reference Hastorf and Cantril1954): (1) Bias is real but small; (2) accuracy is very high; and (3) the conclusions greatly overstated the power and pervasiveness of bias.

5.3. The replication failures

Many classic studies in the expectancy-confirmation literature have proven difficult to replicate. Attempts to replicate Snyder et al.’s (1977) self-fulfilling physical attractiveness stereotype study, Darley and Gross's (Reference Darley and Gross1983) social class stereotype bias study, and Snyder and Uranowitz's (Reference Snyder and Uranowitz1978) stereotype-based reconstructive memory studies all failed (Andersen & Bem Reference Andersen and Bem1981; Baron et al. Reference Baron, Albright and Malloy1995; Belezza & Bower 1981). In contrast to Rist's (Reference Rist1970) conclusions, social class biases found in large-scale, quantitative studies of teacher expectations have consistently been nonexistent (Jussim et al. Reference Jussim, Eccles and Madon1996; Madon et al. Reference Madon, Jussim, Keiper, Eccles, Smith and Palumbo1998; Williams Reference Williams1976).

Several lines of research followed up on the Snyder and Swann (Reference Snyder and Swann1978b) study finding that people seek to confirm their social expectations by asking people leading questions that essentially remove from targets the opportunity to do anything except provide confirmatory answers. These have generally focused, not on attempts at exact replication, but on the validity of Snyder and Swann's (Reference Snyder and Swann1978b) conclusion that people are heavily biased towards confirming their social expectations. Snyder and Swann (Reference Snyder and Swann1978b) only gave people the opportunity to ask leading questions. Numerous follow-up studies, however, recognized this limitation and addressed it either by allowing people to make up their own questions or to select from both leading and diagnostic questions (e.g., Devine et al. Reference Devine, Hirt and Gehrke1990; Trope & Bassok Reference Trope and Bassok1982; Reference Trope and Bassok1983). When left to their own devices, or given adequate choice, people overwhelmingly ask diagnostic questions, and they almost never ask the type of leading questions found in Snyder and Swann (Reference Snyder and Swann1978b). There does appear to be a slight tendency to ask questions to which a “yes” answer will confirm perceivers’ expectations, and combined with a slight tendency on the part of targets to acquiesce, social hypothesis-testing may indeed be slightly biased in favor of confirming perceivers’ hypotheses (Zuckerman et al. Reference Zuckerman, Knee, Hodgins and Miyake1996).

Nonetheless, Snyder and Swann (Reference Snyder and Swann1978b) is cited more than all these other studies put together, and the most common pattern is to cite it as demonstrating biased social hypothesis testing, without citing any of the research showing that people generally ask diagnostic questions (e.g. Deaux & Major Reference Deaux and Major1987; Miller & Turnbull Reference Miller and Turnbull1986). Similar citation patterns characterize much of the expectancy literature. Dramatic demonstrations of bias or self-fulfilling prophecy typically receive abundant attention whereas the failures to replicate that finding, and demonstrations of accuracy and rationality are largely overlooked.

This, then, is another route demonstrating the tripartite conclusion – bias is real but generally small; people are mostly accurate and rational; results demonstrating bias are overstated. In these cases, however, it is not necessarily the original researchers who overstate the result. Rather, the overstatement occurs because attention (citations) primarily focus on, and conclusions primarily emphasize, results of one dramatic (though flawed) demonstration of bias, and the more abundant and generally higher quality research demonstrating small (or irreplicable) bias and high accuracy/rationality is typically overlooked or ignored.

5.4. Quest for the powerful self-fulfilling prophecy

Having discovered this tripartite pattern repeated over and over, it seemed important to try to discover if there were any conditions under which truly powerful self-fulfilling prophecies in the classroom occurred. Thus, we embarked on a quest to systematically search for conditions under which large expectancy effects occurred (Jussim et al. Reference Jussim, Eccles and Madon1996; Madon et al. Reference Madon, Jussim and Eccles1997). Using a data set including over 100 teachers and over 1,000 students, we found a slew of powerful self-fulfilling prophecies, with effect sizes (standardized regression coefficients) ranging from about .40 to about .60. Powerful self-fulfilling prophecies occurred among:

1. 1. African-American students

2. 2. Students from lower SES backgrounds (regardless of ethnicity)

3. 3. Students with histories of low prior achievement who were from lower SES backgrounds (these.6 effects are among the most powerful ever found in social psychology)

4. 4. Students with histories of low achievement who were the target of high expectations. High expectations uplifted such students more than they uplifted high achievers, and more than low expectations harmed achievement.

Although powerful self-fulfilling prophecies are the exception rather than the rule, they systematically occurred among students from stigmatized social backgrounds. Interestingly, in our data, they seemed to ameliorate more than cause social inequalities (uplifting students with histories of low achievement).

5.5. Do self-fulfilling prophecies accumulate or dissipate?

In light of findings that expectancy-based biases and self-fulfilling prophecies are occasionally large but generally quite modest, researchers seeking to maintain a view of self-fulfilling prophecies as powerful and pervasive contributors to social problems needed to generate new arguments for doing so. The seemingly most compelling of these was that self-fulfilling prophecies may accumulate over time and/or over multiple perceivers (e.g., Claire & Fiske Reference Claire, Fiske, Sedikedes, Schopler and Insko1998; Fiske Reference Fiske, Gilbert, Fiske and Lindzey1998). The logic of accumulation is straightforward:

1. 1. Small effects are typically obtained in both short-term laboratory studies of self-fulfilling prophecies and teacher expectation studies conducted over a school year.

2. 2. Although small in such contexts, many targets may be subjected to the same or similar erroneous expectations over and over again. Social stereotypes, widely assumed to be widely shared and erroneous, are often presented as an obvious reason to predict that targets from stigmatized groups will be subjected to repeated self-fulfilling prophecies from multiple perceivers over long periods of time. Thus, effects of expectancies on any particular target are likely to be much higher than demonstrated in any particular study.

There are, however, also compelling reasons to predict that, rather than accumulating, self-fulfilling prophecies will dissipate, including regression to the mean, self-verification (Swann & Ely Reference Swann and Ely1984), and accuracy (see the book for a full discussion of each). Thus, regardless of how “compelling” the accumulation argument may seem at first glance, the issue is an empirical one. Do self-fulfilling prophecies accumulate?

Every teacher expectation study that has assessed whether self-fulfilling effects that occurred in one year accumulate over time has found the exact opposite: They dissipate over time. Self-fulfilling prophecies dissipated in the original Rosenthal and Jacobson (Reference Rosenthal and Jacobson1968) study, where the IQ difference between bloomers and controls was about four points in the first year, and under three points in the second year. Rist (Reference Rist1970) is often cited as evidence of accumulation, but he found neither accumulation across years nor self-fulfilling prophecy. West and Anderson (Reference West and Anderson1976) followed 3,000 students through high school, and found that teacher expectation effects declined from .12 the first year to .06 in the final year (standardized regression coefficients). We also tested accumulation over five to six years in math (from sixth or seventh grade through twelfth grade), and, instead, found dissipation (Smith et al. Reference Smith, Jussim and Eccles1999). The typically modest self-fulfilling prophecies found in sixth and seventh grade (.10, .16, respectively) declined to 0 and .09, respectively, by twelfth grade. Dissipation has also been found when research has followed students from first through fifth grade, in both reading and math (Hinnant et al. Reference Hinnant, O'Brien and Ghazarian2009).

Compelling stories can and have been told about how the accumulation of self-fulfilling prophecy upon self-fulfilling prophecy constitutes a major mechanism by which social stereotypes confirm themselves and maintain unjustified systems of oppression and status (e.g., Claire & Fiske Reference Claire, Fiske, Sedikedes, Schopler and Insko1998; Darley & Fazio Reference Darley and Fazio1980; Snyder Reference Snyder1984; Weinstein et al. Reference Weinstein, Gregory and Strambler2004) – typically without consideration or review of the considerable evidence indicating that self-fulfilling prophecies dissipate. Nonetheless, there is currently no clear evidence supporting such an analysis, and a great deal of evidence disconfirming it.

5.6. Conclusion: The less than awesome power of expectations to create self-fulfilling prophecies, and bias perception, judgment, and memory

Do expectations lead to self-fulfilling prophecies and biases in judgment, perception, and memory? Yes, at least sometimes. But even the early blush of research on expectancy effects – the era filled with “classics” in the study of self-fulfilling prophecies and bias – never showed that such effects are, on average, inevitable, powerful, or as pervasive as often claimed. Such effects are not only relatively small, on average, but they tend to be quite fragile, in the sense that seemingly small changes in experimental procedure, geography, type of dependent variable, or researcher often seem to lead such biases to mostly or completely evaporate, and sometimes, to completely reverse.

Just because bias tends to be small, however, does not necessarily mean that accuracy tends to be high. Evaluating the accuracy question is simultaneously very simple and dauntingly complex. Therefore, the complexities of studying accuracy are summarized next.

6.1. Political objections

Some have criticized accuracy research because it can be used to justify inequality. For example, Stangor (Reference Stangor, Lee, Jussim and McCauley1995) explains why stereotype accuracy is not worthwhile to study, in part this way: “As scientists concerned with improving the social condition, we must be wary of arguments that can be used to justify the use of stereotypes.” And then later in the same paragraph: “[…] we cannot allow a bigot to use his or her stereotypes, even if those beliefs seem to them to be accurate” (Stangor Reference Stangor, Lee, Jussim and McCauley1995, pp. 288–89). This is an explicitly political criticism of accuracy research. It refers quite bluntly to political power rather than science (“cannot allow a bigot”). People in power make decisions about what is allowable, whereas, presumably, scientific research does not.

Opposition to accuracy research on political grounds has a kernel of truth. Accuracy cannot explain social problems. Demonstrating that people's sex stereotypes are accurate (Swim Reference Swim1994) or that people's racial stereotypes are accurate (McCauley & Stitt Reference McCauley and Stitt1978) does nothing to alleviate or explain injustices associated with sexism or racism. Worse, demonstrating social perceptual accuracy can be viewed as not merely documenting high acumen in perceiving individual and group differences, but as implicitly reifying and justifying those differences. To characterize a belief that some kid is not too bright, or is a klutz on the basketball court, or is socially inept as “accurate” has a feel of “blaming the victim.” Blaming the victim is a bad thing to do – it means we have callously joined the perpetrators of injustice.

Nonetheless, this argument fails to threaten accuracy research. First, scientific conclusions should be based on empirical evidence, and not be subject to political litmus tests. Second, it cannot be logically possible to reach conclusions about inaccuracy – and the four-decades–long emphasis on error and bias in social cognition provides ample evidence that social psychologists do indeed often wish to reach conclusions about inaccuracy – unless we can also reach conclusions about accuracy. Third, if we think we are curing a social problem (e.g., inequality) by treating the wrong disease (the supposedly inaccurate expectations whose accuracy social psychologists rarely assess and which, therefore, may be far more accurate than many seem to assume) we may not get very far.

Furthermore, there will be no way to assess our success at leading people to adopt more accurate beliefs, unless we have techniques for assessing accuracy. By understanding what leads people astray, and what leads them to accurate judgments, we will be much more capable of harnessing those factors that lead to accurate judgments, and therefore, reduce social problems resulting from inaccurate beliefs. Thus, even on the political grounds of aspiring to reduce inequality, political objections fail to provide a serious scientific threat to the study of accuracy.

6.2. Theoretical objections

Not all objections to accuracy research are political. Next, therefore, I consider some of the most common substantive and theoretical objections to accuracy research.

6.3. Criteria and construct validity

8.1. The logical incoherence of defining stereotypes as inaccurate

A claim that all beliefs about all groups are inaccurate is logically incoherent. It would mean that:

(1) Believing that two groups differ is inaccurate; and (2) believing two groups do not differ is inaccurate. Both (1) and (2) are not simultaneously possible, so we can reject any claim that all beliefs about groups are inaccurate.

If stereotypes are the subset of beliefs about groups that are inaccurate, then only inaccurate beliefs about groups can be considered stereotypes. Accurate beliefs about groups have been defined away as not stereotypes. This has the (probably unintended) effect of defining away nearly all existing research on stereotypes. Why? Because vanishingly few studies of stereotypes have actually first demonstrated that the beliefs about groups under study are inaccurate. Holding social psychology to this interpretation of “stereotypes are inaccurate” means concluding that decades of research framed as addressing stereotypes really has not done so. There would be no studies of the role of stereotypes in expectancy effects, self-fulfilling prophecies, person perception, subtyping, memory, and the like.

There are additional logical problems with defining stereotypes as inaccurate. No scholarship that has done so has also identified the point at which a belief crosses over from being an “accurate” belief about a group, to being a “stereotype.” Absent a standard for (in)accuracy, this means that we cannot know whether any belief is a (defined as inaccurate) stereotype. Similarly, if one claims that accuracy cannot or should not be assessed, or that existing research fails to validly assess accuracy (Fiske Reference Fiske, Gilbert, Fiske and Lindzey1998; Reference Fiske2004; Stangor Reference Stangor, Lee, Jussim and McCauley1995), one has dismissed all evidence that bears on accuracy and therefore precluded one's self from making any statements about stereotypes’ (in)accuracy. In summary, defining stereotypes as inaccurate is severely problematic no matter what the definer means. Any scientist who wishes to maintain such a definition needs to precisely articulate how each of these forms of logical incoherence have been overcome.

8.2. A viable, logically coherent definition

I concur with the minority of scientists who have left inaccuracy out of the definition of stereotype (e.g., Ashmore & Del Boca Reference Ashmore, Del Boca and Hamilton1981; Judd & Park Reference Judd and Park1993; Ryan Reference Ryan2002), and who have generally defined stereotypes as beliefs about the attributes of social groups. This allows for many possibilities not explicitly stated. Stereotypes may or may not:

• • be accurate and rational

• • be widely shared

• • be conscious be rigid

• • exaggerate group differences

• • assume group differences are essential or biological

• • cause or reflect prejudice and discrimination

• • cause biases and self-fulfilling prophecies

• • play a major role in some social problems.

This definition retrieves accuracy from premature foreclosure by definition and turns it into a scientific empirical question. How well do people's beliefs about groups correspond to what those groups are actually like?

8.3. The rigorous assessments of stereotype (in)accuracy

To be included here, empirical studies assessing the accuracy of stereotypes needed to meet two major criteria. First, they had to relate perceivers’ beliefs about a target group with some measure of what that group was actually like. This may seem obvious, but the social psychological discourse on stereotypes has often drawn conclusions about the inaccurate or unjustified nature of stereotypes based entirely on evidence addressing social cognitive processes – illusory correlations, priming, expectancy effects, attributional patterns, and so forth. Such research, although important on its merits, does not directly address accuracy, which can only be done by comparing beliefs about groups to criteria regarding those group's characteristics.

Second, studies needed to use an appropriate target group. If the stereotype is of “American women,” the target group should be a representative sample of American women; it cannot be a convenience sample (Judd & Park Reference Judd and Park1993). Studies that met both of these criteria were included; those that did not were excluded.

8.4. Four types of stereotype (in)accuracy

Accuracy is often a multidimensional construct (e.g., Judd & Park Reference Judd and Park1993; Kenny Reference Kenny1994), as can be readily illustrated with a simple example. Consider Fred, judging the average height of male Americans, Columbians, and Dutch. Fred estimates the average heights, respectively, as 5′8″, 5′5″, and 5′10″. Let's say the real average heights are, respectively, 5′10″, 5′7″, and 6′0″. In absolute terms, Fred is inaccurate – he consistently underestimates height by two inches. However, in relative terms, Fred is perfectly accurate – his estimates correlate 1.0 with the actual heights. Although Fred has a downward bias in perceiving the absolute heights among men in the different countries, he is superb at perceiving the relative height differences.

Discrepancy from perfection refers to how close people's beliefs about groups are to those groups’ actual mean characteristics on criteria. These are assessed with discrepancy scores. Correspondence with differences refers to how well people detect either variations between or within groups on some set of attributes. These are assessed with correlations between beliefs and criteria. Personal stereotypes are the beliefs about groups held by a particular individual. Consensual stereotypes are the overall, or average, beliefs about a group held by some group of perceivers. This creates four types of stereotype (in)accuracy: Personal discrepancies (how discrepant a single person's stereotypes are from a criterion); consensual discrepancies (how discrepant a sample's or group's mean stereotypes are from a criterion); personal correspondence (the correlation of a single person's stereotypes with criteria); and consensual correspondence (the correlation of a sample's or group's stereotypes with criteria).

8.5. What is a reasonable standard for characterizing a stereotypic belief as “accurate”?

Discrepancies from perfection (discrepancy scores) and correspondence with real differences (correlations) capture different but important aspects of accuracy. Because both can and have been used to assess accuracy, there needs to be two separate standards for characterizing a belief as accurate – one for discrepancy scores, and another for correlations.

8.6. Pervasive stereotype accuracy

I use the term “pervasive stereotype accuracy” here to refer to the widespread evidence of at least some accuracy, and sometimes quite high accuracy, found in nearly every study that has assessed stereotype accuracy. Nonetheless, there is no evidence that stereotypes are perfectly accurate. Furthermore, even within a single study, accuracy levels may vary, not just across judgments or perceivers, but, because accuracy is a multi-faceted construct, across the four types of accuracy described previously. The few studies of highly inaccurate stereotypes are also reviewed. Nonetheless, the evidence of pervasive stereotype accuracy is inconsistent with virtually all perspectives defining stereotypes as inaccurate, or emphasizing their inaccuracy.

8.6.1. Ethnic and racial stereotypes

Table 2 summarizes the results of all studies assessing the accuracy of racial/ethnic stereotypes that met the criteria for inclusion. Consensual discrepancies are mostly accurate or near misses. For example, in McCauley and Stitt (Reference McCauley and Stitt1978), out of 70 judgments about Americans (in general) and African-Americans, 34 were accurate, and another 30 were near-misses. Ashton and Esses (Reference Ashton and Esses1999) found a similar pattern in the consensual ethnic stereotypes held by Canadian college students’ – judgments about the academic achievement of eight of nine ethnic groups were accurate. Ryan (Reference Ryan1996) found evidence of both accuracy and inaccuracy in African-American and White college students’ consensual racial stereotypes over all six perceiver group-target group combinations (African-American and White perceivers making judgments about African-Americans, Whites, and their differences): 34 judgments were accurate, 20 were near misses, and 48 were inaccurate.

Table 2. The Accuracy of Racial and Ethnic Stereotypes

Ryan's (Reference Ryan1996) results refer to her stereotypicality results, not her dispersion results. Exaggeration means the perceived differences between groups exceeded the group differences on the criteria. Underestimation means the perceived differences between groups was smaller than the group differences on the criteria. Individual correlations involve computing, for each individual perceiver, the correlation between their judgments (stereotypes) and the criterion. Studies performing this analysis typically report the average of those correlations. Aggregate correlations refer to the correlation between the overall average perceived difference between the groups (for the whole sample) and the group difference on the criteria.

Table 2 Notes:

1. Except where otherwise stated, all discrepancy results occur at the consensual level. Accuracy means within 10% of the real percentage or within .25 of a standard deviation. Exaggeration means the perceived differences between groups exceeded the group differences on the criteria. Underestimation means the perceived differences between groups was smaller than the group differences on the criteria. Except where otherwise noted, only one word is entered in this column when one pattern (e.g., “accuracy”) occurred for a majority of results reported. When there was no majority, the top two results (most frequent first) are reported here except where otherwise noted.

2. If the study reported more than one individual level (average) correlation, their correlations were averaged to give an overall sense of the degree of accuracy.

3. These correlations do not appear in the original article, but are computable from data that was reported.

4. For each group of perceivers, the first correlation is the correspondence between their judgments and the self-reports of their own groups; the second correlation is the correspondence between their judgments and the self-reports of the other group; and the third correlation is the correspondence between the perceived difference between the groups and the difference in the self-reports of the two groups.

5. These are personal discrepancies. Ashton & Esses (Reference Ashton and Esses1999) computed a personal discrepancy score for each perceiver, and then reported the number of perceivers who were within .2 standard deviations (sd) of the criteria, the number that exaggerated real differences (saw a difference greater than .2 sd larger than the real difference) or underestimated real differences (saw a difference more than .2 sd smaller than the real difference).

Furthermore, the results from these studies provide little consistent support for the idea that stereotypes exaggerate real differences. Exaggeration of real differences occurred more often than underestimations in some studies (Ashton & Esses Reference Ashton and Esses1999; Ryan Reference Ryan1996), but underestimation occurred more often in others ((McCauley & Stitt Reference McCauley and Stitt1978; Wolsko et al. Reference Wolsko, Park, Judd and Wittenbrink2000). Even the evidence of exaggeration, however, was more mixed than this summary suggests. For example, the only study to assess the accuracy of personal discrepancies (Ashton & Esses Reference Ashton and Esses1999) found that a plurality of people were generally accurate (n = 36), and that more exaggerated (n = 34) than underestimated (n = 25). If stereotypes are defined as exaggerations of group differences, a definition I reject but which has deep roots in social psychology (e.g., G. W. Allport Reference Allport1954/1979; Campbell Reference Campbell1967), then in the Ashton and Esses (Reference Ashton and Esses1999) study, 61 of 95 people did not hold stereotypes.

Stereotype accuracy as correspondence between belief and criteria was generally very strong. Consensual stereotype accuracy correlations ranged from .53 to .93. Personal stereotype accuracy correlations ranged from .36 to .69. Although the participants in these studies were only fairly good at identifying the precise level of some characteristic of racial and ethnic groups, their perceptions of differences both within and between groups across the different attributes were quite high.

8.6.2. Gender stereotypes

Table 3 summarizes the results of studies of gender stereotypes accuracy. In most cases, at least a plurality, and often a majority, of consensual stereotype judgments were accurate, and accurate plus near miss judgments predominate in every study. For example, in the Swim (Reference Swim1994) study, of 33 judgments, 18 were accurate and 7 were near misses. There was no support for the hypothesis that stereotypes generally lead people to exaggerate real differences. As with race, underestimations counterbalanced exaggerations.

Table 3. The Accuracy of Gender Stereotypes

Individual correlations involve computing, for each individual perceiver, the correlation between their judgments (stereotypes) and the criterion. Studies performing this analysis typically report the average of those correlations. Aggregate correlations refer to the correlation between the overall average perceived difference between the groups (for the whole sample) and the group difference on the criteria. Only one word is entered in this column when one pattern (e.g., “accuracy”) occurred for a majority of results reported. When there was no majority (or the majority could not be determined from their data), the top two results, in order of frequency (most frequent first) are reported here.

Table 3 Notes:

1. Except where otherwise stated, all discrepancy results occur at the consensual level. Accuracy means within 10% of the real percentage or within .25 of a standard deviation. Exaggeration means the perceived differences between groups exceeded the group differences on the criteria. Underestimation means the perceived differences between groups was smaller than the group differences on the criteria. “Near miss” means perceivers were more than 10% wrong, but no more than 20% wrong. Except where otherwise noted, only one word is entered in this column when one pattern (e.g., “accuracy”) occurred for a majority of results reported. When there was no majority, the top two results (most frequent first) are reported here, except where otherwise noted.

2. These correlations do not appear in the original article, but are computable from data that was reported.

3. Swim (Reference Swim1994) sometimes reported more than one meta-analysis as a criterion for a perceived difference. In that case, I simply averaged together the real differences indicated by the meta-analyses in order to have a single criterion against which to evaluate the accuracy of the perceived difference.

4. For Beyer (Reference Beyer1999), all results are reported separately for men and women perceivers, except the individual correlations for GPA. Because there was no significant sex of perceiver difference in these correlations, Beyer reported the results separately for male and female targets.

5. For simplicity, if the study reported more than two correlations, I have simply averaged all their correlations together to give an overall sense of the degree of accuracy.

Across the studies summarized in Table 3, consensual stereotype accuracy correlations were quite high, ranging from .34 to .98, with most falling between .66 and .80. The results for personal stereotypes were more variable. Once they were inaccurate, with a near zero correlation with criteria (Beyer Reference Beyer1999, perceptions of female targets). In general, though, they were at least moderately, and sometimes highly accurate (most correlations ranged from .40 to .60 – see Table 3). A recent multi-national study found a fundamentally similar pattern, with consensual gender stereotype accuracy correlations ranging from .36 to.70, and showing no evidence of exaggeration (Löckenhoff et al. Reference Löckenhoff, Chan, McCrae, De Fruyt, Jussim, De Bolle, Costa, Sutin, Realo, Allik, Nakazato, Shimonaka, Hřebíčková, Graf, Yik, Ficková, Brunner-Sciarra, Leibovich de Figueora, Schmidt, Ahn, Ahn, Aguilar Vafaie, Siuta, Szmigielska, Cain, Crawford, Anwar Mastor, Rolland, Nansubuga, Miramontez, Benet-Martínez, Rossier, Bratko, Marušić, Halberstadt, Yamaguchi, Knežević, Martin, Gheorghiu, Smith, Barbaranelli, Wang, Shakespeare-Finch, Lima, Klinkosz, Sekowski, Alcalay, Simonetti, Avdeyeva, Pramila and Terracciano2014).

8.6.3. Other stereotypes

Empirical research on stereotype accuracy has also addressed a wide variety of other stereotypes (e.g., occupations, college majors, etc.; see Table 4), and found essentially the same broad and general patterns as obtained for race, ethnicity, and gender: high levels of accuracy, and little or no general tendency to exaggerate real differences.

Table 4. The Accuracy of Other Stereotypes

Table 4 Notes:

1. Except where otherwise stated, all discrepancy results occur at the consensual level. Except where otherwise noted: (1) Accuracy means within 10% of the real percentage or within .25 of a standard deviation; (2) exaggeration means the perceived differences between groups exceeded the group differences on the criteria; and (3) underestimation means the perceived differences between groups was smaller than the group differences on the criteria. Only one word is entered in this column when one pattern (e.g., “accuracy”) occurred for a majority of results reported. When there was no majority (or the majority could not be determined from their data), the top two results, in order of frequency (most frequent first) are reported here.

2. If the study reported more than one individual level (average) correlation, their correlations were averaged to give an overall sense of the degree of accuracy.

3. Neither percentages nor standard deviations were reported. I characterize the main results of their discrepancy analyses as “accurate” because seven of eight mean discrepancies are all less than 1 scale point (on a seven point scale).

4. These correlations do not appear in the original article, but are computable from data that was reported.

8.7. Inaccurate stereotypes

Despite the impressive and surprising evidence of the accuracy of stereotypes, there is some evidence of inaccurate stereotypes. In the United States, an early study found little accuracy in political stereotypes (Judd & Park Reference Judd and Park1993). More recent research on the stereotypes of the moral beliefs held by liberals and conservatives, found a more mixed picture, of both accuracy and exaggeration (Graham et al. Reference Graham, Nosek and Haidt2013).

A large scale study conducted in scores of countries found that there is also little evidence of accuracy in national stereotypes regarding personality (Terracciano et al. Reference Terracciano, Abdel-Khalek, Adám, Adamova, Ahn, Ahn, Alansari, Alcalay, Allik, Angleitner, Avia, Ayearst, Barbaranelli, Beer, Borg-Cunen, Bratko, Brunner-Sciarra, Budzinski, Camart, Dahourou, De Fruyt, de Lima, del Pilar, Diener, Falzon, Fernando, Ficková, Fischer, Flores-Mendoza, Ghayur, Gulgoz, Hagberg, Halberstadt, Halim, Hřebičková, Humrichouse, Jensen, Jočič, Jonsson, Khoury, Klinkosz, Knežević, Lauri, Leibovich, Martin, Marusič, Mastor, Matsumoto, McRorie, Meshcheriakov, Mortensen, Munyae, Nagy, Nakazato, Nansubuga, Oishi, Ojedokun, Ostendorf, Paulhus, Pelevin, Petot, Podobnik, Porrata, Pramila, Prentice, Realo, Reátegui, Rolland, Rossier, Ruch, Rus, Sánchez-Bernardos, Schmidt, Sciculna-Calleja, Sekowski, Shakespeare-Finch, Shimonaka, Simonetti, Sineshaw, Siuta, Smith, Trapnell, Trobst, Wang, Yik, Zupančič and McCrae2005). However, Heine et al. (Reference Heine, Buchtel and Norenzayan2008) found that, when behavioral rather than self-report data were used as the criteria, far more evidence of stereotype accuracy emerged for the conscientiousness factor (correlations between consensual stereotypes and behavior averaged about .60). A recent replication (McCrae et al. Reference McCrae, Chan, Jussim, De Fruyt, Löckenhoff, De Bolle, Costa, Hřebíčková, Graf, Realo, Allik, Nakazato, Shimonaka, Yik, Ficková, Brunner-Sciarra, Reátigui, Leibovich de Figueora, Schmidt, Ahn, Ahn, Aguilar-Vafaie, Siuta, Szmigielska, Cain, Crawford, Mastor, Rolland, Nansubuga, Miramontez, Benet-Martínez, Rossier, Bratko, Marušić, Halberstadt, Yamaguchi, Knežević, Purić, Martin, Gheorghiu, Smith, Barbaranelli, Wang, Shakespeare-Finch, Lima, Klinkosz, Sekowski, Alcalay, Simonetti, Avdeyeva, Pramila and Terracciano2013) addressing many of the issues raised by Heine et al. (Reference Heine, Buchtel and Norenzayan2008), but still using self-reports as criteria, again showed almost no accuracy in national character stereotypes. The fairest conclusion, therefore, seems that the (in)accuracy of national character stereotype remains contested and unresolved.

8.8. Strengths and weaknesses of research on the accuracy of racial, ethnic, and gender stereotypes

Stereotype accuracy correlations are among the largest and most replicable effects in all of social psychology (see Table 5), and are typically far larger than the effect sizes routinely interpreted as support for more famous social psychological hypotheses. Several methodological aspects of these studies are worth noting because they bear on the generalizability of these results. Only one (Judd & Park Reference Judd and Park1993) was based on a nationally representative sample, so that, although evidence of stereotype accuracy is common in the data, the generalizability of those findings is currently unknown. Although most studies assessed the accuracy of undergraduates’ stereotypes, several assessed the accuracy of samples of adults (see Tables 2 and 3). Some of the highest levels of accuracy occurred with these adult samples, suggesting that the levels of accuracy obtained do not represent some artifact resulting from the study of undergraduate samples.

Table 5. Social Stereotypes are more valid than most social psychological hypotheses

Table 5 Notes:

1. Date obtained from Richard et al.’s (Reference Richard, Bond and Stokes-Zoota2003) review of meta-analyses of all of social psychology, including thousands of studies. Effects are in terms of the correlation coefficient, r.

2. From Tables 2 through 4: Within parentheses, the numerator is the number of stereotype accuracy correlations meeting the criteria for that row (exceeding .30 or .50) and the denominator is the total number of stereotype accuracy correlations. Because Table 2 summarizes the results for five studies for McCauley et al. (Reference McCauley, Thangavelu and Rozin1988), the .94–.98 figure is counted five times. These numbers probably underestimate the degree of stereotype accuracy, because all single entries in Tables 2 through 4 only count once, even though they often constitute averages of several correlations found in the original articles, and because I did not use the r-to-z transformation (this table describes the data, it does not report tests of statistical significance). Furthermore, the simple average of correlations is conservative, tending to underestimate the true correlation (Silver & Dunlap Reference Silver and Dunlap1987).

Second, the studies used a wide variety of criteria: U.S. Census data, self-reports, Board of Education data, nationally representative surveys, locally representative surveys, U.S. government reports, and so forth. The consistency of the results across studies, therefore, does not reflect some artifact resulting from use of any particular criteria.

Third, the studies examined a wide range of stereotype content: beliefs about demographic characteristics, academic achievement, personality and behavior. The consistency of the results across studies, therefore, does not reflect some artifact resulting from a particular type of stereotype content.

Fourth, personal discrepancies were the least studied of the four types of accuracy. Thus, the studies do not provide much information about the extent to which individual people's stereotypes deviate from perfection. Because of wisdom of crowd effects (Surowiecki Reference Surowiecki2004), it is likely that consensual discrepancies are more accurate than individual discrepancies, though identifying and understanding sources of individual discrepancies remains an important area for future research.

9.1. Stereotype “biases” that increase accuracy

When individuating information is unavailable or ambiguous, reliance on an accurate stereotype to make a guess or inference about a target will lead one to be as accurate as possible (e.g., Jussim Reference Jussim1991) and more accurate than will ignoring the stereotype. Whether this is morally desirable and whether acting on such judgments is always legal is a complex issue addressed in the book.

When stereotypes have substantial degrees of accuracy, then, in the absence of completely diagnostic individuating information, people will generally arrive at more accurate judgments when using than when ignoring their stereotypes. This is exactly what has been found in studies of occupational stereotypes (C. Cohen Reference Cohen1981), college dorm stereotypes (Brodt & Ross, Reference Brodt and Ross1998), role stereotypes (Macrae et al. Reference Macrae, Milne and Bodenhausen1994), and gender stereotypes (Gosling et al. Reference Gosling, Ko, Mannarelli and Morris2002; Jussim et al. Reference Jussim, Eccles and Madon1996; Madon et al. Reference Madon, Jussim, Keiper, Eccles, Smith and Palumbo1998). Research published after Social Perception and Social Reality similarly shows that reliance on stereotypes regarding mothers increases empathic accuracy (Lewis et al. Reference Lewis, Hodges, Laurent, Srivastava and Biancarosa2012), and that reliance on gay stereotypes increased perceiver's ability to identify whether a target was gay from a photograph of the face alone, if the gay target appeared feminine, but not if the gay target appeared masculine (Stern et al. Reference Stern, West, Jost and Rule2013).

Stereotype “biases” sometimes increase rather than decrease the accuracy of person perception. Because only a handful of studies have addressed this issue, psychologists are not yet in a position to reach broad conclusions about the generality or pervasiveness of this pattern. Nonetheless, the data that has actually assessed person perception accuracy contrast sharply with the common interpretation in social psychology that any influence of a stereotype on person perception is an unjustified distortion (Brown Reference Brown2010; Fiske Reference Fiske, Gilbert, Fiske and Lindzey1998; Stangor Reference Stangor, Lee, Jussim and McCauley1995).

9.2. Why is the evidence of stereotype accuracy and rationality important and useful?

1. 1. Claims, such as “stereotypes are inaccurate” and “it is hard to get people to individuate” are unjustified by existing data. Unqualified claims that stereotypes “exaggerate real differences” are not justified. The consensual stereotypes that have been studied demonstrated extraordinarily high levels of accuracy, so that unqualified suggestions that stereotypes are false cultural myths are not justified. Stereotype accuracy has been obtained by multiple independent research teams. It is also one of the largest effects in all of social psychology.

2. 2. Stereotypes are a central component of how people think about other people. This is part of many core definitions of social psychology and social cognition.

3. 3. Allowing for the possibility that some stereotypes may have some degree of accuracy leads to a coherent understanding of past and future research. Absent a recognition that on both logical and empirical grounds stereotypes may be accurate, past research will be haunted by a scientifically incoherent definitional tautology, which is, that people who believe in stereotypes are in error because stereotypes are erroneous beliefs.

Recognizing the existence of stereotype accuracy raises interesting and important theoretical and empirical questions. When do stereotypes flexibly change in response to changes in social reality and when do they remain rigidly resistant to change? When do people's person perception judgments correspond to, or deviate from, Bayesian rationality? Are the deviations predictable from implicit or explicit prejudice? What characteristics of perceivers and targets (both individuals and groups) moderate degree of (in)accuracy?

Acknowledging accuracy and rationality in stereotypes and stereotyping neither contests nor diminishes the importance of scientific research on stereotype biases or sources of oppression or inequality. Accuracy and bias can and often do co-exist (Jussim Reference Jussim1991; Reference Jussim2012). Accuracy, construed here as a question of degree, rather than something absolute, leaves open ample room for inaccuracy and bias. Demonstrating accuracy rarely precludes the possibility of bias, even socially important ones; demonstrating bias does not preclude high levels of accuracy. And, perhaps even more important, if stereotypes are often reasonably accurate, it highlights the question: what other phenomena create or maintain inequality?