“ambiguous” person is an assailant, decision makers inevitably exaggerate the weight properly accorded to this fact. Whatever merit there is to the contention that it is appropriate to consider a person’s race as one—just one—of the factors defining the “kind” of person who poses a danger, the racial factor assumes overriding psychological significance when the supposed assailant is Black.
For White Bayesians, cultural differences increase the danger of overestimating the threat posed by a supposed Black assailant. Nonverbal cues such as eye contact and body communication, for instance, vary significantly among subcultures, and thus may fail in intercultural situations.14 If the female bank patron in our opening hypothetical scenario were White (her racial identity is intentionally undefined), her misinterpretation of the Black victim’s eye and body movements as furtive and threatening may have resulted from cultural differences in nonverbal cues, illogically distorting her perception of danger.
Even if we accept the Bayesian’s insistence that his greater fear of Blacks results wholly from unbiased analysis of crime statistics, biases in the criminal justice system undermine the reliability of the statistics themselves. Racial discrimination in sentencing, for example, causes arrest statistics to exaggerate what differences might exist in crime patterns between Blacks and Whites, thus undermining the reliability of such statistics.15 A 1996 New York State study revealed that 30 percent of Blacks and Hispanics received harsher sentences than Whites in New York for comparable crimes, and that approximately four thousand Blacks and Hispanics are incarcerated each year for crimes under circumstances that do not lead to incarceration for Whites. Further exaggerating differences between Black and White crime rates is discrimination by police officers in choosing whom to arrest.16 Thus, although the rate of robbery arrests among Blacks is approximately twelve times that of Whites, it does not necessarily follow that a particular Black person is twelve times more likely to be a robber than a White.
Although biases in the criminal justice system exaggerate the differences in rates of violent crime by race, it may, tragically, still be true that Blacks commit a disproportionate number of crimes. Given that the blight of institutional racism continues to disproportionately limit the life chances of African Americans, and that desperate circumstances increase the likelihood that individuals caught in this web may turn to desperate undertakings, such a disparity, if it exists, should sadden but not surprise us. As Guido Calabresi, former dean of the Yale Law School and current federal appeals court judge, points out: “[O]ne need not be a racist to admit the possibility that the stereotypes may have some truth to them. I don’t believe in race, but if people are treated badly in a racist society on account of an irrelevant characteristic such as color or language, it should not be surprising if they react to that treatment in their everyday behavior.”17
The media spin on the comments of the Reverend Jackson decrying Black-on-Black crime used Jackson’s call for Blacks to take action on crime in their communities as an admission by the civil rights leader that racism and economic injustice have nothing to do with the crime problems of those communities. Columnist Mike Royko, for example, reported that Jackson believes it’s “a waste of time to expect government to reduce … urban mayhem.”18 From this standpoint, self-help and government investment are mutually exclusive. Anyone advocating antibias programs or federal aid to cities is portrayed as “making excuses” for Black people’s own self-destructiveness. Accordingly, when Jackson expressed fear that his ideas would be misconstrued by media and politicians looking for scapegoats, and further reiterated his long-standing insistence that both government help and self-help are needed for the African American community, he was widely derided. “[R]ather than grant [Whites the absolution for which they hungered]—and reap the enormous good will and political cooperation such a move might bring—Jackson has pulled back,” declared U.S. News.19
To the extent that Blacks do commit disproportionate numbers of violent street crimes, socioeconomic status largely explains such overrepresentation. Crime rates are inextricably linked to poverty and unemployment. Genetic explanations of crime statistics founder on the fact that crime and delinquency rates of the African American middle class are virtually identical to those of Whites similarly situated.20
Recognizing the socioeconomic factors that drive violent street crime, the Bayesian may insist that he views race merely as a proxy for information with admittedly greater predictive value—such as income, education, and prospects for the future—but that costs more to obtain. “Thus,” says the Bayesian, “I consider a wealthy Black graduate of the Harvard Law School who is making six figures at a major Wall Street law firm to pose a lower risk of armed robbery than a poor and illiterate White high school dropout with little hope of gainful employment.”
“However,” he continues, “ascertaining an individual’s schooling and income may require a personal interview and reference checks. The costs of obtaining such particularized information may be prohibitive in many situations. Surely you can’t expect shopkeepers, cabdrivers, or people in the position of our hypothetical bank patron to incur such costs, to get a person’s life story before he fingers his buzzer, stops his taxi, or uses deadly defensive force against a ‘suspicious’ person. When obtaining such information is prohibitively costly, we must economize by using stereotypes and playing the odds.”
Viewed in this light, the Bayesian’s claim that race can serve merely as a proxy for socioeconomic status might seem persuasive. But if race is a proxy for socioeconomic factors, then race loses its predictive value when one controls for those factors. Thus, if an individual is walking through an impoverished, “crime-prone neighborhood,” as the Reverend Jackson may have had in mind, and if he has already weighed the character of the neighborhood in judging the dangerousness of his situation, then it is illogical for him to consider the racial identity of the person whose suspicious footsteps he hears. For he has already taken into account the socioeconomic factors for which race is a proxy, and considering the racial identity of the ambiguous person under such circumstances constitutes “double counting.”21
Since our hypothetical scenario takes place in a predominantly White upper-middle-class neighborhood, it does not seem to implicate the double-counting problem. Further, the discussion shall proceed on the basis of two assumptions: first, that the rate of robberies is “significantly” higher for Blacks than for non-Blacks; second, and most unrealistic, that the defendant’s greater fear of Blacks results entirely from his analysis of crime statistics. Given these assumptions, what objections to the argument of the Bayesian remain? Surely admitting statistics, carrying logic and objectivity on the rising and plunging curves of their graphs like Vulcans on dolphinback, better promotes the accuracy, rationality, and fairness of the fact-finding process than not admitting them.
Why Rational Discrimination Is Not Reasonable
The most readily apparent objection to the reasonableness claim of the Bayesian challenges the statistical method he employs to assess the victim’s dangerousness. Neither private nor judicial judgments about a particular member of a class, the argument goes, should rest on evidence about the class to which he or she belongs. Despite the attractiveness of this principle, and occasional court admonitions to avoid statistical inferences about individuals, private and judicial decision makers routinely rely on statistical evidence to judge past facts and predict future behavior. Lenders use statistics concerning age, marital status, location or residence, income, and assets to predict whether a borrower will repay a loan. Parole commissions may also use statistical techniques to predict parole success, considering factors such as number of prior convictions, type of crime, employment history, and family ties. And courts consider nonindividualized statistical probabilities when deciding whether to allow injured litigants to use epidemiological proof of causation in their lawsuits.
To accept the usefulness of statistical generalizations as a general matter, however, is not to agree that such generalizations are appropriate everywhere. For the use of statistical generalizations entail significant social costs, notwithstanding obvious benefits to defendants. The fatal flaw in the Bayesian’s argument lies in his failure to take account of the costs of acting on his racial generalizations. Instead, he assumes