I RELY HEAVILY in this book on the elementary observation that, in the first instance, “race” is a mode of perceptual categorization people use to navigate their way through a murky, uncertain social world. I want us to think about people as being hungry for information, constantly seeking to better understand the social environment in which they are embedded, searching always for markers, guideposts, clues that can equip them to make wiser choices on matters of consequence. This is a cognitive, not a normative activity. Information-hungry human agents—in making pragmatic judgments, to be sure, but also as a necessity for survival—will notice visible, physical traits presented by those whom they encounter in society: their skin color, hair texture, facial bone structure, and so forth. There is neither shame nor mystery in this. The practice of grouping people together on the basis of their common possession of visible bodily marks is a universal aspect of the human condition.1 One of the ways that we generate and store social information is to classify the persons we encounter—that is, form broad categories between which contrasts can be drawn and about which generalizations can be made—so we can better know what is to be expected from those with whom we must deal, but about whom all too little can be discerned. So I would like to begin with a few observations about the act of “racial classification.”2
As befits an economist, I employ the concept of classification in the decision-theoretic sense: Decision-makers (agents) act in ways that affect others (subjects) on the basis of what can be observed about those toward whom their actions are directed. An employer hires, a banker lends, a landlord rents, a neighbor moves, a suitor woos, or not, and so on. As a purely cognitive matter, agents, surveying the field of human subjects, endeavor to discern relevant distinctions among subjects in that field in order to refine their actions, that those actions may better serve their ends. To make a distinction of this kind is to engage in an act of “classification” in the sense that I intend here. When distinctions are based in some way on a subject’s “race,” then we are dealing with an act of “racial classification.”
Two things should be obvious straightaway: First, whether “race” is a part of the calculation or not, classifying human subjects in this general way is a universal practice, one that lies at the root of all social-cognitive behavior. There can only be the question of how, not whether, human agents will classify those subject to their actions. Second, at this level of generality, the normative status of even a race-based classification cannot be definitively assessed absent some consideration of the purposes on behalf of which the classifying act has been undertaken. That is, the simple fact that a person classifies others (or herself, for that matter) in terms of “race” is in itself neither a good nor a bad thing. Normative judgment must, at the very least, entail some analysis of the goals of classifying agents.
I stress this last point because it has apparently eluded some commentators on race relations in American life who argue that, since no exact biological taxonomy can vindicate the “race” idea, any use of this category for classificatory purposes is, ipso facto, morally dubious. But that cannot be correct. Even the current U.S. Supreme Court, in its doctrine of “strict scrutiny,” recognizes that whether or not the reasons for a racial classification are compelling should constitute one of the tests to determine if it passes muster in constitutional terms. Yet the point is much more general. For both the racist employer (bent on holding blacks down) and the diligent public servant (intent on enforcing the laws against discrimination) will alike and necessarily be “guilty” of classifying the field of human subjects in racial terms as they carry forward their respective, diametrically opposed projects.3 It follows that the cognitive act of so classifying is insufficient, by itself, to allow a normative judgment.
By focusing on racial classification I pose the question in a manner that may be contrasted with the more traditional focus on racial discrimination. In doing so I do not imply that no problems of racism or race-based unfairness in U.S. society still exist. To the contrary, although the extent of overt racial discrimination against blacks has obviously declined over the last half-century, it seems to me equally obvious that racial injustice in U.S. social, economic, and political life persists, though less transparently so, and in ways that are more difficult to root out. So, at any rate, I will be arguing here. I give pride of place to classification over discrimination for two reasons—generality and utility: Classification is the logically prior concept; and thinking in terms of classification yields greater insight into the problem at hand than thinking in terms of discrimination. The first point is a trivial one, once we see that discerning a difference is a necessary condition for acting on one. That my approach is more useful will, I trust, be clear when the reader has finished this book.
I want now to say more formally what I intend by the term “race.” In this book I use that term to refer to a cluster of inheritable bodily markings carried by a largely endogamous group of individuals, markings that can be observed by others with ease, that can be changed or misrepresented only with great difficulty, and that have come to be invested in a particular society at a given historical moment with social meaning. This definition has three aspects: ease of identification, relative immutability, and social signification. While physical markings on the human body are central to my notion of “race,” I stress (in keeping with Axiom 1) that nothing turns on the underlying biological factors that may engender those markings. I only require that the pertinent physical traits are passed on across generations, are easily discerned, and are not readily disguised. Moreover, what is “essential” here is that these physical traits are taken to signify something of import within an historical context. “Race,” on my account, is all about embodied social signification. As such, much depends on the processes through which powerful meanings come to be associated with particular bodily marks. Obviously, these will have to be historically specific, culturally mediated processes.4
There has been much recent discussion in philosophy and cultural studies about the ontological status of “race”—are there any things in the world that may be taken as corresponding to the word “race,” and so forth. It has become fashionable to put the word in quotes, by way of emphasizing its problematic scientific and philosophical status. The core claim in this literature is that there exist no objective criteria—biological, cultural, or genealogical—through use of which the set of human beings can be consistently partitioned into a relatively small number of mutually exclusive, collectively exhaustive subsets that may be taken as races. Belief in the existence of races, on this view, is rather like belief in the existence of witches—just mischievous superstition, nothing more.5 I do not follow this line of argument here.
Of course, neither do I dispute the core claim—Axioms 1 and 2 declare as much. But I find little of interest in this philosophy of language exercise. Rather, I am impressed, as any good social scientist would be, by the fact that so much human behavior has become organized around racial categorization, despite its evident lack of any basis in biological taxonomy. This, it would appear, is what must be explained. One has no need for objective rules of racial taxonomy to study, as I do here, the subjective use of racial classifications. It is enough that influential observers (passersby on the street, new neighbors before the moving van arrives, policemen, employers, bankers, and so on) hold schemes of classification in their minds, and act on those schemes. They need not make their schemes explicit; their methods of classification may well be mutually inconsistent, one with another. And while it may be true that these agents could not give cogent reasons for adopting their schemes, it is also the case that they are unlikely ever to be asked to do so.
Still, if a person is aware that others in society are inclined to classify him on the basis of certain markers and if, in turn, this classification constitutes the basis of differential actions affecting his welfare, then these markers will become important to him. He will attend to them, become conscious (and, I dare say, self-conscious) in regard to them. He will, at some level, understand and identify himself as being “raced.”6 This will be a rational cognitive stance on his part, not a belief in magic, and certainly not a moral error.7
Moreover, whatever the scientific status of the race concept, the social convention of classifying people on the basis of their bodily markings will typically have profound, enduring, and all too real consequences. This ubiquitous practice can, at one and the same time, be eminently consistent with reason, stubbornly resistant to change, and a formidable barrier to the attainment of social justice. To illustrate how and why this can be so, I want now to consider in some detail the inner workings of what I will call “self-confirming racial stereotypes.”
A “self-confirming stereotype” is a statistical generalization about some class of persons regarding what is taken with reason to be true about them as a class, but cannot be readily determined as true or false for a given member of the class. Furthermore, this generalization is “reasonable” in the specific sense that it is self-confirming: Observers, by acting on the generalization, set in motion a sequence of events that has the effect of reinforcing their initial judgment. And so a “self-confirming racial stereotype” is simply a generalization of this kind about a class of persons defined in part or altogether on the basis of whatever categories of racial classification happen to be operative in observers’ minds. I wish to consider the rationality, durability, efficiency, and fairness of self-confirming racial stereotypes.
Obviously, a generalization about some group can be supported by evidence without that evidence having in any way been influenced by the actions of those making the generalization. Thus not all “reasonable” stereotypes will be self-confirming. However, I am interested here in the special circumstance in which those making a surmise about some group of persons have within their power the ability to act so as to influence the population being observed. For reasons that will become clear, I see this particular circumstance as being highly relevant to the task of understanding and evaluating the social problem of persistent racial inequality in the United States.
I acknowledge that this use of the term “stereotype” diverges from common parlance. Webster’s New World Dictionary defines “stereotype” as “A fixed idea or popular conception about how a certain type of person looks, acts, etc.” One senses a connotation of “unreasonableness” in that definition—the stereotype being a false or too simplistic surmise about some group: “blacks are lazy,” “Jews are cunning,” and so on. While I do not dispute that this crude overgeneralizing behavior occurs, it is not my subject here. Rather, my model of stereotypes is designed to show the limited sense in which even “reasonable” generalizations, those for which ample supporting evidence can be found, are fully “rational.” I argue that such generalizations often represent instances of what I will refer to as “biased social cognition.”
The self-confirming property of stereotypes as defined here is, therefore, crucial to my argument. I will be positing situations in which stereotypic thinking seems plausible, so I can go on to show that, even then, where race is involved things may not be quite as they appear.
To illustrate, if agents hold a negative stereotype about blacks they may think (correctly) that, on the average and all else equal, commercial loans to blacks pose a greater risk of default or black residential neighborhoods are more likely to decline. But this can hardly be the end of the story. What about the possibility that race conveys this information only because agents expect it to, and then act in ways that lead to the confirmation of their expectations? What if blacks have trouble getting further extensions of credit in the face of a crisis, and so default more often? Or what if nonblack residents panic at the arrival of blacks, selling their homes too quickly and below the market value to lower-income buyers, thereby promoting neighborhood decline?
If under such circumstances observers attribute racially disparate behaviors to inherent limitations of the stereotyped group—thinking, say, that blacks do not repay their loans or take care of their property because they are just less responsible people on average—these agents might well be mistaken. Yet, given that their surmise about blacks is supported by hard evidence, they might well persist in the error. Now, notice one thing: This mistake would be of great political moment. For attributing an endogenous difference (a difference produced within a system of interactions) to an exogenous cause (a cause located outside that system) leaves one less interested in working for systemic reform. This is the effect I am after with the models to be elaborated below, and this is why I am willing to employ an apparently loaded phrase like “biased social cognition”: It is a politically consequential cognitive distortion to ascribe the disadvantage to be observed among a group of people to qualities thought to be intrinsic to that group when, in fact, that disadvantage is the product of a system of social interactions. My contention is that in American society, when the group in question is blacks, the risk of this kind of causal misattribution is especially great.
Now, whether race is involved or not, the logic of self-confirming stereotypes as I conceive them entails three key components:
How can we relate this abstract way of thinking to the subject at hand? In the broadest terms, this stereotype-logic provides an analytic template to illustrate how the cognizance of race comes into existence and is reproduced through time in society. This logic, in other words, provides insight into how and why observers use racial categories for their classifying purposes. The point is that the inferential, self-confirming logic just outlined can easily be contingent on the racial characteristics of subjects, such that altogether different outcomes occur for subjects belonging to different races—that is, distinguishable by observable bodily marks. Although these race-markers may be of no intrinsic significance, they nevertheless can serve as useful indices around which human agents organize their expectations.
One way to think about race conventions, then, is to see them as the equilibria that emerge when subjects and agents in the habit of noticing certain racial markers interact with one another on matters of consequence under conditions of limited information.8 It becomes “rational” for agents to classify a subject using functionally irrelevant (racial) markers because this allows them more accurately to assess that subject’s functionally relevant but unobservable traits. Physical traits matter because observers (correctly) expect them to matter. This expectation induces agents to interact with subjects in a manner that depends on race, thereby creating different incentives for subjects in racially distinct population subgroups. Responding to these incentives, subjects adapt according to how they expect to be perceived, which is to say, they adapt differently depending on their race. In the equilibrium, this race-varying behavior by subjects is consistent (on the average) with observing agents’ initial beliefs, confirming the agents’ supposition that a subject’s race would be informative. Race conventions emerge as byproducts of the happenstance of observable morphological variability in human populations. Put differently, race matters “in the equilibrium” (as we economists would say) as a result of the inexorable logic of self-confirming feedback loops.
At this (admittedly high) level of generality, a “race” could be constructed around any cluster of inheritable physical markers shared by a largely endogamous human sub-population that are easy for observers accurately to assess and that can be misrepresented only with difficulty. Observers, doing the best they can under trying circumstances, end up partitioning the field of human subjects in such a way that a person’s hard-to-observe but functionally relevant (say, economic) traits can be effectively estimated by conditioning on that person’s evidently informative though functionally irrelevant (racial) traits.
This, then, is my “model” of self-confirming racial stereotypes.
We are clearly in need of examples at this point. A few thought experiments will illustrate the logic just outlined.
Imagine a group of employers who harbor the a priori belief that blacks are more likely than others to be low-effort trainees. Suppose they observe the number of mistakes any employee makes on the job, but not the effort exerted by that employee during the training period. Let employers have the option of terminating a worker during the training period, and suppose they find it much more difficult to do so later on. Then employers will set a lower threshold for blacks than for other employees on the number of mistakes needed to trigger dismissal, since, given their prior beliefs, they will be quicker to infer that a black worker has not put in enough effort to learn the job. Mistakes by black workers early in their tenure will provide evidence of the employers’ worst fears, more so than an equal number of mistakes by other workers. Employers will, therefore, be less willing to extend the benefit of the doubt to blacks during the training period.
But how will black workers respond to such behavior by employers? It is costly to exert effort during the training period, and the reward for doing so can only be realized if an employee escapes termination. Knowing they are more likely to be fired if they make a few mistakes, an outcome over which they cannot exert full control, more black than other workers may find that exerting high effort during the training period is, on net, a losing proposition for them.9 If so, fewer of them will elect to exert themselves. But this will only confirm the employers’ initial beliefs, thereby bringing about a convention in which the employers’ racial stereotype—“blacks tend to be low-effort trainees”—will (seem to) be entirely reasonable.
Alternatively, suppose most taxi drivers refuse to stop for young black men after a certain hour because they fear being robbed, though a few drivers will stop for anyone. Let there be two types of young men—those merely trying to get home late at night and those intent on robbery—and let us suppose that the relative number of the two types does not depend on race.10 Now, for most young men, anticipating a long wait will discourage dependence on taxi transportation. They may arrange to ride with friends, take public transport, or bring a car, and this is especially so if a young man is simply trying to get home. But a person bent on robbery will not be so easily deterred. Even though he knows most cabs are unlikely to stop, he only needs one to do so to get in his night’s work.11 Given that taxi drivers treat blacks differently, stopping less frequently for them, and that robbers are less easily deterred than are the law abiding, the drivers’ reluctance to stop will discourage relatively more of the law abiding than of the robbers among blacks from relying on taxi transportation. This effect will not be present for nonblacks, since drivers are quite willing to stop for them. Hence, through a process that economists call “adverse selection,” the set of young black men actually seen to be hailing taxis after dark may well come to contain a noticeably larger than average fraction of robbers, precisely the circumstance presumed by the drivers in the first place.
Notice what is happening here: The drivers’ own behaviors have created the facts on which their pessimistic expectations are grounded. Indeed, in the context of this thought experiment, were most drivers as willing to stop for young black men as for others, the set of blacks hailing cabs would be no more threatening than the overall population average. But then it would be reasonable for drivers to pay no heed to race when deciding whether or not to stop! So is it “rational,” or not, for drivers to use race as a signifier of danger? Clearly, once a convention employing the self-confirming stereotype has been established, the drivers’ beliefs and actions are defensible on the basis of reason. And yet the deeper conclusion—that there is an intrinsic connection between race and crime—is altogether unjustified. I think it is safe to assume that this subtle distinction will elude most cab drivers, politicians, Op-Ed writers, and not a few social scientists!
Consider another example. Suppose automobile dealers think black buyers have higher reservation prices than whites—prices above which they will simply walk away rather than haggle further. On this belief, dealers will be tougher when bargaining with blacks, more reluctant to offer low prices, more eager to foist on them expensive accessories, and so on. Now, given that such race-based behavior by dealers is common, blacks will come to expect tough dealer bargaining as the norm when they shop for cars. As such, a black buyer who contemplates walking away will have to anticipate less favorable alternative opportunities and higher search costs than will a white buyer who entertains that option. And so the typical black buyer may find it rational to accept a price rather than continue searching elsewhere, even though the typical white may reject that same price.12 Yet this racial difference in typical behavior by buyers is precisely what justified the view among dealers that a customer’s race would predict bargaining behavior. Thus, even if there are no intrinsic differences in bargaining ability between the two populations, a convention can emerge in which the dealers’ rule of thumb, “be tougher with blacks,” is all too clearly justified by the facts.
Here is a final example. Suppose black and white students apply for admission to a group of professional schools, and that the schools are keen to admit what they think of as an adequate number of blacks. Suppose further that, in the experience of the admissions officers at these schools, there is a substantial disparity in the academic merit of black and white applicants, on the average, and that the use of a uniform standard for the two racial groups would not yield an adequate number of black admissions. Accordingly, in order to meet their racial diversity goals, these admissions officers are convinced that they must accept some blacks with test scores and/or grades that would lead to rejection if submitted by a white applicant. Let most schools follow this policy, and consider the incentives for achievement that will have been created in the racially distinct student populations. Blacks will (correctly) anticipate that the level of performance sufficient for them to gain admission is lower than the level (correctly) presumed necessary by whites. If students are, at least to some extent, responsive to these differing incentives, then those anticipating tougher standards may (on average) exert greater effort than those anticipating more relaxed admission standards.13 If this is so, the initial belief by admissions officers—that different standards were necessary to achieve enough diversity—may have been a self-fulfilling prophecy: There may have been no difference in the underlying tendency of the two groups of students to achieve high test scores and grades. Using race-dependent admissions standards may have set in motion a sequence of events that, in the end, confirmed in the officers’ minds that their preferential handling of black applicants was required. Had the officers steadfastly stuck to racially uniform standards, they might thereby have created a factual circumstance in which their diversity goals could be met without any use of race in the admissions process.14
At this point, a reader may be asking some questions, such as:
These questions go to the heart of the matter, and dealing with them leads naturally into a discussion of racial stigma. Before going further, however, I wish to make two observations.
First, I want to declare that I do not see the feedback mechanism just illustrated as the be-all and end-all of race-based behavior in society. As will become evident, I believe people attend to racial markers because they convey social meanings, and not just social information. Still, I think that to study conventional stereotyping is an empirically relevant and analytically useful exercise. It yields insight into how racially disparate outcomes can be understood without recourse to essentialist assumptions of innate racial difference. It shows how acquired differences in capabilities between members of different racial groups—due for instance to unequal access to resources critical for human development—can be magnified into even larger differences in social outcomes. It clarifies why “The data bear me out when I say ‘those people’ are really less productive” is no good answer to the complaint that widely disparate group outcomes should be a cause of concern for anyone interested in social justice. This way of thinking at least hints at how great the cost may be—for members of a racially marked group, to be sure, but for the entire society as well—when widely held negative stereotypes about a visibly distinct subset of the population are allowed to persist indefinitely. And it shows why broad-based, system-wide interventions may be the only way to break into the causal chain that perpetuates racial inequality over time.
Second, I want to discourage any rush to moral judgment about the behavior of observing agents (the employers, cab drivers, automobile dealers, or admissions officers) in the examples just offered. As I see it, we are dealing with deep-seated cognitive behavior here. There is no getting around classification—it is the way of the world. People will classify when the stakes are high enough. Thus, our imaginary taxi driver stands to gain $10 from a law-abiding fare but to lose, say, $10,000 on average if he stops for a robber (allowing for possible loss of life or limb). With those stakes, even if the probability of robbery is on the order of one chance in a thousand, a small difference in the behavior of racial groups may shift a driver’s cost-benefit calculus from a “stop” to a “do not stop” decision. With the stakes so high, with information so limited, and given that a real correlation between “race” and “chance of robbery” is there to be observed, why should we condemn this taxi driver?
However, consider a traffic cop sitting in a $50,000 cruiser, who has received $100,000 worth of training, is backed by a big bureaucracy, and has a computer at his fingertips that allows him, by simply reading a license plate, to instantly generate reams of information. This is an observer with no excuse for allowing his behavior to be driven by racial generalizations. So my purpose here is to analyze, not moralize. I am arguing neither “for” nor “against” stereotypes. I seek merely to grasp their consequences; to fathom how racial stereotyping creates the facts that are its own justification; to understand how reasonable people, who base their surmises on hard evidence, can nonetheless hold the pernicious idea that blacks are different from others in some deeper (than race) way that accounts for their lowly status. The social production of such outcomes must first be understood. Only then, it seems to me, will it be possible to engage in effective social criticism.
But what about those good questions? I will address them in turn.
The self-confirming feedback process just illustrated treated each instance in isolation from the others, made no mention of history, and ignored factors like prejudice and misinformation—indeed, willful misinformation. Nor did it allow for any interaction between, on the one hand, reasonable information-based distinctions and, on the other hand, maltreatment of persons due to old-fashioned, unreasoning racial antipathy. And crucially, it did not ask whether persons subject to marker-based discrimination would have their ideas about their own worth or that of others with the same markers affected in any way. It is clear that, in the case of African Americans, all of these are counterfactual omissions. Taking such factors into account would, I submit, go some way toward answering Good Question #1.
Now, consider Good Question #2, which might be expanded as follows: Why don’t people learn about the self-confirming feedback mechanism and intervene so as to break the production of racial stereotypes, or disrupt their reproduction through time? Why doesn’t somebody do something about the entrenchment and reification of this way of thinking? If race-based classification is a human product—a social construction—then shouldn’t humans be able to control it? This question goes to the core of my concerns in this book, so I will take it up at some length.
To aid in this reflection, consider the key distinction between “competitive” and “monopolistic” observing situations. A competitive situation is one in which there are a large number of observing agents, each encountering subjects from an even larger, common population, each taking actions in relation to these subjects but knowing that, owing to their relatively insignificant size, no action they can take will affect the population’s characteristics. A monopolistic situation is one in which a single observing agent (or a quite small number) acts on a population of subjects. Examples of competitive observing situations include the taxi drivers encountering prospective fares and deciding whether or not to stop on the basis of their estimates of the likelihood of being robbed, and the low-wage labor market of a big city where many small employers hire from a common labor pool and use “race” as an indicator of likely worker reliability. Examples of monopolistic observing situations include a police department setting policy about how its officers should conduct traffic surveillance, the labor market in a small city where one or a few big employers dominate the hiring, and a huge bureaucracy like the military that deals with millions of people on a national scale.
Now, a monopolistic observer might upon reflection become aware of how his behavior (the use of racial markers to formulate race-dependent estimates of subjects’ hard-to-observe traits) produces feedback effects in the distinct populations in a way that ends up confirming his initial beliefs. That is, a monopolistic observer might try to take into account the incentive effects sketched earlier so as to improve the equity and the efficiency of the subject-observer interactions. But this would not be possible for a competitive observer. Even if a small employer or a taxi driver learned or was told about such feedback effects, there would be nothing to be done because, in a competitive situation, an individual’s action has so little impact on the overall observing environment. So Good Question #2 is most relevant in monopolistic observing situations.
This terminology—referring to “monopolistic” and “competitive” observing situations—is borrowed from economics in analogy with the distinction between sellers who do and those who do not have the power to set market prices. The analogy can be taken one step further: Even when sellers lack market power they can still act in concert, with the aid of government regulations, to set and enforce a minimum price. Likewise, in my model, even though competitive agents cannot influence the observing environment on their own, their collective action via government remains a possibility. So Good Question #2 may still be relevant in competitive observing situations, once allowance is made for the possibility of a coordinated response implemented through public policy. In any event, the question is certainly relevant in monopolistic observing situations, and these are numerous enough. So why, we must ask, do those observers who have “the power to create facts” not learn and intervene?
To venture an answer, and to hint further at the role of racial stigma in my overall argument, suppose this observer can credit two qualitative causal accounts of what produces his data. The first is the story just told, in which “race” predicts behavior only because, thinking it will do so, the observer uses the race-marker to discriminate, thereby inducing a statistical association between functionally irrelevant though easily observable marker and functionally relevant but unobservable trait. The second account posits that the marker itself is intrinsically relevant in some way. That is, the second account credits to some extent the view (racial essentialism) that I explicitly rejected in Axiom 2. Now, if a monopolistic observer believes mainly the first account, he will see the racially disparate outcome as being anomalous or surprising. He may therefore find it to be in his own self-interest to experiment, so as to learn about the structure that is generating his observations. He may be led in this way to reduce his reliance on the racial marker and, in so doing, to unmake the factual circumstance that initially justified its use.
However, if this monopolistic observer credits mainly the second, essentialist account, he will not see much of a benefit to be garnered from experimentation.15 (We need not assume that the observer wholly believes one story or the other; he may think either possible. My argument works so long as the essentialist account is given sufficient weight.) In this case, the observer’s experience does more than simply confirm his beliefs; it comports with his inchoate sense of the natural order of things. “Those people just don’t make good workers,” he will conclude, and he will continue to view them with the skepticism that, on the unsurprising (and uninspiring) evidence at hand, they seem so richly to deserve, looking down on their feeble and ineffective strivings, to borrow a phrase from W. E. B. Du Bois, with “amused contempt and pity.”
Now, a rationalistic account could be developed in which an agent experiments even though with current beliefs this seems unlikely to pay, because the agent thinks those beliefs may be wrong and realizes that experimentation may uncover the error. This, for instance, is one way that scientific communities function so as to advance the frontiers of knowledge. However, as Thomas Kuhn observed long ago, experimentation of this kind generally requires an observer to encounter events that are anomalous, challenging previously taken-for-granted understandings (Kuhn 1962). And whereas the incentives facing scientific investigators are structured precisely so as to encourage this “anomaly hunting,” it is something of an understatement to observe that the incentives facing those who employ low-skilled workers or who run police departments are not so structured.
The key point here, and the answer I have for Good Question #2, is this: Learning to discard an erroneous or incomplete causal explanation in matters of race is generally not a straightforward undertaking. If, on Kuhn’s account, highly disciplined scientific communities have trouble abandoning an outmoded paradigm, we may be sure that less formal social aggregates will as well. Revision of beliefs may well be a cognitive activity, but that cognition is always rooted in a social context and influenced by the taken-for-granted suppositions that agents hold.16 As a result, if a racial disparity does not strike a powerful observer as being disturbing, anomalous, contrary to his unexamined and perhaps not even consciously espoused presumptions about the nature of his social world, then he may make no special effort to uncover a deeper (than race) cause of the disparity. Certainly, the possibility that his own behavior has helped to engender the problem will be unlikely to occur to him. However reasonable (that is, nonarbitrary, grounded in evidence) his beliefs may be, the process through which he arrives at and holds on to those beliefs need not, and generally will not, be “rational” at all.
Indeed, in the academic field of decision theory the study of how rational agents learn is an underdeveloped area of research. Moreover, a powerful critique has been made of the notion that rational choice theory can, even in principle, give a foundational account of learning. The sociologist Barry Barnes eloquently makes this case, and, since it bears importantly on my own argument here, I quote Barnes at length:
In order to act as ER (economically rational) individuals we have to be knowledgeable. But knowledge … cannot be taken as given … Much of it has to be learned. The problem then arises of how far learning is rational action. It arises in the most acute form in the case of the new member, just arrived in the world with very little cognitive baggage at all, about to acquire language, knowledge and culture ab initio … Consider an ER baby, lying in its cot facing its new world of threat and opportunity. Will it consult its preference schedule and reflect: “Well what should I do now? A restful nap perhaps, or a spot of healthy foot-kicking for muscle development. Or perhaps a cry for mother, to do a bit of language-learning. Hang on a minute! How can I be thinking this when I haven’t any language yet to …?” New members … cannot be rational in quite the way that existing members perhaps are, because rationality requires mastery of a repertoire of symbols and reference to a body of knowledge, neither of which new members possess. The activities of new members, therefore, and in particular their learning activities, pose a special challenge to an individualist social theory with which they are prima facie incompatible … Evidently acts of learning, classification and inference … raise pervasive problems for individualism … It would seem that individuals must act in a certain sense arbitrarily if they are to act at all; what is routinely identified as rational action begins to look as if it is conventional action. (Barnes 1995, 34–36)
Barry Barnes makes a good point that applies far beyond the confines of a nursery or a playpen: We cannot hope to explain all of human behavior with a cost-benefit calculus. Specifically, when we ask how people acquire the mechanisms of symbolic expression peculiar to the communities in which they are embedded, we must consider the meaning of their relations with others. Plausibly, much social learning will come about as a by-product of social activity undertaken for its own sake: One wants to get along in the world, in this community, with other people. So one undertakes to see the world as others do—not because the benefit of doing so outweighs the cost, but because that is the way of being in the world with these people. This kind of thinking suggests that it is futile to look for “rationality” at the foundation of all social action.
But although Barnes’s critique is powerful, we need not adopt so comprehensive a view to appreciate a key point. We can stick with a more or less rational account of learning, and simply observe that people have to take a “cognitive leap of faith” with respect to how they specify the environment in which their learning is to take place. That choice of specification, plausibly, cannot be a fully rational act. Intuitively, the cognition underlying it is more a “pattern recognition” than a “deductive” type of cognition. It is as if the agent is thinking: “This fits. This feels about right. I think this framing of the problem is more or less accurate. Now, having so framed, I will go on to make a deductive calculation about whether this or that alternative hypothesis, seen from within my adopted frame of reference, makes sense.”
I admit that this is far from a rigorous social-psychological argument. I am aware that, by advancing it, I step rather far out on the proverbial limb. But as the force of Good Question #2 makes clear, some speculation of this sort is warranted, given the stakes. For if agents do not learn about mechanisms within their control that reproduce racial inequality through time, the results may be tragic. Consider the possibility that learning about the ultimate (not proximate) causes of a group disparity fails to occur for one division of the population (black/white) because, when told that the blacks are lagging, people’s sense is: “They are about where we expected them to be.” But learning does take place for a different division of the population (male/female) because, upon hearing that the girls are lagging, people instinctively harbor the thought: “Something must be dreadfully wrong.”
This is no simple accusation of “racism.” Nor am I charging the American people with caring more about gender inequality than about racial inequality—though it may be that they do so. Rather, I am making what I take to be a pertinent observation about the cognitive-adaptive possibilities implicit in various social situations, in which observers try to discern how the facts on which they base their decisions have been generated.
Specifically, I want to distinguish two cognitive acts required to process social information—specification and inference. An observer first adopts a specification, within the framework of which an inference is subsequently drawn. Specification refers to the qualitative framework guiding an agent’s data processing. Inference refers to the quantitative calculation of parameters from available data. The language is borrowed from statistics but is intended to apply to the cognitive assessments of ordinary persons, not statisticians.
Now, I assert that the mental processes underlying these cognitive acts are fundamentally different, and that while inference may be well conceived as a fully rational enterprise, specification is best thought of as an intuitive, instinctual, pattern-recognition type of activity. The cognition underlying the self-confirming feedback loops that lead to racial stereotypes, as illustrated in the foregoing thought experiments, is an instance of inference. But the questioning of long-held beliefs, and the willingness to experiment for the sake of learning though this might seem not to pay—these are instances of specification. We should expect nonrational factors—in particular, the taken-for-granted meanings that may be unreflectively associated with certain racial markers—to exert a significant influence on the latter type of cognitive behavior.
Here, then, is my final answer to Good Question #2: “Race” may be a human product, but, because it is a social convention that emerges out of the complex interactions of myriad, autonomous decision-makers, it is not readily subjected to human agency. Between us reflective agents and our social artifacts stand mechanisms of social intercourse that are anything but transparent. Because we filter social experience through racial categories, and given the ancillary meanings with which those categories are freighted, we can be led to interpret our data in such a way that the arbitrariness of the race convention remains hidden from our view, leaving us “cognitive prisoners” inside a symbolic world of our own unwitting construction.
I conclude this chapter by considering Good Questions #3 and #4. Why don’t observers look for nonracial markers to solve their inference problems? And what ensues when people who happen to share some markers are, willy-nilly, grouped into a single racial category, and yet those so grouped are objectively very different persons? Obviously, these two questions are closely related: The more heterogeneous is a racial group, the greater is the gain to an observer from using nonracial markers.
But we have already discussed this problem; looking for nonracial markers is merely another way of experimenting with one’s specification of the process generating one’s data. It is, in the colloquialism made famous by the University of Pennsylvania sociologist Elijah Anderson, a way of being “streetwise” (Anderson 1990). I will employ Anderson’s framework to make a final point about racial stereotypes.
Adapting the theory first elaborated by Erving Goffman (1959), Anderson uses the streets of the racially mixed West Philadelphia neighborhoods near the campus of the University of Pennsylvania as a laboratory. He studies the problem of “decoding” which all social actors must solve when meeting others in public. One cannot be entirely certain of the character or intent of “the other”; it is necessary to process such information as may be gleaned from an examination of the external self-presentation of those being encountered. The context of the meeting—time of day, physical setting, whether the individual is alone or in a group, and so on—will affect how these external clues are read. As an encounter unfolds, communication between the parties, ranging from a meeting of eyes (or the avoidance of same) to an exchange of greetings, permits further inferences to be drawn. Race—an easily and instantly ascertainable characteristic—may, as I have suggested, be expected to play a large role in this game of inference. Social class—as conveyed by dress, manner, occupation, and speech—may also be quite important. (In fact, as Anderson’s account makes clear, these two indicia of social identity interact in subtle and complex ways.) An individual’s experience of the social environment is strongly influenced by how he and those he encounters in public negotiate such meetings.
Anderson describes in elegant detail the rules of public etiquette, norms of mutual expectation, conventions of deference, methods of self-protection, strategies of turf-claiming, signals of intention, deciphering of cues, mistakes, biases, bluffs, threats, and self-fulfilling prophecies that are implicit in the interactions he observes. His ethnography is a wonderful illustration (indeed, a vindication) of the theoretical approach I am promoting in this book. He identifies social roles, public routines, and behavioral devices common to the encounters he chronicles. And he suggests compelling explanations for many puzzling features of life on the streets of the communities he has studied.
Anderson’s core concept in this work is the notion of becoming “streetwise,” meaning adept at subtly decoding the markers presented to one in the streets. At the crudest level, a resident uses race, or possibly race combined with class, as a key indicator of danger (or of opportunity, depending on what the observer is on the lookout for). But on Anderson’s account streetwise persons advance beyond this crude level, becoming shrewder at navigating the streets, thereby enabling themselves to sustain deeper and more meaningful relationships across the racial divide. An observer becomes streetwise by experimenting with nonracial markers, or perhaps more accurately, by supplementing racial markers with a wide array of nonracial ones that refine the discriminatory practice and permit more nuance in the treatment of those bearing a negative racial marker. Thus the white lady who runs into her apartment as soon as she sees a black kid approaching from across the street, clutching her bag close, looking furtively over her shoulder, is not “streetwise.” She hasn’t bothered to take any note of the signs: Is the kid carrying a book under his arm? How is he dressed? What are his gait and demeanor? She hasn’t learned about other, nonracial information that might be powerfully informative in that particular situation. “Street wisdom” is just a generalization of that remark.
This behavior—acquiring street wisdom—is surely commendable, one would think. We might all hope (and pray?) that those authorized to use deadly force on America’s city streets will soon acquire greater wisdom in this regard. But notice one thing. By eschewing stereotype-driven behavior and using a more refined set of indices to guide their discrimination, observers encourage the production of those very indices of differentiation by better-off members of the negatively stereotyped group, because they are the ones who gain most by separating themselves from the masses.17 I do not say here that this is necessarily a bad thing, though I can easily imagine circumstances in which it would be.
The strategies of social identity manipulation used by racially marked people to inhibit being stereotyped—their methods of “partial passing”—are endless: affectations of speech; dressing up to shop at a downtown store; writing more equations on the blackboard than needed, to show a skeptical audience that one does indeed have complete command of the discipline’s technical apparatus; “whistling Vivaldi” while walking along a city’s mean streets so as not to be mistaken for a thug18—most generally, adopting styles of self-presentation that aim to communicate “I’m not one of THEM; I’m one of YOU!” Such differentiating behavior can undermine a pernicious equilibrium in which the use of an intrinsically irrelevant racial trait has become institutionalized as a social convention.
But such strategies can also be a way to undermine solidarity in the race-marked population, and to encourage the selective out-migration (through subtle forms of “partial passing”) of the most talented. And they can promote a fractured ego, an insider’s own-group antipathy—“if only THEY would get their acts together, then people like ME wouldn’t have such a problem”—which is anything but pretty. When this process results in the pursuit of social mobility in a racially marked group by means of directed marriage patterns intended to preserve lightened skin tones over the generations (a commonplace of African-American society in years gone by which has yet to fully dissipate, and which can be found in Brazilian society even today); or when the better-off classes of the racially marked indulge themselves by preening obsessively over minute symbols of their relatively superior status (as the sociologist E. Franklin Frazier described nearly a half-century ago; Frazier 1957), the nature of the problem becomes apparent.
I wish to avoid misunderstanding. I do not intend these remarks to be an attack on the act of “partial passing” as such. Nor do I see that act as some kind of immoral betrayal. But neither can I celebrate it blithely as a glorious exercise in individual liberty, or as a god-sent mechanism for subverting an otherwise oppressive racial order. To the contrary, the tragedy of the selective out-migration of a relative few from the marked population through partial passing is that it places the burden of reforming a racially stereotypic order on those with little leverage to alter underlying social structures (letting our monopolistic observers off the hook). And, perhaps more importantly though certainly more speculatively, it promotes a liberal individualist ideology of personal achievement that reinforces, rather than challenges, an order in which the scourge of racial stigma can flourish.
So I hold there to be nothing wrong with individuals doing their best under trying circumstances by “passing” out of their stigmatized racial group through artful acts of selective self-presentation. Those doing so are not rightly indicted as reactionaries. But by the same token their behavior is unlikely to provide much political inspiration.
With these speculations in view, the key role that racial stigma plays in my argument should now be easier to see. In the next chapter I propose that durable racial inequality be understood as the outgrowth of a series of “vicious circles of cumulative causation.” The story goes something like this: The “social meaning of race”—that is, the tacit understandings associated with “blackness” in the public’s imagination, especially the negative connotations—biases the social cognitions and distorts the specifications of observing agents, inducing them to make causal misattributions detrimental to blacks. Observers have difficulty identifying with the plight of a people whom they mistakenly assume simply to be “reaping what they have sown.” This lack of empathy undermines public enthusiasm for egalitarian racial reform, thus encouraging the reproduction through time of racial inequality. Yet, absent such reforms, the low social conditions of (some) blacks persist, the negative social meanings ascribed to blackness are thereby reinforced, and so the racially biased social-cognitive processes are reproduced, completing the circle. As they navigate through the epistemological fog, observing agents find their cognitive sensibilities being influenced by history and culture, by social conditions, and by the continuing construction and transmission of civic narrative. Groping along, these observers—acting in varied roles, from that of economic agent to that of public citizen—“create facts” about race, even as they remain blind to their ability to unmake those facts and oblivious to the moral implications of their handiwork.
Calling this behavior racism, while doing little violence to the language, also fails to produce much insight. How, we should ask, will this self-reinforcing process be contested? Epithets are unlikely to be of much help. Subtle dynamics underlie racially biased social cognition—dynamics that are not much illuminated when conceived as some form of anti-black enmity. Note, for instance, that the argument to this point has made no reference to the race of the observer. Whereas a theory grounded in racial enmity would have trouble explaining anti-black sentiments held by other blacks, nothing in my theory prevents a black from succumbing to the same cognitive biases as anyone else, when drawing inferences about the underlying causes of racial inequality. Nor would I dismiss the possibility that perceptions by blacks of the larger society—of the opportunities available to them for upward mobility, for instance—might be distorted by racially conditioned causal misattributions on their part.
Here is yet another reason to resist the temptation to moralize when discussing these issues. I hold that it is more fruitful to focus on the cognitive rather than the normative aspects of this problem, attending to how people—often unreflectively—think about social information. So, anyway, I hope to persuade the reader with the argument to follow.