Abstract
There is a lively ongoing debate among philosophers and social scientists about the reality of race and among social scientists about the reality of caste and ethnicity. This paper tries to sort out what the issues are and makes some preliminary suggestions about what the evidence shows. Standard philosophical analyses try to find the necessary and sufficient conditions of our concept of race. I argue that this is not the best way to approach the issue and that the reality of these concepts should be taken as a scientific realism question; that is, do our best social scientific accounts of these phenomena show that appealing to the concepts of race, caste, and ethnicity is essential to successful social science explanation? I argue that in some cases that is the case and lay out the empirical issues involved.
Keywords
Philosophers have much discussed questions about the reality of race as well as what its reality consists in if indeed there are races. Parallel, though more empirical, debates have occurred in the social sciences. The social sciences also have ongoing debates about the reality of groups such as ethnic groups and castes that philosophers have not addressed, though many of the issues are the same as the questions about race. The general concern is whether any of these categorizations pick out something real when there seems to be an obvious sense in which these groups are socially constructed, when these categories are often ambiguously applied, when they are prone to changing meanings over time and context, and numerous other problems.
This paper addresses these debates. However, it does so in a rather different way than is standard in the philosophy literature. Philosophers have generally been concerned with our ordinary notions of race and the extent to which we should take those concepts realistically. My interest is instead in the philosophy of social science questions that groups such as races, castes, and ethnic groups raise rather than in the analyses of common-sense concepts. Social research on groups need not necessarily work with common-sense concepts any more than physics is restricted or constrained by folk physics. Of course, common-sense concepts can and probably should figure in some way in social science explanations—they are likely to be part of the explanans at least—but they are not the arbiters of those explanations. Looking at questions about groups from the philosophy of social science perspective, questions about the reality of race become largely questions of scientific realism—should we take social science accounts invoking group concepts realistically as telling us the way the world is or approximately is?
The paper proceeds as follows. Section 1 introduces the issues. Section 2 provides a general framework for thinking about categorization and classification in the social sciences. That is a big topic, and I only explore as much of it as needed to think about the reality of social groups. I then in section 3 use that framework to show that many of the objections to the reality of race, ethnicity, and caste need not be compelling: either they commit defenders of realism to things they do not have to assert or they raise empirical questions that cannot be answered on general conceptual grounds. Section 4 applies the framework to empirical issues around race, caste, and ethnicity. My final argument is that in some contexts, race, castes, and ethnicity are real. 1
Some admission of modesty is called for in the face of such a large topic. 2 The literature on race, caste, and ethnicity is enormous, the issues complex. Making some progress on sorting out the issues and arguments and giving a first pass look at the evidence I think would be a contribution. I do not claim to do more.
1. The Issues
There is an ongoing debate in the philosophy of social science and the social sciences themselves about the status of social and behavioral science categories and classifications. The worries are that social science categories do not pick out anything objective, that they do not allow for measurement, that they provide arbitrary and vague groupings of individuals, that they have no consistent definition across investigators, that they are social constructions in some sense that lack stability over time, and so on.
To give a couple of examples, the concepts of social capital and culture illustrate the worries. Social capital is a trendy idea in the social sciences. It builds on the idea of human capital which is well developed in economics. Social capital may be an intuitive idea perhaps, but it is one that has many different operationalizations and ambiguities, enough to raise doubts that it is picking out anything real. Similarly, the concept of culture has long faced such criticisms. I am not saying I agree, but you can certainly see where the antirealist sympathies come from. Moreover, it is easy to see how social constructivist accounts of these categories in terms of advancing disciplinary interests in a social context would go. Such arguments certainly have been made about the concept of culture.
These questions about the reality of social categories have produced a lively debate about race among philosophers and social scientists and parallel debates about closely related notions of ethnicity and caste among social scientists, but not philosophers so far as I know. This section describes numerous doubts about the reality of these social groups, documenting many of them in the literature.
Not in order of importance or frequency, doubts about the reality of the categories of race, caste, and ethnic groups claim that the three categories:
do not pick out substantial “things”
are not natural kinds
are not and cannot be collective agents
are not composed of individuals who define themselves as members of the category
are reifications of the individual and social practices that constitute racism and there is no such thing as race but only racism
are not constituted by a common culture
are socially constructed in various ways that show they are not real
exist only if there is collective agreement, common knowledge, collective recognition among those in the category and this usually fails
have unclear and variable criteria for membership
vary over time, context, and nation in how defined
have no biological basis although the concepts essentially require that
do not provide for systematic classification: not every individual in a society belongs to the categories in question and these categories do not form the basic building blocks of society
different social science measures of the categories vary in how they classify individuals
cannot ground causal explanations because these categories are fixed traits of individuals and there is no coherent account of causation for fixed traits
refer to phenomena that are not independently existing but rather the by-product of other social categories such as class and/or more individual level processes
inequality between these categories is not just due to discrimination
progress in ending discrimination due to legal changes or marketplace forces is making these groups increasingly unimportant to understanding social processes.
These claims are not always made about each of race, caste, and ethnicity, though their general logic suggests that they should generalize across all three. These claims are also not all independent and are related in complex ways and different ways from author to author. Yet these claims do form a “family resemblance” or the elements of—to use equally vague but suggestive notions—a research tradition or paradigm in the social sciences and philosophy of the social sciences that has significant influence.
Let me provide some references from the literature. Brubaker (2004) in Ethnicity without Groups and Loveman (1999) in a seminal earlier article on the reality of social groups both advance a number of the theses listed above. So Brubaker (2004, 3) says that races or ethnic groups are not bounded wholes, are not entities, and are not collective actors. They are not things in the world but perspectives on the world (Brubaker 2004, 17). Racial, ethnic, and other social groups are not real things but just categorizations. Pierson (1967) agrees that race does not exist in Brazil.
Loveman provides a similar viewpoint. “‘Race’ should be abandoned as a category of analysis” (Loveman 1999, 2); explanations in terms of race face “three critical pitfalls: (1) confounding categories with groups, (2) reifying ‘race,’ and (3) maintaining the unwarranted analytical distinction between ‘race’ and ‘ethnicity.’” These three flaws undermine the usefulness of a “‘racialized social system’ framework” (Loveman 1999, 2).
Webster (1993) adds to the chorus by claiming that race does not exist but that racial categories are the result of “racialization” by various groups, especially government and social scientists. Wimmer (2013) argues that races and ethnic groups are not substantial entities; what is real is boundary making that is generally contested.
Similar claims about the social construction of groups come from Zack (2003) who claims that race is not a fact in the world. Gracia (2017) sets up the issue about the reality of race as a debate about whether race is outside the mind or a mere social construction. Wimmer claims that those invoking ethnic group concepts can give an adequate account only if these categorizations make for a systematic typology. All individuals must be assigned to one ethnic group and the groups must be mutually exclusive (Wimmer 2013, 47), but racial and ethnic categories do not do that. Furthermore, true groups require subjectively felt belonging and that is often missing.
Among philosophers, at least the following support the antirealist program to some extent and in various ways. Atkin (2017) argues that we need to first get clear on our concept of race, where our concept is the common-sense notion. He also claims that if race is real, it must be a natural kind (Atkin 2012, 71), but it is not. Glasgow (2017) offers similar views, saying that any discussion of race is bound by our ordinary concept. Taylor (2013, 106) argues that groups are real only if they involve individuals that self-consciously recognize themselves as a member of the group and that collective agreement of some type is required. Mallon (2016) affirms this latter requirement in some form as does Elder-Vass (2011). 3
Critics also argue that each of these categories is really proxy for or reducible to other social phenomena. W. Wilson’s (1980) Declining Significance of Race is probably the best-known variant of this claim, with race being replaced with class as a determinant of outcomes. Race it is also suggested may just be a product of ethnicity (Loveman 1999) and ethnicity may just be a side effect of homophily in networks. Wimmer (2013) in Ethnic Boundary Making makes both claims.
So, there is a strong current in contemporary social science and philosophy that is antirealist and skeptical about important social categories and groups. These are doubts the philosophy of social science should be able to address and that are the goal of this paper.
2. A Framework
As I noted above, questions about how to think about social and behavioral science classifications and categories is a big topic; it may well not admit of a uniform analysis. Still, I want to sketch an approach to thinking about race, caste, and ethnicity that differs from some mainstream social science and philosophical analysis and which I think is more fruitful for thinking about these categories.
Discussions of race by philosophers standardly try to clarify our ordinary concept. This leads to various conclusions, but the most common is some form of antirealism about the category because our concept is biologically based in a way that is indefensible. However, it is more helpful I think to look at questions about the reality of race, ethnicity, and caste as questions of scientific realism: 4 does our best social science give us reasons to take these categories, as used in the social sciences, as real?
Working this way, my general approach is naturalist. It rejects the idea that at least some ontological questions can be decided by conceptual analysis of our everyday conceptions. Philosophy does not have a privileged position where it first decides which concepts are legitimate and what they mean prior to or independent of scientific investigation. Rather ontological commitment and fundamental categories are the result of what our best scientific inquiry is committed to. Quine’s “to be is to be a bound variable” is an instantiation of this view. So is the idea that what is real is what can stand in causal relations. Goodman, Dennett, and the psychometric tradition amplify this approach. For Goodman (1954) the real categories were those that were projectable. For Dennett (1991) they ground “real patterns” which he put in terms of nonredundant carriers of information. Cronbach and Meehl (1955) thought of valid categories as those that fit into a “nomological net”—they were predictive of other variables as well as predicted by other variables. Similar positions are advanced by Kitcher (1992) on biological classification and Khalidi (2013) on social classification who talks about “epistemic kinds.” The recent development of inferential approaches (e.g., Suarez 2008) likewise provides grounding: categories get at reality to the extent that they allow us to make successful inferences. Arguably these approaches to classification have historical roots in John Stuart Mill who said, “The ends of scientific classification are best answered, when the objects are formed into groups respecting which a greater number of general propositions can be made, and those propositions more important, than could be made respecting any other groups” (Mill 1882, 871-72).
Cronbach and Meehl represent the psychometric tradition which has been much refined by recent developments in analyzing causation, with the Pearl (2009) graphical modeling approach and embodiments in structural equation modeling (SEM) representing the best of this work in my view. An SEM generally has two parts: a measurement model for the variables and explicit causal links among the variables. The measurement model involves multiple independent indicators of the variables in question, with the variable typically seen as the common cause of the indicators (which are thus independent only when the common cause is screened off). The causal relations among the variables entail and are provided evidence by the specific conditional and nonconditional dependencies and independencies in these data. A well-supported SEM thus gives us what I will call objective explanations. While there are multiple senses of “objectivity,” here I am thinking of categories that we can access or measure by different means. “Explanations” should be read as categories that predict and are predicted by other objective categories, ideally in ways supporting causality. SEMs and causal modeling make projectability, real patterns, nomological nets, and similar notions considerably more concrete and in the best case concretely operational.
This picture of ontological commitment is an ideal. In the social and behavioral sciences we often do not have the kind of data necessary to support an explicit SEM. Instead, we have to use more qualitative data and more ad hoc methods. Yet it is useful to have the ideal even in more messy cases, for it tells us what general considerations we should be looking for when we make arguments that some social science category should or should not be taken realistically. Nonetheless, there is some successful work on race using explicit SEMs in public health and social psychology.
Three other general philosophy of science considerations will be helpful for our discussion. The first is that studies of actual scientific practices (e.g., Cartwright 1983) show that science often works without comprehensive theories and instead provides a patchwork of piecemeal causal accounts. So, we need to be wary of demands that social science accounts of race, caste, and ethnicity provide general theories. The second helpful point is to note that there are many useful scientific categories that cannot be defined in terms of necessary and sufficient conditions but are best thought of as prototypes or cluster concepts, with the gene being a prime example. Finally it is useful to see that we can think of social entities in two different ways: as groupings of individuals as a type sharing some specific common properties or as concrete particulars. Weberian notions of class typically take class to be a set of individuals sharing some common characteristic, for example, occupation. Alternatively, class might refer to a group of individuals with extensive interrelations which are sufficient to think of the collection as an entity; for example, there are good reasons to think that there is a ruling elite in the United States which is best thought of as a concrete particular (Kincaid 2016). Races, castes, and ethnic groups might be thought of in either way and it is important to keep these two conceptions clear.
3. Framework Applied to Some Common Doubts
This general naturalist approach suggests some clear answers to some of the doubts noted above about race and related social categories. The responses generally are that the objections saddle social science research with commitments they do not have to make in order to produce objective explanations and/or that the issues are contingent, open empirical ones that have to be investigated case by case about race, caste, ethnicity and that cannot be by conceptual argument and analysis.
Individuals placed in these categories do not have a common culture and do not uniformly identify with the groups that are invoked, so they cannot be real groups. The extent to which these claims are true depends on the specific applications of the three categories and is obviously an empirical question. There is no a priori reason that a lack of common culture must be true for every use of race, caste, and ethnicity. Furthermore, if our standard is providing objective explanations, then it is also an empirical question whether these characteristics are required to get successful explanations. So, for example, being gay was in many countries probably a fairly predictive category for certain phenomena even if there was not a common culture or self-identification with gays as a group. Similarly, there is good evidence for a ruling elite in the United States (Domhoff 2013; Kincaid 2016) even if they do not see themselves as a ruling elite with which they identify.
A social group that is real requires collective agreement, common knowledge, collective recognition or some such group intentionality that would make them collective agents but that is not present in races, ethnic groups, and castes. Such a requirement stems from a certain sort of analytic metaphysics applied to social entities, which thinks of social entities as real only if they are collective agents in some sense. My naturalist approach is suspicious of analytic metaphysics in general, finding that it is a questionable enterprise because it is based on intuitions with little ties to science. Yet putting that aside, I see no reason to believe that the only social entities that can ground objective explanations are ones that are collective agents in a metaphysically robust sense. We might find it useful for modeling purposes to treat social groups as maximizing a utility function (as is done in economics with firms), but even these more anemic approaches seem unnecessary for groups to be real. Instead, the question is whether social science using racial, caste, and ethnic categories can provide well-confirmed explanations.
The common-sense notion of race is biological, but there is no support for a biological conception of race. So, race is not real. On the naturalist approach, the common-sense notion is not decisive. There can be a social science notion of race that is not biological but nonetheless explanatory. Part of an explanation for how race works may no doubt want to give details about the common-sense notion in explaining individual behavior, but that is not the same as taking the common-sense notion as the only way or the best way to think about the reality of race.
Membership in the three categories is vaguely defined and changes over time. Again, two responses to this objection are appropriate: it is an empirical question to what extent these claims are true and to what extent, if they are true, they prevent objective explanations. Fuzzy categories can be scientifically quite useful; categories can change over time but still provide time-indexed objective explanations.
The three categories are socially constructed. There are three quite distinct possible notions of “socially constructed.” The thesis may be:
There are social processes involved in the social science studying race, ethnicity, and caste.
Those social processes make racial, and so forth, categories impositions, not objective explanations.
There are social processes involved in the phenomena of race, and so on.
Note first that (1) does not entail (2); the social processes involved in social science research might be epistemically reliable and thus not provide evidence that social science concepts are impositions. (3) is of course trivial and does not entail (2). The question then is whether the empirical evidence supports (2) for the three categories. There is probably no all-purpose answer. There certainly have been times when the social processes involved in social science led to the imposition of racial, caste, and ethnic categories, not objective explanations as the history of scientific racism illustrates. Yet it takes evidence and argument that something similar is occurring now for race, caste, and ethnicity. 5
The three categories do not provide a systematic taxonomy and do not constitute the fundamental building blocks of social analysis. They most certainly do not do so, though some have argued aggressively that caste does so for India (Dumont 1980). Yet I pointed out above that naturalist studies of science show that much science proceeds without comprehensive theories and that categories are often applied piecemeal according to context. A systematic taxonomy that grouped all individuals into one race, ethnicity, or caste and made those the fundamental elements of a complete social theory might be nice, but it is hardly required for the social sciences to get on with less ambitious explanatory projects.
Race, ethnicity, and castes are not natural kinds. The idea that scientific categories must pick out natural kinds grouping individuals seems to me a holdover from traditional analytic philosophy of science that was not very attuned to how science actually works. There are doubts that in the strong essentialist sense of necessary and sufficient properties even the elements of the periodic table are natural kinds. As noted earlier, some of our best natural science works with prototypes and fundamental concepts like the gene that have no clear delineation in terms of necessary and jointly sufficient characteristics. Obviously objective explanation can proceed without them. M. Wilson’s (2008) is a nice description of the scaffolding—his term—that is involved in applied physics to get categories tied down to reality that vary across contexts of application. I doubt that the social sciences have anything like this kind of sophistication. Yet the bottom line is whether we can use race, caste, and ethnicity to provided well-confirmed objective explanations and that is a question we have to answer by looking at actual empirical work.
If race, caste, and ethnicity are real, then they must be substances or things, but they are not. This claim is somewhat obscure. I have already noted that we might take the three categories as picking out types of individuals. Is a set of types of individuals a thing or substance? It would seem not. The set of all numbers with the property greater than 3 is a set of types of individual numbers. Is that set a thing? I also noted that at the opposite end of the spectrum we might take categories as referring to concrete particulars. For example, the Catalonians might be such a case where there are dense relationships among individuals, extensive social organization, and so on and thus the kind of traits that make for the cohesion we usually attribute to things. However, all these considerations very much resemble ordinary language or analytic metaphysics, but done in this case by social scientists. I do not see that social scientists have to work out this metaphysics to determine whether the three categories provide objective explanations. If I can provide diverse evidence that caste, for example, influences health outcomes, income, marriage patterns, and the like, then the key question has been answered.
The idea that the concept of race is a reification and that in fact there is no such thing as race but instead only racism is not a coherent position. Racism must involve treating individuals differently based on their race in ways that disadvantage members of racial groupings; society promotes disadvantage in a variety of ways and we have to distinguish among the many social processes by differentiating the different social groups involved if we are to understand discrimination. Discrimination based on racial difference is a separate phenomenon from discrimination based on sex or age, for example. Saying that there is no such thing as race but only discrimination does not allow us to understand the different forms of discrimination that currently exist.
To conclude this section, I have not responded to all the objections listed above and these responses can certainly be further developed. The next section will further expand on some of these and address some that have not been discussed as we turn to look at empirical evidence for objective explanations.
4. Varieties of Empirical Evidence for Objective Explanations
The prima facie evidence that race can ground objective explanations is extensive and it is obviously necessary to be selective in a paper like this. My first strategy is to focus on race in the United States, Brazil, and South Africa, countries where the influence of race is profound and well documented. (Of course, mere correlations do not show causation, but evidence of correlations plus relevant background knowledge can go some way in suggesting causal connections. Ideally causal inferences would be justified by full SEM models as I suggested elsewhere in this paper; such empirical work does exist but much more is needed.)
An overview of the evidence about the United States, Brazil, and South Africa from Gradin (2010) concludes:
In all countries, but more intensely in South Africa, African descendents are more likely than whites to be confined to the bottom of the income distribution . . . The differentials in income are increasing in absolute terms along the income scale in all three countries . . . The relative contribution of educational gap between blacks and whites appears to be the main explanatory factor in South Africa . . . Even if blacks had the same observed characteristics as whites in these three countries, a substantial (conditional) differential would still persist in average incomes.
In plain terms, there is evidence of current disadvantage due to past disadvantage and of current disadvantage due to current discrimination.
In what follows I detail some of that evidence. I will give the greatest attention to the United States, where the documentation and number of studies is greatest. 6
I start with the United States and give in turn evidence that race causes differential outcomes, that there is still ongoing explicit discrimination, and that there is a lock in, feedback systematic racism system resulting from past racial disadvantage that is continuing independently of ongoing explicit discrimination. I then repeat the same sequence of evidence for Brazil and South Africa, but in much less detail, in part for reasons of space but also as much because the amount of relevant research is noticeably smaller.
After having outlined this evidence, I then turn to possible doubts about its force. Here I consider those doubts across all three countries simultaneously.
Why race matters in the United States is an open question, though I think there is clear evidence that there are two transmission mechanisms as it were, one probably stronger than the other in some intuitive sense (comparisons of causal importance are often perplexing as I discuss below in the context of race and class). The weaker mechanism is probably ongoing explicit discrimination resulting from conscious or unconscious decisions by whites to favor their own over African Americans in employment or credit, for example. Stronger systematic forces emerge from the past history of systematic discrimination that has produced lock in-feedback processes that ensure that outcomes will be worse for African Americans. In any case I present evidence for both for the United States.
Active discrimination by whites against African Americans has not gone away, despite numerous claims to the contrary. 7 “Audit studies” are a quite compelling form of evidence. They basically involve sending black and white applicants with identical educational, financial, and so on characteristics out to apply for housing, loans, or employment. The results show quite significant current discrimination in all three areas (Bonilla-Silva 2014). There is good evidence of ongoing discriminatory practices in public education as illustrated, for example, by the treatment of students by teachers (Lewis and Diamond 2013). Statistical studies of wages consistently show evidence of discrimination, and those studies probably underestimate the problem as I explain below. Discrimination in the judicial system is widely documented.
There is also substantive experimental evidence that at least implicit racial biases continue to operate. Numerous social-psychological experiments among U.S. subjects and even functional magnetic resonance imaging (fMRI) studies show that prejudice has deep psychological roots (Mallon and Kelly 2012).
No doubt some significant part of the ongoing discrimination is due to either conscious or unconscious racial attitudes and beliefs. However, it is not an easy matter to separate current disadvantage due to current discrimination from current disadvantage due to legacies of the past. That difficulty, however, does not matter for my argument that race is real because it grounds objective explanations.
So, I turn now to legacies of past discrimination. The basic plot line argues that past discrimination set up a series of conditions that are now locked in—there is a past disadvantage that involves positive feedback loops that ensures that it continues into the indefinite future. So, in the following areas systematic past discrimination makes for continuing discrimination:
4.1. Wealth
The history of slavery and the post-slavery history of legally sanctioned segregation and discrimination means that current African Americans are at a serious disadvantage in terms of inherited wealth which is a significant component of current wealth across the board. The majority of the wealth differences between whites and blacks in the United States can be traced back to wealth levels at the time of emancipation. The Homestead Act gave white families land that was then inherited and constitutes a significant fraction of current household wealth in the United States; blacks were excluded. Housing ownership constitutes a main source of household wealth accumulation. Furthermore, white neighborhood associations with assistance from banks, realtors, and the Federal Housing Authority (FHA) prevented African Americans from buying houses in neighborhoods where home ownership produced the highest returns on investment and generally made home ownership difficult and more expensive for African Americans (Roithmayr 2014). Given the compound nature of returns to wealth, these past effects of race are guaranteed to continue and increase differential levels of wealth indefinitely into the future. Moreover, wealth levels are major influences on the quality of neighborhoods that current individuals can afford and on the ability to obtain postsecondary education. So they build in ongoing differences in racial outcomes.
4.2. Spatial Segregation
Wealth differences, past legal segregation, neighborhood association enforced segregation, past and current federal housing policy, and other mechanisms promoting segregated neighborhoods result in a large percentage of African Americans living apart from whites in largely poor urban neighborhoods. The spatial concentration of poverty is self-perpetuating because these neighborhoods have fewer resources for education, since education is supported by property taxes, and these neighborhoods have fewer government resources, have higher crime rates, and have less access to employment. There is considerable evidence that spatial segregation along with the ensuing poverty, high unemployment, high crime exposure, high exposure to environmental toxins, and poor quality schools are “heritable”—current residents of poor, racially segregated neighborhoods largely have parents who grew up in similar conditions (Sharkey 2013). Of course, there are multiple causes for these conditions aside from race, but race is the major factor. Spatial segregation exists in large part because of race and thus race grounds an enduring mechanism making for ongoing differences in outcomes.
4.3. Social Networks
There is an extensive body of evidence that social networks matter for individual outcomes in a variety of ways. Networks vary in size and in the characteristics of the individuals in them. Networks may vary in terms of the income, status, access to information, levels of education, and power, such as authority status in employment and access to government officials. Perhaps the most important influence of networks is on employment, a literature developed from Granovetter’s (1983) seminal article on networks and employment. The size and quality of a network influences the prospects of getting a job and the quality of jobs gotten (Roithmayr 2014).
Not surprisingly the legacy of past disadvantage means African Americans have significantly smaller and lower quality networks, thus making employment and job quality likely to be inferior to whites. It is easy to see why this would be so, because it is reinforced by other aspects of systematic disadvantage. Spatial segregation and the ensuing lower neighborhood quality, lower quality and extent of education, and lower parental levels of employment, income, and wealth are all likely to lead to smaller and less socially valuable networks.
These mechanisms suggest that we can have evidence for the objective status of race without evidence of explicit racist attitudes and practices. Racial disparities in the United States are institutionalized.
There are interesting lessons here for the individualism-holism debate. The lessons concern the question of just how far we can get in explaining social process without invoking social entities, institutions, organizations, structures, and so forth—how far can we get by simply explaining via the traits of individuals? It is fairly clear in the accounts given above that institutions such as the FHA, political elites, white racist organizations such as neighborhood organizations and vigilante groups, and so on played a fundamental role in producing racial disparities. Obviously all these resulted from the actions of individuals, but putting these explanations purely in terms of individuals would appear to be a pipe dream. This is in part what it means to say that racism is systemic, even if individual attitudes still matter.
I turn now to make a similar case for the objective explanatory power of race in Brazil and South Africa. I provide much less detail, because there is less research, because the outline of the U.S. case makes it clear how the argument would go, and for reasons of space.
Looking at Brazil, there is extensive evidence of similar inequalities to those in the United States. Education, income, access to jobs, and social mobility are statistically and substantively different between whites and nonwhites (Telles 2004). Brazil, like South Africa, has two nonwhite races—brown and black—and there are distinctions in inequality between them, but the most glaring inequality is between whites/nonwhites. This is a nice instance of the categorical pluralism defended above: a three-tier categorization makes for objective explanations as does a two tier and there is no reason they must be incompatible. There is also substantial evidence that systematic racism—locked in effects due to past inequalities—is important (Telles 2004).
Finally, turning to South Africa, there is overwhelming evidence of both structural racial disadvantage and ongoing discrimination. Apartheid involved an explicit state-led policy of favoring whites over blacks and over “coloured” 8 in extreme degrees. Those policies have a lasting legacy which is in a sense directly observable in South Africa. Africans were forcibly spatially isolated to either the worst agricultural land or to townships—shanty towns—in urban areas, though the townships were mostly located some distance from the urban centers. There has been no meaningful land reform, and the spatial segregation imposed by the township structure remains. To give a concrete example, there is little formal or informal employment in the Cape Town townships, and it is nearly a 2-hr taxi ride to Cape Town where there are employment possibilities. The legacy effects are enormous.
There is also evidence of ongoing discrimination in South Africa. 9 Using surveys conducted in 1994, 2001, and 2011, Naidoo, Stanwix, and Yu (2014) find that labor market discrimination has decreased but nonetheless is still significant.
The evidence that race grounds objective explanations in the United States, South Africa, and Brazil may actually be stronger than what has been presented above, because some standard methodologies used in the many studies reported above are based on impoverished pictures of causation and causal inference. These worries will equally hold for studies of caste and ethnicity. Typically, specific outcomes such as wages are treated as dependent, left hand-side variables and multiple regressions are run with possible causes as right-side variables. Race then may be one of those variables and if so, the influence of race is taken as being measured by the size of the race coefficient when other possible causes are included in the regression. 10 It is not uncommon based on this methodology to claim that outcomes for blacks can be shown to be due not to race but instead to other factors.
There are multiple worries about this methodology. The first is that it really only gets at current, ongoing discrimination—that is, it does not get at systematic racism issues. For example, the legacy of disadvantage is an important determinant of current educational achievement in that parent’s education, income, wealth, neighborhood, and so on influence the achievement of their children. Frequently all this is left out of the model.
Second, these methods can provide biased estimates of the current effect of discrimination on outcomes. Unfortunately, the multiple regression methods (or related statistical tools) used are not reliable indicators of causality and in fact make systematic errors in certain circumstances. Consider, for example, the claim that is sometimes made that race does not significantly influence educational outcomes and that, for example, they are instead explained by school quality, socioeconomic status (SES), and related variables. These are the kinds of claims that Wilson makes. It is the “instead” in this analysis that is problematic. A plausible causal picture would be that race is a major causal influence on school quality—given the history of racism and segregation based on race in the United States, skin color was a major factor determining school quality. Thus, a reasonable picture is that:
Race → school quality → educational outcomes.
We now understand that because of advances in causal modeling (Pearl 2009) testing such a picture by multiple regression methods produces serious errors if we are making causal inferences. School quality is an intervening variable. Multiple regression will give us the correlation of race with educational outcomes conditioning on school quality or, in other words, holding it constant. The result will be that the correlation of race with educational outcomes will be screened off and our multiple regression exercise will show little to no correlation between race and outcomes. We will thus conclude that race is not an important influence on outcomes. If race influences SES and other such variables, regressing on them will further diminish the correlation between race and outcomes. If we follow the multiple regression methodology naively (which is not uncommon), we conclude that race is not that important. There are multiple ways to deal with these problems—mediation analysis or fitting full SEMs—but that is not usually done in studies of race. The unfortunate upshot: we wrongly conclude that race is not an important cause of social outcomes.
I do not know how often this kind of error leads to misleading conclusions about the causal effect of race. That would take a larger investigation than can be done here. Nonetheless, the point is that the evidence I have provided that the concept of race allows objective explanations probably understates the case, given that much of the evidence relies on the multiple regression framework, naively interpreted.
The claim that race grounds objective explanations needs to be qualified in various ways, but these qualifications do not mean that the explanations do not argue for taking race as real. A first qualification was mentioned in the previous section: establishing the objective existence of some races at some time and place does not mean those races exist across time and societies. We will address debates below about how universal the black/white phenomena are, but it is perfectly coherent to claim that these races are real in the sense of grounding objective explanations in contemporary American society but not in contemporary India or South Korea or to claim that black/white races did not exist in South Africa in the 19th century. Indeed, there is good evidence that is true. In the 19th century there were instead ethnic groups: Afrikaner, British, South Asian, and various Bantu expansion African groups. It was only in the 20th century that a white/nonwhite racial formation was created, done by Afrikaners and the British together to ensure their joint domination.
A second qualification which follows from the general perspective outlined in the last section is that appeal to races can be explanatory without everyone in a given society having a clear racial classification. In the contemporary United States “Asians” are heterogeneous in ways that make it doubtful that appeal to Asians as a race has much explanatory power. Yet that does not impugn the explanatory value of talking about whites and blacks in the United States. Recall that there is no reason that theories invoking race have to assert that explanations in terms of race form a systematic typology in the traditional taxometric sense nor that races are the basic building blocks or fundamental particles of social theory.
I argued above that the main objections raised against the reality of the three social groups generally depend on contingent empirical issues that have to be evaluated case by case. I turn to that task now for race in the United States, Brazil, and South Africa. I start with one argument that is a direct assault on my claims above that there is evidence of objective explanations grounded in racial categories. The criticism is straightforward: showing causality requires showing how variation in states of the entities in question would produce variations in outcomes for those individuals. This is sometimes called a potential outcomes or counterfactual approach to causality (Morgan and Winship 2014). Since race is fixed, no sense can be made of what would happen if individuals were of a different race and so claims that races are causes are nonsensical. Holland (2008), who is a vocal advocate of the potential outcomes approach to causation, is the intellectual source of this objection, but it has been unfortunately picked up by others (e.g., Zuberi and Bonilla-Silva 2008).
The objection is bogus. Holland himself comes close to showing why. He grants that there is evidence of discrimination, that it is a causal influence on outcomes, and that there is statistical causal interaction between factors (this is his phrasing) associated with discrimination and race. This sounds very much like saying that outcomes for individuals are influenced, that is, caused, by their race.
It has been conclusively established that the potential outcomes approach is logically equivalent with structural equation approaches to causation—each can be translated into the other (Pearl 2009). The counterfactuals of the potential outcomes approach have a natural interpretation in terms of SEMs. However, it is perfectly clear in SEMs that race can be a causal variable. There are dependencies and independencies implied by causal graphs with race as a variable, and these data can either support those dependencies and independencies or not. This is one of our best tools for thinking about causation and it makes it entirely clear that race (and gender and many other invariant characteristics of individuals) can be a cause. After all, the charge of an electron is a fixed characteristic of it, and it would be ludicrous to claim that those charges cannot be causes.
Another common objection to the reality of race is based on measurement—the claims that racial categories make arbitrary assumptions about who falls into racial categories and who does not. I make no claim that this problem is never real for any use of racial categories, only that it does not undermine the evidence I cited above for the reality of the black/white racial categorizations in the United States, South Africa, and Brazil. In both instances the doubt can be answered straightforwardly by two means. For the United States, studies show that self-reported race as African American and race as assessed by surveyors overlap by 90% and that results show that predictive ability holds up across measures (National Research Council 2004). For Brazil with its more complicated racial system, classifications as white/nonwhite by self- assessment and by surveyors have high overlap (generally around 80%) as do these measures with color chart measures of skin tone and of hair type (Telles 2004). Again, predictions hold up whichever measure is used (Telles 2004). For example, levels of education in Brazil differ by race on each of the four measures in roughly the same magnitudes. In South Africa, race is almost always self-assessed in studies, but the large National Income Dynamics Study following over 20,000 individuals over 8 years shows very few individuals changing their reported race over that time period (M. Brown, “pers. com.,” 2017). So, racial categories in these countries seem indeed to be stable.
There are two further responses to worries about vagueness in group membership. The “rainforest realism” of Ross (2004) scan helps us see why multiple measures that are not perfectly correlated should not necessarily bother us. If there can be multiple ways of dividing up the social world that provide information, then it is not surprising that we have measures which nonetheless are not perfectly correlated. Similarly, the recognition that many good scientific concepts are perhaps best thought of as prototypes—one way of thinking of Weber’s ideal types—suggests that categorization can be implemented differently for what seems the same category.
Moving on to other alleged obstacles to the claim that racial classifications provide objective explanations, I want to look at two related objections together: the claims that racial conceptions change over time and that race is socially constructed. As I pointed out above, neither worry is telling automatically. The question is whether time-varying categorizations can provide objective explanations and whether race is socially constructed in a sense that prevents objective explanations. The idea of the skeptics is that racial categories change in their meaning over time and that in addition racial categories are social constructions in the strong sense that shows they have no objective reality. Homosexuality was a socially constructed disorder—a pure labeling exercise when there never was a disorder. Race might be like that.
There is no doubt that current races arose and (probably) will disappear over historical time; they are not the timeless atoms of society. I have already mentioned that the division in South Africa of people into white/nonwhite races only took on full social reality in the late 19th/early 20th century and obviously current racial categories in Brazil and the United States have origins somewhat earlier. It is also certainly true that racial categorization was promoted by governments and elite ruling whites in the cases discussed above of the United States, Brazil, and South Africa and there were changes in how racial concepts got applied over time. There is a line of argument that government categories, especially the Census, and other institutions racialized people and that this shows that races are somehow not real. Loveman (1999) argues this for some Latin American countries and for the United States.
Yet these facts are telling only if racial classification is very time sensitive and does not involve sufficient social reality to ground objective explanations. Again, we have to remember that claims about the reality of race have to get specific. The U.S. Census categorization of Asian or “Other” or the Brazilian classification may be both socially constructed and time variable in such a way that they do not ground objective explanations. However, I think the evidence is very strong that the racial categories of white and black in the United States and South Africa and white and nonwhite in Brazil (recall this distinction in Brazil made at the beginning of this section) are far more than just bureaucratic government impositions. First, in Brazil the government vigorously promoted an ideology of racial mixture—the idea that race had disappeared and that the people of the country were Brazilians without racial divisions (governments and elites had self-interested reasons for doing so), thus undermining the idea that races are socially constructed by government classification systems. Instead, there are deep-seated realities maintained by a host of mechanisms centrally involving government policies and elite white influence in a systematic way which have all the causal outcomes and causal precursors cited at the beginning of the section. If there has been some variance over time in which individuals would be put into which categories, that does not mean that at any specific time the categorization did not make major differences in life chances and so on.
As I noted earlier, it is certainly true in South Africa that whites, especially white elites in the economy and government, promoted racial classifications that helped create social stratification along apartheid lines. Yet the point is that they were successful in many ways, and I provided evidence earlier that racial divisions in South Africa had and still have many significant causal effects and are still sustained by institutional and organizational mechanisms inherited from apartheid and colonialism. There were and still are ways in which the racial classification of some individuals is arbitrary and in that sense a social construct in the more radical sense of classifications without reality. However, that does not show that the basic racial system is an arbitrary classification without social reality and social effects.
Moreover, doubts about the existence of races because they come and go over historical time are also quite unconvincing. Race in the United States and Brazil has been around for over 200 years and for more than a century in South Africa. Throughout that time the evidence has been that race grounds objective explanations.
A final objection to my evidence for objective racial categories claims that these results are actually proxies for some more fundamental relationship. So, it is not race that is influencing life chances but class, ethnicity, spatial networks where individuals just happen to associate with others who are like them, and so on. These are complicated issues, so my goal is just to sketch some prima facie responses.
W. Wilson (1980) is the best-known advocate of the idea that it is largely class, rather than race, that explains the situations of black Americans, though in later work he qualifies these conclusions somewhat. Research in Brazil has a strong tradition of treating race as an epiphenomenon to class (Telles 2004, 22). Seekings and Nattrass (2000) argue that class has replaced race in post-apartheid South Africa when it comes to distribution of resources.
Care is called for here. Making clear causal claims takes work. There are many outcomes that differ between whites and nonwhites—income, education, health, neighborhood quality, and so on—and there is no reason the relation between race and class should be the same for all.
Moreover, the causal interactions between race and class are complex in ways not generally distinguished. Those who think that race has been replaced in determining outcomes generally argue that current discrimination has waned and class, usually thought of in roughly the Weberian sense of occupational location, now determines outcomes. However, “discrimination” is a slippery concept. Those who claim that class trumps race think of discrimination as explicit, largely intentional practices such as favoring whites over blacks in hiring when their qualifications are equal. Yet we saw above that the mechanisms behind racial disadvantage are far more complex than this. If we take discrimination in the broader sense of differential outcomes caused by race regardless of whether the results are motivated by explicitly racist attitudes, then declines in explicit racist treatment do not show that current outcomes are not still the results of discrimination. The results of past explicit intentional racial discrimination are locked in to current circumstances—wealth levels, neighborhood quality, social networks, and the other factors cited above—in ways that guarantee racial disparities in outcomes (Feagin 2006).
Thus, the claim that class matters more than race assumes a very simple and I would say naive causal model. Consider the two possibilities in Figure 1 (below).
Possibility 1 does not include past discrimination. Relative importance of race and class is measured by the size of causal coefficients A and B. Possibility 2 includes past discrimination and shows that judgments about the importance of race and class must consider more than the relative sizes of coefficients A and B.
Possibility 1 is the claim that currently class matters more as signified by the larger coefficient on the causal arrow from class to outcomes compared with that of the race to outcomes causal variable. However, possibility 1 is a very simple model that ignores the past. Possibility 2 includes the past and a much richer model appears. Class is still more important than race looking at contemporary causation. However, possibility 2 includes past causes and produces a much more sophisticated model. Now race has many more causal influences. I do not know in this more sophisticated model how to measure the importance of race compared with class—I do not think anybody does, but would love to be proven wrong—nonetheless, intuitively it seems obvious that in the full model class no longer trumps race.
Thus, arguments that race does not matter or is of minimal importance rest on simplified causal assumptions that undermine the results in ways similar to the problem noted earlier for multiple regression-based studies of the causal influence of race. It is entirely reasonable to think that there is a causal process that runs racial ancestry → income class → outcomes. As we saw, if that is true, then standard studies using multiple regression will find that race has no or little effect, since they screen off the influence of race. Qualitative, verbal arguments will run into the same problem. A researcher can note that the differences in black versus white outcomes are also differences in black versus white class and then conclude that it is class that matters. The same error is at work and this is a problem for W. Wilson’s (1980) arguments. The class differences may be the result of race—or better, the entire racial stratification system of institutions, spatial segregation, and so on—and so the inference that race does not matter but only class does makes a big mistake.
I turn in finishing to the empirical evidence about the objective explanatory power of caste and ethnicity classifications. The previous sections dealt with general objections to all three categories and to that extent caste and ethnicity have been defended as potentially picking out something real. However, I have argued that even if there are no in-principle obstacles to explaining with these three social categories, the key evidence is empirical and case by case. In this section, I have shown in some detail what that kind of evidence would look like for race in three countries. Providing that kind of detail for caste and ethnicity goes beyond the space available. What I can do is instead to give a brief picture of what evidence would be needed for caste and ethnicity.
Looking at caste, we need first to discuss what we mean by caste, for there are multiple conceptions. The most basic distinction is between varna and jati. The varna is a classification supposedly derived from ancient vedic texts which identifies four groups: the Brahmins who are priests, scholars, and teachers; the Kshatriyas who are rulers, warriors, and administrators; the Vaishyas who are merchants; and the Shudras who are laborers and service providers. These are ranked in terms of status, with the Brahmins on top, where status is associated with purity. Distinct from the varna classifications are the jati. There are several thousand jati, which above all are endogamous marriage groups. Jati are traditionally associated with specific professions or jobs, though that does not mean every member of a jati has that specific profession. Jatis also differ in status along purity lines, though relative status varies across regions and changes over time.
When talking about the reality of caste, there are multiple different questions we might ask. We might want to know for all of the many jatis—and there are hundreds—if they are real. Alternatively, we might instead ask if the caste system itself is real without requiring a commitment to the reality of specific castes. Finally, we might want to know if the varna distinctions are real, in particular whether the untouchable/higher castes distinction picks out real social entities.
The clearest evidence will be for the existence of the caste system and the untouchable/higher caste grouping, with questions about the reality of each of the specific jatis being a project that is immensely large. For the first two claims the evidence is fairly overwhelming. Marriage, rituals, social association, and residential location are still very much determined by caste membership (Bayley 1999). Occupation is likewise still strongly influenced, though less so than in earlier history. Outcomes are also still strongly influenced by caste. So, for dalits and other scheduled classes (these are the “untouchables” which in a sense have no ranking in the caste system), there is clear evidence that:
Health outcomes are significantly different between dalits and higher castes (Borooah 2010).
Wages and access to occupations in urban labor markets differ strongly across dalits and higher castes (Madheswaran and Attewell 2010).
In rural markets, there is differential treatment by caste of both producers and consumers, with dalits having diminished access to production inputs, higher prices for those inputs, lower prices for outputs, and lower wages for agricultural labor (Thorat, Mahamallik, and Sadana 2010).
In general dalits are denied property rights, with strong consequences for self-employment (Thorat, Kundu, and Sadana 2010).
Dalits are denied equal access to education and when they do obtain schooling there is discrimination in terms of treatment and access to resources (Nambissan 2010).
In the case of caste, perhaps unlike race in the United States, Brazil, and South Africa, ongoing discrimination and systematic disadvantage built in by past legacies are both obviously present. There is little doubt that the past caste status of one’s parents influences one’s prospects due to restrictions on access to resources, education, networks, and so on. However, I think the ongoing active differentiation between people based on caste is much stronger than differential treatment of races in Brazil and the United States.
Compared with race, the skeptical objections about the reality of the classifications involved in caste should be easier to respond to. There is a literature arguing for social construction, pointing to the role of the British in codifying caste categories (Dirks 2001). Yet even if specific jatis were solidified by British practices, nonetheless it is doubtful that those castes are not of enduring influence and it is certainly clear that the caste system and the distinction between higher castes and dalits predates and exists independently of colonial influences (Bayley 1999). While castes certainly can have economic benefits and consequences, the prospects for reducing caste to class seem slim.
These are the kinds of considerations that would go into showing that the category of caste grounds objective explanations. Of course, a full defense would need to be much longer, however the discussion here should give an idea how the story would go.
If caste seems an easier case than race, ethnicity is probably the most complicated of the three categories in terms of arguing for reality. Given the range, diversity, and widespread ethnic phenomena globally, a selective sampling looking at the strongest and weakest cases would be necessary. Of course, earlier sections answered all-purpose skeptical objections. Still needed is specific empirical evidence about the role of ethnicity in various cases and objections to it.
As with race and caste, the ideal evidence is for objective—independently measurable—categories that ground explanations, ideally causal explanations. In the case of ethnicity, evidence of independent measures can come from showing that groups of individuals have, for example, common history, marriage patterns, social and economic relationships, culture, institutions, language, residential patterns, and religion. These are intuitively independent indicators that can pick out objective groups.
The majority of evidence for the causal importance of ethnic groups is historical, claiming to show that major societal events were driven by ethnic membership (Horowitz 2000). Internationally there is prima facie a wealth of instances where ethnic groups were causes and consequences of important social phenomena—ethnic conflict, regime change, migration, and so on.
As with race and caste, there are various doubts about the above empirical evidence on ethnicity. For one, there is a wide range of ethnic phenomena, and there is certainly no guaranteeing that all of them will have the kind of evidence just described. Clearly in some cases ethnic membership does not explain much, though of course again this is a case by case empirical issue.
One major doubt about the empirical evidence parallels doubts about race which claim to show that race is not an important causal variable but rather is just a proxy for some deeper variable. Brubaker (2004), for example, says that ethnic groups do not cause violence but violence causes ethnic groups via specific organizations and individuals pursuing their own agendas promoting ethnic divides. Horowitz (2000) presents a variety of evidence for the claim that ethnic group phenomena are not just the product of broader populations being manipulated by elites. This is one example of the kind of empirical issues that must be addressed to decide whether ethnic groups ground objective explanations.
5. Conclusion
There is a significant trend among philosophers and social scientists to doubt the reality of social groups. Philosophers have doubted the reality of race, and social scientists the reality of race, caste, and ethnicity. These doubts are on general, conceptual grounds that are misguided. Serious doubts have to be based on specific empirical evidence showing that these categories do not provide evidence for objective explanations invoking these categories. There is considerable tentative empirical evidence that the three categories do indeed pick out real groupings in a range of circumstances. When and where that is the case is an open and important question for investigation.
Footnotes
Acknowledgements
Thanks to the Philosophy of Social Science Round Table group for helpful comments and for being an important contributor to the field for many years and to Ken Smith and two anonymous referees for valuable comments.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
1
The form of social constructivist realism defended here is not new to me. Khalidi (2013) and 
defend versions of the basic position. What is relatively new is the detailed look at doubts from the social sciences, the detailed examination of social science evidence, and in the case of Hardimon, arguing from a philosophy of science and scientific realism stance rather than traditional conceptual analysis. Also, I should note that there is a tough ethical issue in writing on race. Arguing that there is a legitimate notion of race may contribute to legitimating harmful racial stereotyping. However, denying that there is any such thing as race cuts the other way, promoting visions of the social world where race plays no role. My sense is that the latter problem much outweighs the former.
2
Modesty is also called for because there is no discussion of the role of gender in thinking about race, caste, and ethnicity. I can in part point to the fact that the social science (and philosophical) work on these categories has made little to no systematic attempt to bring in gender. That is not an altogether happy excuse, and bringing gender issues into the debate over these categories will hopefully be a topic for future research.
3
Haslanger (2005) is an interesting case in this regard. She is clearly in the conceptual analysis tested against intuition tradition that I believe to be a dead end. However, she wants to move from that tradition to something more subtle by combining externalist semantics, Foucault-style genealogy, and normative political philosophy. I appreciate the attempt—certainly an empirically realistic account of how racial categories work in practice is valuable and much of the social science literature on race that I cite makes efforts in that direction. However, in her article () none of that literature is cited. Moreover, the well-known objections to Kripke style semantics, the fuzziness and difficulty of doing Foucault-style genealogies, and the complexities that come from basing accounts of race on political principles make her project suspect.
4
Realism is multiple different doctrines. The version in the literature that comes closest to the realism advocated here says that we have good reason to believe our current theories are true. However, that version claims more than I want to here, for on my view much of the social sciences are about theories in only an anemic sense. Rather social science often produces piecemeal causal claims instead. I think that is true of work invoking race, ethnicity, and caste.
5
This kind of moderate social constructivism has been well defended elsewhere by others (Elder-Vass 2011; ).
6
An interesting and perhaps unanswerable question is whether we can say race is more important in one country than another. That may be an unanswerable speculative question, but my gut reaction is that South Africa stands out. I think it would be possible to do comparative racial differences in outcomes across countries by race.
7
8
“Coloured” refers to individuals of mixed origin whose ancestors were Indonesian, white, and Khoisan.
9
says that “Race remains very important in cultural and social terms, but no longer structures economic advantage and disadvantage.” However, no substantial evidence is provided that race is no longer economically important, and given both the obvious structural racism legacy and the existence of clear studies showing continuing ongoing labor market discrimination, the claim is ludicrous.
10
There are multiple, more complex ways of looking for discrimination. Considering them is beyond my scope here but a project for future investigation.