Limitations of Civil Rights Law

Federal equal employment opportunity (EEO) laws attempt to prevent and redress discrimination in employment, a major source of economic and social inequality. A growing body of literature analyzes the capacity and limitations of EEO law to ameliorate inequality in the workplace. EEO law has, to some extent, improved women's and nonwhites' position in the labor market and workforce (Reference BellerBeller 1982; Reference Burstein and EdwardsBurstein & Edwards 1994; Reference Eberts and StoneEberts & Stone 1985; Reference Heckman and VerkerkeHeckman & Verkerke 1990; Reference LeonardLeonard 1984, Reference Leonard1986; Reference SkaggsSkaggs 2008, Reference Skaggs2009). However, a large body of legal and sociolegal literature highlights the many limitations of civil rights law in redressing inequalities at work. Some limitations stem from power disparities in litigation. Sociologists of law point out that the rights-mobilization process, in which victims must generally perceive rights violations and endure the prolonged and costly process of litigation in order to realize the benefits of legal rights, tends to penalize precisely those who rights are intended to benefit: individuals with fewer social, economic, and political resources (Reference AlbistonAlbiston 1999, Reference Albiston2005; Reference BumillerBumiller 1987, Reference Bumiller1988; Reference Felstiner1980/1981; Galanter Reference GalanterFelstiner et al. 1974; Reference MarshallMarshall 2005; Reference Miller and SaratMiller & Sarat 1980; Nielsen Reference Nielsen, Fleury-Steiner and Nielsen2004a, Reference Nielsen and Sarat2004b; Reference NielsenNielsen et al. 2010; Reference ScheingoldScheingold 1974). A second set of limitations of EEO law's potential to redress inequalities stems from mismatches between discrimination as conceptualized by law and discrimination as experienced in the workplace. EEO law generally conceptualizes discrimination as intentional, disregards its structural forms, and fails to recognize how employment practices sustain patterns of market-based discrimination (Reference Haney LopezHaney Lopez 2000; Reference KriegerKrieger 1995; Reference Nelson and BridgesNelson & Bridges 1999; Reference SchultzSchultz 1990; Reference Schultz and PettersonSchultz & Petterson 1992; Reference SturmSturm 2001; Reference Sturm and GuinierSturm & Guinier 1996). Additionally, courts may help to institutionalize ineffective organizational responses to law by deferring to compliance structures that symbolize attention to law but are ineffective at combating discrimination (Reference Edelman, Nelson and NielsenEdelman 2005, Reference Edelman and O'Brien2007; Reference EdelmanEdelman et al. 1999, Reference Edelman2011; Reference KalevKalev et al. 2006).

In this article, we turn our attention to a stumbling block for EEO law that has received much attention in critical race and feminist scholarship but relatively little attention in empirical analyses of inequality in the courts: intersectionality. As described below, intersectionality disadvantages plaintiffs both as a source of inequality in litigation and as a mismatch between legal conceptualizations and actual experiences of discrimination.

Twenty years of work by sociolegal scholars suggests that plaintiffs who face multiple disadvantages fare less well in civil rights litigation than do plaintiffs who suffer a single form of social disadvantage (Reference AustinAustin 1989; Reference CaldwellCaldwell 1991; Reference Carbado and GulatiCarbado & Gulati 2001; Reference CrenshawCrenshaw 1989, Reference Crenshaw1991, Reference Crenshaw1992; Reference HarrisHarris 1997; Reference RobertsRoberts 1991; Reference SmithSmith 1991; Reference WeiWei 1996; Reference WilliamsWilliams 1991). The key insight of intersectionality theory is that discrimination and disadvantage are not just additive; categories may intersect to produce unique forms of disadvantage. For example, an employer might hire both white women and black men but refuse to hire black women because he stereotypes them as desperate single mothers (Reference KennellyKennelly 1999); since this stereotype is specific to black women, it cannot be explained as the summed effects of racism and sexism.Footnote ¹

Intersectionality theorists have suggested two distinct processes through which people facing multiple disadvantages are subordinated in the courts, but have not explicitly distinguished between these types, causing some confusion. To highlight the fact that intersectional disadvantages comprise two distinct mechanisms of subordination, we formulate and investigate two different constructs: demographic intersectionality, in which the courts are the site of intersectional disadvantages or discrimination, and claim intersectionality, in which the law does not adequately redress intersectional discrimination that occurs in the labor market. Demographic intersectionality can be thought of as a type of inequality in litigation, while claim intersectionality can be thought of as a mismatch between discrimination as conceptualized by law and discrimination as experienced in the labor market.

Demographic Intersectionality

Demographic intersectionality disadvantages occur when discrimination and/or stereotyping targets plaintiffs who occupy the intersection of two or more demographic categories. For these plaintiffs, overlapping axes of disadvantage may add up to more than the sum of their parts. Articulating the mechanisms through which intersectional stereotypes operate, social psychologists find that people sometimes merge information from multiple categories to create subcategories with attendant stereotypes (Reference BodenhausenBodenhausen 2010; Reference Roccas and BrewerRoccas & Brewer 2002), and that information about characteristics or roles can take on new meanings when nested within other categorical formations (Reference Hutter and CrispHutter & Crisp 2005; Reference Kunda and ThagardKunda & Thagard 1996; Reference Richards and HewstoneRichards & Hewstone 2001; Reference RidgewayRidgeway 1997; Reference StangorStangor et al. 1992). By demonstrating that status characteristics are not always perceived independently, this experimental research suggests that intersectional stereotypes are likely to emerge and to influence social perception and judgment.

Indeed, recent empirical research on hiring and discrimination provides evidence that employers hold discrete stereotypes for various intersectional categories. Employers may stereotype inner-city blacks (but not necessarily other blacks or white inner-city residents) as lazy and dangerous (Reference Kirschenman, Neckerman, Jencks and PetersonKirschenman & Neckerman 1991; Reference Moss and TillyMoss & Tilly 2001). Employers also hold different stereotypes about black men and black women. They sometimes stereotype black women negatively as desperate single mothers (Reference KennellyKennelly 1999) or positively as responsible “matriarchs” (Reference ShihShih 2002: 111). Black men, on the other hand, are stereotyped as “unmanageable workers [who are] more likely to resist authority” (Reference ShihShih 2002: 102). Using census data, Reference KaufmanKaufman (2002) concludes that employers often have preconceived notions about which race and sex combinations are right for a job and tend to select job applicants who match these stereotypes. Interview-based research and audit studies confirm that employers prefer to hire white men for low-skilled jobs (Reference Moss and TillyMoss & Tilly 2001; Reference TurnerTurner et al. 1991).

Judges, juries, and lawyers are subject to the same institutionalized stereotypes as are employers. If they introduce these stereotypes into legal decisionmaking, the types of stereotypes discussed in the literature on labor-market discrimination may also affect court outcomes, with courts replicating the discriminatory practices that operate in the labor market.

Claim Intersectionality

Claim intersectionality is present when plaintiffs allege discrimination on the basis of two or more ascriptive characteristics. Critical race theorists have argued that since antidiscrimination law organizes demographic traits into formal, one-dimensional categories—race, sex, national origin, and so forth—legal doctrine often fails to capture the types of discrimination suffered by intersectional subjects (Reference AustinAustin 1989; Reference CaldwellCaldwell 1991; Reference CrenshawCrenshaw 1989; Reference HarrisHarris 1990; Reference RobertsRoberts 1991). So, for example, sex discrimination is conceptualized in antidiscrimination case law as a problem affecting all women equally and in the same ways (with white women as the prototypic case), while race discrimination is understood as affecting all blacks (prototypically male) in the same ways (Reference CrenshawCrenshaw 1989; Reference HarrisHarris 1990). Intersectionality theorists argue that this one-dimensional, categorical approach to understanding discrimination prevents civil rights law from adequately protecting members of groups that experience more than one axis of prejudice.

For example, an employer might be willing to hire black men and white women as retail salespeople but unwilling to hire black women because he thinks that customers will stereotype them in disparaging ways that will harm his business (Reference SmithSmith 1991: 28). Or, as another example, an employer might fire a black female employee because the employer is discomfited by her Afrocentric feminine attire or hairstyle (Reference CaldwellCaldwell 1991). Their employees might make what we call intersectional claims: allegations that they were discriminated against due to more than one ascriptive characteristic. But since these types of discrimination would not affect minority men or white women, under some interpretations of EEO law, the employer could parry a claim of race discrimination by pointing to the hiring of men belonging to the plaintiff's racial group and deflect a claim of sex discrimination by pointing to his hiring of white women (Reference CrenshawCrenshaw 1989; Reference HarrisHarris 1990; Reference SmithSmith 1991).

Thus, plaintiffs who make intersectional discrimination claims may be less likely to win their cases not only because they are members of particularly derogated subgroups, but also because, given the categorical nature of discrimination law, intersectional claims are particularly hard to establish. While demographic intersectionality can produce unequal outcomes in all arenas of social life, claim intersectionality is a mechanism of disadvantage that is particular to civil rights litigation.

Lack of Empirical Research on Intersectionality and Litigation Outcomes

Although a substantial and rich literature describes the nature of intersectionality and demonstrates how intersectionality has penalized plaintiffs in particular cases, there has been no systematic effort to determine the extent to which intersectionality penalizes plaintiffs in litigation generally. Regarding demographic intersectionality, some studies have compared litigation outcomes across gender or racial groups, but most studies have examined race and gender disparities separately, comparing all men to all women or comparing all racial minorities to all whites (Reference BabcockBabcock 1993; Reference NelsonNelson 1994; Reference ResnikResnik 1991; Reference SchafranSchafran 1998; Reference SelbySelby 1999). Since they ignore the intersection of race and gender, these studies may elide important differences among subgroups, and we remain in the dark about whether race disparities are constant across gender, and vice versa. One exception to this trend is Reference OppenheimerOppenheimer (2003), who examined a sample of 334 employment-discrimination and wrongful-discharge cases tried in the California state courts and found that black women had low win rates in cases alleging sex discrimination or race discrimination. While these results suggest that plaintiffs with intersecting disadvantaged statuses may fare worse in the California state courts, the study did not test whether the differences were statistically significant or control for other characteristics of the cases. To fully examine demographic intersectionality, we must statistically compare outcomes for white men, white women, nonwhite men, and nonwhite women.

Claim intersectionality has attracted more scholarly attention in the form of several qualitative analyses of employment discrimination opinions by critical race theorists (Reference CrenshawCrenshaw 1989; Reference CunninghamCunningham 1998; Reference Scales-TrentScales-Trent 1989; Reference SmithSmith 1991). This work highlights a series of judicial opinions with widely varying treatment of intersectional claims. In some cases, judges have not recognized intersectional claims as being legally cognizable and have dismissed them at the outset. Reference CrenshawCrenshaw's (1989) foundational article on intersectionality, for example, centers on the case of DeGraffenreid v. General Motors (1976), in which the plaintiffs established that GM had not hired any black female workers before 1964 and that all the black women hired after 1970 had lost their jobs in a later seniority-based layoff. The court ruled that black women were not a protected class under Title VII. Since the company had hired white women, no sex discrimination had occurred. Since the company had hired black men, no race discrimination had occurred either. A similar fate befell black female plaintiffs in Jewel C. Rich v. Martin Marietta (1975), Ella Logan v. St. Luke's Hospital Center (1977), and Mary M. Love v. Alamance County Board of Education (1984), where the judges also considered race and sex claims separately and disregarded statistical evidence of discrimination against black women as legally irrelevant.Footnote ²

However, in other opinions, the courts have been sympathetic to intersectional claims. In Jefferies v. Harris County Community Action Association (1980), the court ruled that “discrimination against black females can exist even in the absence of discrimination against black men or white women” (quoted in Reference Scales-TrentScales-Trent 1989: 17). Similarly, in reversing the district court's grant of summary judgment in Lam v. University of Hawai'i (1994), the U.S. Court of Appeals for the Ninth Circuit criticized the district court for imagining that racism and sexism can be evaluated separately. Finally, some opinions are mixed. In Judge v. Marsh (1986), the court stated that it would consider intersections of two characteristics but not three or more, out of the concern that considering too many intersections would turn Title VII into a “many-headed Hydra” and make it impossible to make any employment decisions “without incurring a volley of discrimination charges” (quoted in Reference Scales-TrentScales-Trent 1989: 37).

These cases reveal the courts' varying responses to intersectional claims but leave broader patterns obscure: How many plaintiffs are bringing intersectional claims? On the whole, are plaintiffs who bring intersectional claims less likely to win their cases? And if so, can any other characteristics of the cases explain this disadvantage? These questions can best be addressed through quantitative empirical research. However, to date, there has been very little empirical research on the effects of claim intersectionality on litigation outcomes. To our knowledge only one preliminary empirical analysis speaks directly to this question. Reference KotkinKotkin (2009) examined 26 employment discrimination summary judgment adjudications in the federal district courts for the Southern and Eastern Districts of New York in which plaintiffs presented multiple-basis claims. These plaintiffs lost the defense motion for summary judgment 96 percent of the time, which is higher than plaintiff loss rates reported by other studies of summary judgment outcomes, providing suggestive evidence that plaintiffs alleging multiple bases of discrimination fare poorly. However, the study did not include a comparison group of single-basis claims, thus leaving it unclear whether plaintiffs alleging single bases of discrimination fared better, and if so, whether the differences were statistically significant. Documenting claim-intersectionality disadvantages requires systematically testing whether intersectional claims fare worse than other claims do.

Thus, no existing research systematically compares intersectional and nonintersectional cases and tests whether intersectional cases fare significantly worse. Our analysis addresses the lacuna of empirical data and findings on this issue by investigating intersectionality disadvantages using a probability sample of federal civil rights opinions.

The Politics of Methods in Intersectionality Scholarship

The lack of quantitative empirical work on intersectionality is due in part to methodological conflicts within critical race scholarship. The first point of controversy centers on how to use categories in research on intersectionality. Scholars who believe that the main contribution of intersectionality theory is the documentation of the detrimental effects of categorization are loath to use their own work to divide people into categories. Additionally, some scholars suggest that intersectionality cannot be captured through an interaction effect because the social construction of categories is contingent and fluid (Reference CollinsCollins 1990; Reference KingKing 1988; Reference West and FenstermakerWest & Fenstermaker 1995). However, other researchers defend the importance of using categories to document inequalities (Reference McCallMcCall 2001, Reference McCall2005).Footnote ³

The second point of disagreement centers on research methods. Critical race scholars who study intersectionality use almost exclusively qualitative and interpretive methods (Reference AbramsAbrams 1994; Reference AustinAustin 1989; Reference CaldwellCaldwell 1991; Reference CrenshawCrenshaw 1989, Reference Crenshaw1991, Reference Crenshaw1992; Reference DelgadoDelgado 1995; Reference RobertsRoberts 1991; Reference SmithSmith 1991; Reference WilliamsWilliams 1991). Many critical race scholars criticize quantitative research as overly simplistic and positivist (Reference DavisDavis 2008; Reference McCallMcCall 2005). However, a smaller group of researchers criticizes the exclusive use of qualitative methods for providing inadequate documentation of inequalities (Reference NashNash 2008), argues for the use of statistics to document differing outcomes among groups (Reference Baldus, Woodworth and PulaskiBaldus et al. 1990; Reference McCallMcCall 2005), and calls for greater dialogue between critical race theory and empirical research on law and society (Reference Gomez and SaratGomez 2004).

We argue that these conflicts can be resolved by recognizing that the best method of analysis depends on the nature of the research question and the dependent variable. While qualitative research is most appropriate for in-depth studies of experiences and identities (Reference HarveyHarvey 2005; Reference Hondagneu-SoteloHondagneu-Sotelo 2001), quantitative research may be best suited to documenting the aggregate patterns that constitute between-group inequalities (Reference McCallMcCall 2001, Reference McCall2005; Reference Yuval-DavisYuval-Davis 2006). While racial, sex, or other categories certainly do not richly describe people's experiences and identities, differing outcomes across these categories are important indicators of structural inequality and social stratification. Additionally, as the social categories on which discrimination is often based and through which legal claims must be pursued, these categories have real effects.

While there is an extensive research literature exploring intersectional experiences and identities, researchers have rarely sought to document the effects of intersectionality on inequality (Reference Browne and MisraBrowne & Misra 2003). Attributing this neglect to methodological preferences, McCall argues that suspicion of statistics has “restrict[ed] the scope of knowledge that can be produced on intersectionality” (Reference McCallMcCall 2005: 1772). This neglect is so extreme that the hypothesis that Reference CrenshawCrenshaw (1989) introduces in her foundational article on intersectionality—that intersectional plaintiffs fare worse in discrimination lawsuits—has not been systematically tested. Our study is designed to test this hypothesis.

Methods

Coding

We coded each opinion for court (which circuit or which district), judges, plaintiff characteristics, defendant characteristics, statutory claims involved in the case, challenged actions, legal theories on which the claims were based, which party prevailed (and the extent to which they prevailed), and a variety of other variables. Table 1 presents descriptive statistics for the variables used in the analysis. Given the complexity of our coding scheme, we took numerous measures to ensure intercoder reliability. We developed and refined the coding scheme through an iterative procedure involving trial coding of opinions by five researchers over a period of about one year and then ran a series of reliability checks to ensure that there were no systematic differences among the coders.Footnote ⁹

Table 1. Descriptive Statistics

¹ More than one ascriptive characteristic(race, sex, age, or national origin)

² E.g., race and retaliation

³ E.g., hiring, termination, compensation

Independent Variables

Claim Intersectionality

Claim-intersectionality theory suggests that plaintiffs are at a disadvantage when they allege intersectional discrimination, regardless of their demographic characteristics. We observe claim intersectionality when a plaintiff alleges discrimination based on more than one of the following ascriptive characteristics: race, sex, age, or national origin.Footnote ¹¹ Table 2 presents descriptive statistics for the specific intersections in our sample. The intersection of race and sex accounted for one-third of intersectional claims, and race and age, race and national origin, sex and age, and three- and four-way intersections also appeared relatively frequently.

Table 2. Number of Sampled Cases with Various Bases of Discrimination

Instead of using an interaction effect, as we do to measure demographic intersectionality (see below), we measure claim intersectionality using three mutually exclusive dummy variables. We distinguish intersectional claims from cases where plaintiffs allege only one basis of discrimination and also from cases where plaintiffs allege multiple nonintersectional bases of discrimination (that is, more than one basis of discrimination of which only one or none are ascriptive characteristics). For instance, a case with allegations of race and sex discrimination is coded as intersectional, while a case with allegations of race discrimination and retaliation is not. By this definition, 18 percent of cases in our sample make intersectional claims (see Table 1).Footnote ¹²

Demographic Intersectionality

Demographic intersectionality theory yields the hypothesis that various axes of disadvantage (race, sex, age, sexuality, and so on) do not operate independently in the courts. All our data on plaintiffs' demographics are coded from judicial opinions, which rarely mention demographic characteristics other than race or sex. Thus, although demographic intersectionality theory could apply to the intersection of any demographic characteristics, we can only test it for the intersection of race and sex. Our data are further limited by the fact that the plaintiffs' race and sex are not always mentioned in the judicial opinions. In 40 opinions, we could not determine the plaintiffs' sex. Additionally, 57 cases involved a mixed-sex group of plaintiffs or an organizational plaintiff. We combined these two types of cases into a “missing sex” group that makes up 9.5 percent of the sample (see Table 1).Footnote ¹³ Our data on the plaintiffs' races are sparser. Thirty-two cases involved a racially mixed group of plaintiffs. In an additional 579 opinions, the plaintiffs' races were not mentioned. We combine racially mixed groups of plaintiffs and plaintiffs whose race was not mentioned into a “missing race” category that comprises 60 percent of our sample (see Table 1). We originally coded plaintiffs' races into six categories,Footnote ¹⁴ but our sample is too small to examine each racial category separately. Therefore, we compare the 60 plaintiffs we can identify as white to the 353 plaintiffs we can identify as nonwhite, a category that includes blacks, Hispanics, Asians and Pacific Islanders, Native Americans, and plaintiffs of other nonwhite races (see Table 1). Most of the plaintiffs we can identify as nonwhite are black (84 percent). To see whether patterns differed between blacks and other minorities, we also ran our models with three racial categories and found no important differences.Footnote ¹⁵

As might be expected, most opinions that mentioned the plaintiffs' races resulted from cases alleging race or national origin discrimination. Of the 353 plaintiffs we can identify as nonwhite, 320 alleged discrimination on the basis of race or national origin. We were concerned that the 33 opinions in which judges mentioned the plaintiff's race even though the plaintiff was not alleging racial or national origin discrimination might be unusual cases or might differ in some way from race and national origin cases. Therefore, we reran all analyses with interaction effects between the race variables and whether the case included race or national origin claims; our substantive results were unchanged.Footnote ¹⁶

Thus, our data on plaintiffs' demographic characteristics are more limited than our data on intersectional claims: we only have measures of plaintiffs' race and sex, and even that information is missing for a large proportion of plaintiffs. However, since we take care to distinguish between plaintiffs with missing and nonmissing data, we can draw some conclusions about the effects of race and sex on litigation outcomes. To test the hypothesis that race- and sex-based disadvantages do not operate independently in the courts, we ran regressions with variables for plaintiffs' race and sex and an interaction effect between race and sex. This modeling strategy distinguishes between white male, white female, nonwhite male, and nonwhite female plaintiffs, and plaintiffs with unknown race or sex.

Overlap between Claim and Demographic Intersectionality

Claim intersectionality and demographic intersectionality overlap but are not perfectly correlated. Nonwhite female plaintiffs are more likely to make intersectional claims than are other plaintiffs. However, not all nonwhite female plaintiffs make intersectional claims (for instance, a black female plaintiff might sue for race discrimination alone), and not all intersectional claims are brought by nonwhite women (for instance, a claim of race and age discrimination might be brought by a black man, or a white man might claim reverse race and sex discrimination). Table 3 shows the extent to which the categories overlap. We hypothesize that demographic and claim intersectionality are two separate processes of disadvantage: intersectional claims will be less likely to succeed, regardless of plaintiffs' demographic characteristics, and race and sex disadvantages will intersect in the courts regardless of whether plaintiffs make intersectional claims.

Table 3. Overlap between Nonwhite Female Plaintiffs and Intersectional Claims

Mediating Variables

While we theorize that demographic and claim intersectionality can operate separately, we also test whether either one mediates the other by testing their effects on case outcome separately and then including both in the same model. Additionally, we explore several other potential mediating variables.

Reference SmithSmith (1991) suggests that plaintiffs may be especially unlikely to prevail when they file an intersectional claim under Title VII and a race claim under Section 1981, which applies only to race discrimination. When plaintiffs fail to prove race discrimination under Section 1981, judges may throw out the race portion of their Title VII claims.Footnote ¹⁷ Smith argues that by mistakenly assuming that intersectional race and sex claims under Title VII can be considered separately, these judges prevent intersectional claims from getting a fair hearing. To test Smith's hypothesis that the combination of Section 1981 and Title VII is disadvantageous for plaintiffs making intersectional claims, we created a variable indicating whether the case includes claims brought under both statutes and an interaction effect between this variable and claim intersectionality.

Another potential mediating variable is judicial deference to institutionalized organizational structures. Judicial deference occurs when judges take the mere presence of organizational structures as evidence of an organization's compliance with civil rights law irrespective of whether the structures actually protect employees (Reference EdelmanEdelman et al. 1999; Reference EdelmanEdelman et al. 2011). For example, despite the fact that civil rights law neither mandates the creation of organizational grievance procedures nor specifies that these structures constitute evidence of nondiscrimination, and despite the fact that many organizational grievance procedures are ineffective at reducing discrimination (Reference EdelmanEdelman 1992; Reference EdelmanEdelman et al. 1993), courts have become increasingly likely over time to accept the presence of formal grievance procedures as evidence of nondiscrimination without evaluating their effectiveness (Reference Bisom-RappBisom-Rapp 1999, Reference Bisom-Rapp2001; Reference EdelmanEdelman et al. 1999). Given research that suggests that judges are often intuitive decision makers and that intuitions are often flawed (Reference Guthrie, Rachlinski and WistrichGuthrie et al. 2007), institutionalized organizational structures may provide a heuristic mechanism through which judges are more likely to assume fair governance when they are more skeptical of the plaintiffs or the claims. Reference KotkinKotkin (2009) notes that some federal judges treat plaintiffs who make intersectional claims like “the child who cried wolf” (Reference KotkinKotkin 2009: 1458). If judges are skeptical about the existence of intersectional discrimination, they may be predisposed to seek out signs that employers charged with intersectional discrimination are rational and nondiscriminatory. Increased likelihood of judicial deference—using structures as symbolic indicia of fair treatment instead of considering evidence as to whether they reduce discrimination—may indicate a subtle shift in judges' symbolic construction of plaintiffs and defendants. Thus, if deference mediates the effect of claim intersectionality, this would indicate that judges' subjective constructions of intersectional claims account for some of intersectional plaintiffs' disadvantage.

Another set of potential mediating variables measures plaintiffs' resources. While we cannot measure all aspects of plaintiffs' resources, we coded for three variables that shed some light on resources: unionization, involvement of government or public-interest organizations, and occupational prestige. Our first measure, unionization, is a dummy variable equal to one if the judicial opinion mentions that any of the plaintiffs were union members. Second, since government or public-interest organizations can provide a substantial legal advantage to plaintiffs, we created a dummy variable equal to one if (a) the plaintiff is a government agency or public-interest organization, (b) the plaintiff is represented by a public-interest organization, or (c) a government or public-interest organization filed an amicus brief for the plaintiff. Third, we coded for plaintiffs' occupational prestige using 1980 census occupational categories and the 1989 General Social Survey (GSS) occupational prestige scale (Reference Nakao and TreasNakao & Treas 1992).Footnote ¹⁸ While no single measure can capture all aspects of occupational status, we expect occupational prestige rankings to be correlated with economic and cultural resources that can help plaintiffs succeed in court. If plaintiffs with particular race and sex characteristics or plaintiffs who bring intersectional claims tend to have fewer resources, then controlling for unionization, organizational involvement, and occupational prestige should decrease the size of the intersectionality coefficients.

Crenshaw and colleagues argue that doctrinal barriers and evidentiary hurdles diminish the success rates for intersectional claims. Doctrinal barriers stem from the categorical nature of discrimination law, and evidentiary hurdles from the difficulty of proving complex claims. For example, proving that nonwhite women were less likely to be promoted requires a large enough sample of nonwhite women, white women, nonwhite men, and white men in the workplace (Reference KotkinKotkin 2009), making it more difficult to document intersectional discrimination than it is to document discrimination based on only one characteristic. While we cannot measure doctrinal barriers and evidentiary hurdles directly, we examine whether the opinion results from a motion to dismiss or a defendant's motion for summary judgment. In a motion to dismiss, the employer argues that there is no need to consider the facts of the case, since the plaintiff's claims are inconsistent with legal doctrine. In a motion for summary judgment, the defendant argues that the plaintiff's claims can be rejected because the plaintiff has failed to adduce evidence in discovery from which a reasonable jury could find in their favor at trial. If many intersectional cases are losing on summary judgment instead of on considerations of the facts of the case, this could indicate that procedural barriers or evidentiary hurdles are preventing intersectional claims from succeeding.

Control Variables

We control for several variables predicted to be correlated with employee victory: whether the case is in the circuit or district court, whether the opinion is legally published,Footnote ¹⁹ the length of the opinion in pages, the number of employer actions the employee is challenging, and the passage of time.Footnote ²⁰ To account for change over time, we control for whether the case was decided after 1986, when the Celotex trilogy of decisionsFootnote ²¹ made it easier to obtain summary judgment and the Meritor Savings Bank v. Vinson (1986) decision stated that employers would be more likely to prevail in some types of EEO cases if they had established certain employment structures, such as antiharassment policies and grievance procedures (Reference EdelmanEdelman et al. 2011). We also ran all models with dummy variables for each year to completely account for any time patterns; our results were substantively unchanged.

Results

The Increasing Prevalence of Intersectional Claims

Intersectional claims have increased dramatically over time. In the 1970s and 1980s, less than 10 percent of EEO opinions dealt with intersectional claims. As shown in Figure 1, the proportion began rising around 1990, and by the second half of the decade, more than a quarter of EEO opinions involved intersectional claims.Footnote ²³ The proportion dropped somewhat in 1999, but since this is the last year we observe, we cannot discern whether this was a change in the trend or a temporary aberration. Since the total number of EEO opinions rose dramatically during this time period, the increasing proportion of intersectional claims indicates an even more dramatic increase in real numbers: we estimate that there were fewer than 100 intersectional cases per year in the district and circuit courts in the 1970s and 1980s and over 1,000 per year by the second half of the 1990s. This increasing prevalence highlights the importance of learning how these claims are faring.

Figure 1. Percent of EEO cases with intersectional claims.

Both Demographic and Claim Intersectionality Disadvantage Plaintiffs

Bivariate relationships between both claim and demographic intersectionality and case outcomes yield strong support for intersectionality theory. First, plaintiffs making intersectional claims are less than half as likely to win their cases as are other plaintiffs (15 percent compared to 31 percent; see Table 4). Second, race and sex disadvantages do not operate independently. White male plaintiffs were more likely to lose their cases than white women were (61 percent as compared to 55 percent; see Table 4). This female advantage, however, does not apply to black women, who are slightly more likely than black men to lose their cases (71 percent as compared to 69 percent; see Table 4).Footnote ²⁴

Table 4. Case Outcome by Claim and Demographic Intersectionality

The bivariate relationships between both types of intersectionality and employee victory provide suggestive evidence of an intersectionality penalty. Next, we conducted generalized ordered logistic regressions to control for other features of cases that might account for the relationships (see Table 5). In each model we present, the first column of coefficients denotes each variable's effects on the odds that the plaintiff will achieve at least a partial victory (a mixed outcome or a complete victory), and the second column of coefficients focuses on the odds of a complete victory.

Table 5. Generalized Ordered Logistic Regressions of Employee Victory on Demographic and Claim Intersectionality

Standard errors in parentheses.

All models control for the number of challenged actions, circuit vs. district court, legal publication, length in pages, and post-1986.

^*** p < 0.001, ^**p < 0.01, ^*p < 0.05, ^tp < 0.1.

Model 1 (Table 5) shows the effects of claim intersectionality on the likelihood of employee victory. Even when controlling for multiple aspects of the case, compared to plaintiffs who allege a single basis of discrimination, plaintiffs making intersectional claims have only about half the odds of attaining at least a partial victory and approximately one-third the odds of a complete victory.Footnote ²⁵ All else equal, we predict that plaintiffs alleging only one basis of discrimination will win their cases 28 percent of the time, whereas plaintiffs bringing otherwise identical cases that allege intersectional bases of discrimination will win only 13 percent of the time.Footnote ²⁶ This finding provides strong evidence for the hypothesis that EEO law disadvantages plaintiffs who allege intersectional discrimination.

One interpretation of this finding might be that cases alleging multiple types of discrimination were intrinsically weaker, with desperate plaintiffs adding bases of discrimination and hoping that one would be successful. At least one federal judge adopts this view, suggesting that plaintiffs who allege multiple bases of discrimination are “throwing spaghetti at the wall to see what sticks” (district court judge Ruben Castillo, quoted in Reference KotkinKotkin 2009: 1442). We test for this possibility in three ways. First, we control for multiple challenged actions because desperate plaintiffs might be just as likely to challenge multiple employer actions (e.g., compensation and promotion) as to allege multiple bases of discrimination. We do find a significant negative effect of the number of challenged actions on complete plaintiff victory, but the effect's magnitude is much smaller than is the claim intersectionality effect. Second, if desperate plaintiffs were likely to add both challenged actions and bases of discrimination to their cases, then controlling for the number of challenged actions would decrease the size of the coefficient for intersectional claims. In fact, including the number of challenged actions in the model has no such effect. Third, if alleging multiple bases of discrimination were an indicator of intrinsically weak cases, it should not matter whether or not the additional alleged bases of discrimination are based on ascriptive characteristics. Our results show that whether the bases of discrimination are ascriptive or not matters: we find a large significant negative effect for intersectional claims (those alleging discrimination based on multiple ascriptive characteristics), but only a small, marginally significant effect for cases alleging multiple nonintersectional bases of discrimination (that is, cases in which only one or none of the alleged bases of discrimination is an ascriptive characteristic). These findings suggest that intersectional claims are not the result of plaintiffs' frivolously adding additional claims.Footnote ²⁷

Model 2 in Table 5 shows the effects of plaintiffs' demographics on the likelihood of employee victory without considering demographic intersectionality. Besides the control variables, Model 2 includes only the main effects for race and sex, which are measured by dummy variables for nonwhite plaintiffs and plaintiffs with missing race data (white plaintiffs are the omitted category) and variables for female plaintiffs and plaintiffs with missing sex data (male plaintiffs are the omitted category). Model 2 shows that all else equal, nonwhite plaintiffs have less than one-third of white plaintiffs' odds of achieving complete victories. Female plaintiffs are slightly more likely than male plaintiffs to achieve at least partial victories, but this coefficient is significant only at the.1 level.Footnote ²⁸ If we stopped at Model 2 (hence ignoring demographic intersectionality), as do previous studies, we would likely conclude that there are no important differences between men's and women's outcomes in EEO litigation.

Model 3 improves on Model 2 (and on previous research) because it accounts for demographic intersectionality by including a variable set to one if the plaintiff is a nonwhite woman. The interaction effect has a negative and statistically significant effect on plaintiffs' odds of at least partial victory and on plaintiffs' odds of complete victory. When we include it in the model, the main effect for sex becomes a significant predictor of at least partial plaintiff victory.

Based on Model 3, and holding all control variables constant at their means or modes, white women have the highest predicted probability of a full victory (38 percent), followed by white men (31 percent), nonwhite men (15 percent), and nonwhite women (11 percent).Footnote ²⁹ This intersectional relationship between race and sex can be understood in two ways. First, there are larger race effects for women than for men: nonwhite women fall further behind white women than nonwhite men fall behind white men. Second, there are different gender effects for whites and nonwhites: white women get ahead of white men, while nonwhite women fare similarly to nonwhite men.Footnote ³⁰ Our findings, then, suggest that studies that fail to account for demographic intersectionality miss the fact that sex and race disadvantages do not operate independently in the courts.

Whereas Models 1 and 3 in Table 5 consider demographic and claim intersectionality separately, Model 4 includes both, thus allowing us to test whether either intersectionality effect is an artifact of omitted variable bias. Given that plaintiffs making intersectional claims are disproportionately likely to be nonwhite women (see Table 3), the effect of claim intersectionality observed in Model 1 might actually reflect nonwhite women's disadvantage. Alternatively, the negative coefficient for nonwhite women in Model 3 might be explained by the fact that nonwhite women are disproportionately likely to make intersectional claims (see Table 3). Model 5 shows, however, that each type of intersectionality has an independent effect on plaintiffs' likelihood of winning. The claim intersectionality coefficient remains statistically significant and decreases only slightly. Regarding demographic intersectionality, the interaction effect between plaintiffs' race and sex is only significant at the p < 0.1 level in Model 4, but its magnitude is virtually unchanged. The fact that the coefficients for both measures of intersectionality remain large and at least marginally statistically significant when included in the same model suggests that demographic and claim intersectionality represent two distinct pathways of disadvantage for plaintiffs. Demographic and claim intersectionality are each associated with dramatically reduced odds of plaintiff victory.

Figures 2 and 3 show the magnitude of the effects of each type of intersectionality, net of the other, with all other variables held constant at their means or modes. Figure 2 shows the effects of claim intersectionality net of demographic intersectionality by giving the predicted probabilities of complete victories for nonwhite female plaintiffs who do and do not assert intersectional claims. Even among nonwhite female plaintiffs, intersectional claims are predicted to prevail only half as often as single claims. Figure 3 shows the effects of demographic intersectionality net of claim intersectionality by giving the predicted probabilities of complete victories in single-claim cases for plaintiffs at each intersection of race and sex. Even when controlling for their increased likelihood of making intersectional claims, nonwhite female plaintiffs still have the lowest predicted probability of winning their cases.

Figure 2. Effects of claim intersectionality, net of demographic intersectionality: predicted probabilities of complete employee victory for nonwhite women, holding controls constant at their means or modes.

Figure 3. Effects of demographic intersectionality, net of claim intersectionality: predicted probabilities of complete employee victory for single-claim cases, holding controls constant at their means or modes.

Mediating Variables: Mechanisms for the Intersectionality Penalty

To explore how intersectionality disadvantages plaintiffs, we consider several variables that might mediate the effects of demographic and claim intersectionality on plaintiff victory. In Table 6, we sequentially add mediating variables to Model 4 to see whether they decrease the size and significance of the claim intersectionality coefficient or the race/sex interaction effect. Model 4 is repeated in Table 6 for comparison of the coefficients. Mediating variables that decrease the effects of the intersectionality variables tell us something about how the intersectionality penalty works.

Table 6. Mechanisms for Intersectionality Effects

All models control for the number of challenged actions, circuit vs. district court, legal publication, length in pages, and post-1986.

¹ Government or public-interest involvement, occupational prestige, unionization, summary judgment, and motion to dismiss.

Standard errors in parentheses. ^***p < 0.001, ^**p < 0.01, ^*p < 0.05, ^tp < 0.1.

First, to test Reference SmithSmith's (1991) hypothesis that intersectional claims fare especially poorly when plaintiffs bring claims under both Section 1981 and Title VII, Model 5 adds a dummy variable set to one when the plaintiff brings claims under both statutes and an interaction effect between this variable and intersectional claims. Consistent with Smith's argument, there is a large negative interaction effect between intersectional claims and the Title VII/Section 1981 combination on the odds of employee victory. The coefficient is large, but is only significant at the.1 level; it becomes larger and statistically significant with the inclusion of more control variables in Models 6 and 7. The Title VII/Section 1981 combination is actually advantageous when not combined with intersectional claims; plaintiffs alleging only one type of discrimination under both Title VII and Section 1981 have a higher predicted probability of a full victory (37 percent) than plaintiffs who do not combine the two statutes (30 percent). Intersectional claims have lower rates of plaintiff victory even without this particular combination of statutes (19 percent), and plaintiffs who make intersectional claims under both Title VII and Section 1981 have the lowest win rate of all (10 percent).Footnote ³¹ Thus, Smith is correct that plaintiffs who make intersectional claims under the two statutes are especially unlikely to win their cases.

Next, we explore whether judicial deference to organizational structures mediates the effects of intersectionality. As mentioned earlier, legal endogeneity theory (Reference EdelmanEdelman et al. 1999; Reference EdelmanEdelman et al. 2011) suggests that judges may be especially likely to view institutionalized structures such as grievance procedures or formal notice policies as evidence of nondiscriminatory treatment when they are skeptical of the plaintiffs or their claims, making judicial deference a potential mechanism for the intersectionality penalty. Model 6 in Table 6 explores this possibility by including a dichotomous variable indicating whether any structures were deferred to.Footnote ³² When we include judicial deference in the model, the coefficients for plaintiffs' demographics remain virtually unchanged; thus, judicial deference does not mediate the effects of demographic intersectionality. Likewise, the interaction effect between claim intersectionality and the Title VII/Section 1981 combination remains large and statistically significant. However, the main effect for claim intersectionality is reduced by half and loses statistical significance. Thus, in cases with intersectional claims that do not involve both Title VII and Section 1981, judges are more likely to defer to employers' structures, and this increased likelihood of deference is an important mediator of the disadvantage faced by plaintiffs bringing intersectional claims.

In Model 7, we add three indicators of the resources available to plaintiffs (unionization, involvement of government or public-interest organizations, and plaintiffs' occupational prestige) and two variables related to the procedural posture that may indicate the presence of doctrinal barriers or evidentiary hurdles (motion to dismiss and summary judgment). When these variables are included, the coefficients for the claim intersectionality main effect decrease further; however, they were already small and nonsignificant before the addition of these controls. Including the new variables has little effect on any of the coefficients for plaintiffs' demographics and actually increases the coefficient for the claim intersectionality/Title VII/Section 1981 interaction effect. Thus, our rough measures of resources, doctrinal barriers, and evidentiary hurdles explain little to none of the claim and demographic intersectionality penalties.Footnote ³³

Discussion

Our analysis provides the first systematic empirical test of intersectionality theory by examining the effects of both demographic intersectionality and claim intersectionality on plaintiff win rates in employment discrimination cases. We find strong support for the ideas that race and sex disadvantages do not operate independently in the courts (demographic intersectionality) and that antidiscrimination law provides less protection in cases that involve intersecting bases of discrimination (claim intersectionality).

Our results suggest, moreover, that these two types of intersectionality represent two distinct processes of disadvantage. Although nonwhite women are more likely to bring intersectional claims, this does not explain all of the disadvantage they face in court. Likewise, claim intersectionality harms plaintiffs' chances of winning, regardless of their demographic characteristics. EEO law itself seems to disadvantage intersectional plaintiffs, above and beyond any discrimination they may face in the courtroom on the basis of their race or sex.

Previous research suggested three main reasons why intersectional claims might disadvantage plaintiffs: (1) the categorical nature of discrimination law creates doctrinal barriers to intersectional claims, (2) there are evidentiary hurdles to demonstrating intersectional discrimination, and (3) judicial skepticism about intersectional claims may make intersectional plaintiffs less likely to win their cases. Our findings suggest that the Title VII/Section 1981 combination does pose a doctrinal barrier that mediates the effects of claim intersectionality and thus supports Reference SmithSmith's (1991) argument that judges tend to believe that intersectional claims can be neatly separated and that a ruling against the plaintiffs' race allegations under Section 1981 prohibits consideration of race intersections under Title VII. Our findings also suggest that judicial deference may be an important mechanism through which claim intersectionality penalizes plaintiffs. Controlling for judicial deference explains most of the claim intersectionality penalty for cases that are not brought under both Title VII and Section 1981. The fact that judicial deference is especially likely in cases involving intersectional claims elaborates Reference EdelmanEdelman et al.'s (1999, Reference Edelman2011) suggestion that judges treat institutionalized organizational structures as symbolic indicia of employers' rationality and compliance. In sum, although our data do not permit a direct test of the mechanisms through which intersectionality operates, we find some support for the ideas that intersectional claims are held back by a combination of doctrinal barriers and judicial interpretations.

Our findings have important implications for several theoretical debates in the intersectionality literature. One point of disagreement among scholars is whether intersectionality applies only to members of traditionally disadvantaged groups or whether all identities are intersectional (Reference Browne and MisraBrowne & Misra 2003; Reference DavisDavis 2008; Reference NashNash 2008). Some researchers subscribe to a “multiple jeopardy” approach that assumes that women of color are more disadvantaged than other groups in all contexts (Reference KingKing 1988; Reference Ransford, Jeffries and RansfordRansford 1980). However, recent critiques of the intersectionality literature from within argue that when limited to the case of black women, intersectionality is insufficiently developed as a general theory (Reference NashNash 2008), and view intersectionality as a broader theory that can apply “to any grouping of people, advantaged as well as disadvantaged” (Reference Yuval-DavisYuval-Davis 2006, 201; see also Reference Chang and CulpChang & Culp 2002; Reference KwanKwan 1996; Reference ZackZack 2005). A related debate focuses on whether intersectional disadvantages are “ubiquitous or contingent” (Reference Browne and MisraBrowne & Misra 2003: 492). Some scholars argue that intersectionality affects outcomes and experiences in every social setting (Reference CollinsCollins 1990; Reference WeberWeber 2001), while others suggest that its effects are contingent, with single categories sometimes dominating (Reference McCallMcCall 2005; Reference Yuval-DavisYuval-Davis 2006). Our findings demonstrate that intersectionality is context dependent. Whereas intersectionality theory generally presumes that white men tend to fare the best and nonwhite women are the most disadvantaged, our findings suggest a somewhat different pattern. In our sample, white female plaintiffs had the highest chances of winning their cases, a pattern that is likely specific to the context of EEO litigation.

Our analysis also has methodological implications for the intersectionality literature. Most of the intersectionality literature to date has employed rich qualitative analyses, which have revealed much about the nature of intersectional disadvantage and intersectionality's effects on particular plaintiffs and in particular cases. But until now, we have known little about the general effects of intersectionality on litigation outcomes. Our work demonstrates the power of quantitative methods and categorization for documenting inequalities in litigation and judicial decisionmaking.

In addition to contributing to the intersectionality literature in critical race theory, our work has important implications for emerging research in social psychology and stratification, which demonstrates that intersectional discrimination occurs in the labor market. Our findings regarding demographic intersectionality demonstrate that intersecting demographic characteristics shape outcomes in the courts as well as in the labor market. Moreover, our findings regarding claim intersectionality establish that EEO law provides little protection for plaintiffs facing intersectional discrimination in the labor market.

Plaintiffs who suffer multiple disadvantages in society fare worse than do singly disadvantaged plaintiffs when they seek to assert their civil rights in court. This disadvantage operates both through demographic intersectionality, where the intersection of race and sex puts black women at a disadvantage, and through claim intersectionality, where those who assert two or more types of discrimination fare worse than do those whose cases are simpler. By assuming that disadvantages based on race, sex, and other ascriptive characteristics operate independently, civil rights law perpetuates intersectional disadvantages.