Race, gender, and place: How judicial identity and local context shape anti-discrimination decisions

Abstract

While federal anti-employment discrimination laws have helped diminish inequality at work, discrimination persists, in part perhaps due to unequal handling of equal employment lawsuits. Prior research demonstrates that the definition of discrimination can vary based on local normative ideas, while another line shows that a judge's race or gender can shape how related lawsuits are handled. In this article, I draw on a set of EEOC workplace discrimination cases prosecuted in Federal Court and combine it with locality data, to analyze: (1) the impact of local context, specifically rurality, local political context, and southerness and; (2) how judges' race and gender interact with the local cultural-milieu. Findings reveal that plaintiffs of colour in race discrimination cases fair worse in rural courts or before white judges. Meanwhile, white judges in conservative areas are more defendant friendly than those outside the in more liberal areas. Black judges, in comparison, are more plaintiff friendly in conservative areas when compared to black judges in more liberal areas. While female judges are generally more plaintiff friendly than male judges in sex cases, location has no discernible effect.

Introduction

Title VII of the Civil Rights Act of 1964 has reduced employment discrimination, with successful claims offering a powerful potential remedy for plaintiffs (Bisom-Rapp, 1999; Donohue III & Heckman, 1991; Kalev & Dobbin, 2006; Leonard, 1990; Pedriana & Stryker, 2017; Skaggs, 2008, 2009). Lawsuits, according to recent research, can provide deterrence via increased employer awareness of misconduct and by impacting company reputation with consequences for financial well-being (Hirsh & Cha, 2015; Skaggs, 2009). Moreover, effective enforcement can reduce race and gender disparities in the company sued and in nearby companies (Burstein & Edwards, 1994; Kalev & Dobbin, 2006; Skaggs, 2009) and change employment norms (Hirsh, 2009; Skaggs, 2008). Employment discrimination persists; however, raising questions about effective enforcement of EEOC law and whether legal processes can, in and of themselves, remedy discriminatory treatment (Bisom-Rapp, 1999; Bobbitt-Zeher, 2011; Roscigno et al., 2007).

A constant impediment to consistent enforcement of Civil Rights protections is the ambiguity of anti-discrimination law. While discrimination in the workplace on the basis of a protected class status is formally prohibited, precisely what behavior constitutes discrimination is less clear. Judges consequently have substantial interpretational discretion as to what conduct is legal versus nonlegitimate. In this article, I address this by focusing on the degree to which the race and gender of judges and local community norms shape enforcement of anti-discrimination law. I build on the recent work of Berrey et al. (2017) and Edelman (2016) in examining how social context influences the legal processes and biases they have effectively documented.

My data entail a large sample of federal EEOC cases coupled with personal data on judges and data for the community in which the case was heard. My analyses add to the existing research in at least two important regards. First, while previous research suggests that judges race and gender are relevant to employment discrimination cases, it is equally essential to recognize that judges work in local contexts—contexts (e.g., rurality, region, etc.) that may intersect with judges' own background status attributes in a way that shapes decisionmaking. Secondly, my analyses consider two novel measures of local social context—that is, citizen partisanship and rurality—that, to date, have not been incorporated into analyses of employment discrimination litigation. These two extensions are important from a general social scientific standpoint: If we are to understand law as a socially constructed institution, it is essential to examine how judges behave within local social contexts. This article does so by integrating ideas from both an institutional approach to enforcement of anti-discrimination and the court communities perspective.

Constructionism and Employment Discrmination

My approach derives from a legal constructionist framework, as articulated by Nielsen and Nelson. A constructionist framework views the law as “a culturally and structurally imbedded social institution” in which parties continually construct and redefine legal meaning (Nielsen & Nelson, 2005; Suchman & Edelman, 1996). Such a framework contrasts with legal formalism, arguably the dominant mode of thought for judges and lawyers, which holds that to understand the law means understanding and implementing clear and unambiguous law on the books based on the facts of the case (Moran, 2010; Roberts Jr., 2005).

Under a constructionist approach, the law, including written statutes and judicial precedent, are best conceived not as dictators of the outcome of a case, but rather as guides on the breadth of discretion. After the passage of the Civil Rights Act, courts grappled with the meaning of the law, but over time, the degree of consensus rose, narrowing, the range of judicial discretion. Deeming something illegal discrimination is not a separate act from defining what illegal discrimination is. Instead, the act of deeming an act discriminatory, contributes to the definition.

A constructionist framework argues that the law arises via interactions of parties, each of whom attempts to define the law “in specific ways in specific contexts” (Nielsen & Nelson, 2005). Discrimination is thus not a universally recognized and objective behavior but is instead subject to a shifting and continually contested definition, based on the parties and context. This is not to say the law is absent any relevant precedent or constraints on discretion. Instead, it means a definition capable of predetermining the outcome of any discrimination suit does not exist and instead, the question of what constitutes illegal discrimination varies by context and individual (Hirsh & Sabino, 2008; Huffman & Cohen, 2004; Marshall, 2005). Furthermore, contemporary discrimination is more subtle than the blatant discrimination of the past (Dietch et al., 2003; Sturm, 2001). Therefore, to understand the construction of any law, we much understand both the actors and the context in which they operate. In the present employment discrimination research, the relevant actor is the judge ruling in the employment discrimination case.

According to the framework described by Nielsen and Nelson, the construction of meaning when it comes to anti-discrimination law occurs in six fields: “We can think of anti-discrimination law as being produced in at least six distinct fields that are worthy of analysis in their own right. The fields include: (1) the judicial field, which consists of the federal and state courts that interpret anti-discrimination law; (2) the legislative field, which includes the United States Congress; (3) the regulatory field, including the Civil Rights Commission, the EEOC, and the Office of Federal Contract Compliance Programs (OFCCP); (4) employing organizations; (5) the academic field; and (6) politics and public attitudes” (Nielsen & Nelson, 2005).

This paper is prompted in part by Nielsen and Nelson's argument that while studying individual fields is important, a full understanding of employment discrimination requires studying the areas where fields come together. This is where they begin to reciprocally shape each other (Nielsen & Nelson, 2005). Such an insight is core to my analyses and reflects an important extension to prior research surrounding employment discrimination.

Previous research can be broadly placed in three categories: (1)Work that examines how a field internally constructed concepts about employment discrimination, (2) Research that looks at the definition of illegal discrimination that emerges when two fields come to contact with each other, and (3) Examinations of how individual identity shapes how one conceives of employment discrimination. From this, a fourth category seemingly should exist, which focuses on how individual identity shapes conceptions of employment discrimination at the intersection of fields. This is what the present research aims to accomplish, by studying the intersection of the judicial field and the field of politics and public option, while also recognizing that judges are individuals with identities that could mediate how the meaning of discrimination is generated at this intersection.

Constructions within a field

Research on constructions within an individual field tend to focus on employer organizations. In Inventing Equal Opportunity, Frank Dobbin traced how employer organizations, grappled with the ambiguous new federal mandate that prohibited employment discrimination and over time arrived at what he described as “a legal code internal to the corporation” This code consisted of equal opportunity rules that existed only within the human resource manual. These policies and programs, while not present in any federal or state statute, began to define what equal employment and anti-discrimination policy was. For example, concepts such as objective performance evaluations, publication of jobs beyond the existing workforce, diversity training, and official grievance protocols were constructed by personnel experts and not judges or legislators (Dobbin, 2009). Lauren Edelman similarly traced how employers created largely symbolic rules and offices that created symbols of compliance, without creating any substantive changes (Edelman, 1992, 2016; Edelman & Cabrera, 2020).

Constructions at overlapping fields

Edelman's work however also highlights the fact that individual fields rarely construct meaning without any substantive input from overlapping fields. In the case of Edelman's work she highlights how the judiciary increasingly deferred to the employer constructed meanings, eventually giving them greater weight, which resulted in greater proliferation of these symbolic compliance measures by employer (Edelman et al., 2011). This interaction, known as the endogeneity of law, examines how employers, the group intended to be regulated by EEO law, have shaped the judiciary's opinions on how that law is to be applied. (Bisom-Rapp, 1999; Dobbin, 2009; Edelman, 2005; Edelman et al., 1999; Edelman et al., 2011; Nelson et al., 2008; Suchman & Edelman, 1996). This intersection is not one way however, as the judiciary has been successful in reducing discriminatory business practices via findings of employer liability (Burstein & Edwards, 1994; Donohue III & Heckman, 1991; Hirsh, 2009; Skaggs, 2008, 2009). A growing body of research however has started to address how politics and public attitudes shape employer and employee conceptions of discrimination by signaling what behavior is deemed discriminatory (Beggs, 1995; Edelman, 1990; Garnett, 2012; Skaggs, 2008, 2009; Stainback & Tomaskovic-Devey, 2012). Tomaskovic-Devey and Stainback detailed in their book Documenting Desegregation how national political attitudes and leadership shaped how and by how much workplace segregation changed over time following the passage of the Civil Rights Act (Stainback & Tomaskovic-Devey, 2012).For example, as national politics moved towards neo-liberalism, de-segregation stalled, and as Reagan administration policies moved away from racial equality and towards gender equality, woman achieved greater gains in desegregation than Black Americans (Stainback & Tomaskovic-Devey, 2012). Related research looking at local attitudes found that workplace desegregation regulation has a more significant impact in politically liberal jurisdictions (Guthrie & Roth, 1999; Kalev & Dobbin, 2006; Skaggs, 2008, 2009). Employers in more egalitarian minded communities and with more same-sex couples, tend to have less inequality (Beggs, 1995; Garnett, 2012; Skaggs, 2008). Greater levels of voting for Democrats is correlated with lower rates of occupational segregation by race and gender (May & McGarvey, 2017; McVeigh & Sobolewski, 2007).

Individual constructions of discrimination

Finally, previous research has demonstrated the variety of ways in which individuals construct their understanding of discrimination and the factors that influence these definitions. Personal attributes such as one's age, income, and education influence perceptions of unfair treatment (Bobo & Suh, 2000). Black Americans, Latinos, and women are more likely to perceive discrimination at work (Hirsh & Lyons, 2010; Mccord et al., 2017).

Social context can also shape individual employees' perception of discrimination. Research shows that differences in personal, community, and organizational characteristics influence whether an employee constructs various behaviors as illegal discrimination (Avery et al., 2008; Marshall, 2005; Stainback et al., 2011). Employees are more likely to identify workplace discrimination when they perceive societal wide discrimination as higher, when local norms reject gender discrimination, and when the parties involved differ in race or gender. (Avery et al., 2008; Eyer, 2012; Kartolo & Kwantes, 2018; Schaffer & Riordan, 2013; Stefanie et al., 2013).

Organizational factors, such as levels of integration, formal diversity training and standardized employee screening procedures also impacts whether behavior is constructed as discriminatory (Avery et al., 2008; Hirsh & Kmec, 2009; Hirsh & Lyons, 2010; Hirsh & Sabino, 2008; Stainback et al., 2011). Finally, an employer's construction of their anti-discrimination policy can also shape employees' definition of discrimination (Marshall, 2005). If lay-persons' perceptions of what constitutes discrimination is shaped by personal factors and local norms, why should not judges be subject to the same? Research in other realms already suggests that judges are not much different from lay-persons when it comes to the influence of background beliefs on decisionmaking (Englich et al., 2006; Guthrie et al., 2001; Miller, 2019).

In the next section, I focus on how both community level factors and the race and gender identity of judges contribute to the construction of discrimination. I then look at how local community and judicial identity interact to construct a person and place specific understanding of discrimination.

Discrimination, Court Communities and Judicial Identities

Local normative environments and legal decisionmaking

A theoretical framing related to the legal constructionist approach is the community court's perspective. This approach recognizes that courts are embedded in communities, which can impact how cases are handled within the courtroom. This article conceives of this perspective as an application of the overlap between the public and political opinion fields and the judiciary fields. Peter Nardulli, Roy Flemming, and James Eisenstein's decades long mixed methods study of trial courts in Illinois, Michigan, and Pennsylvania laid the foundation for showing how local norms and existing relationships shape the handling of legal disputes (Flemming et al., 1987; Nardulli, 1986; Nardulli et al., 1980; Nardulli et al., 1984, 1985). The majority of contemporary research on the Court Communities perspective has focused on criminal law, but the insights remain valuable within a discussion of employment discrimination. Research in the field of criminal justice suggests the likelihood of conviction, sentence length, and capital punishment, especially for Black defendants, varies by local factors such as region of the country and percent republican (Baumer & Martin, 2013; Feldmeyer & Ulmer, 2011; Helms & Jacobs, 2002; Jacobs & Carmichael, 2004). Given that employment discrimination and criminal punishment are forms of social control of minority members (Blalock, 1967; Huffman & Cohen, 2004; Johnson et al., 2011), it is little stretch to suspect that the same factors that increase criminal punishment would also cause lax enforcement of anti-discrimination laws.

The Court Communities perspective recognizes that local communities shape even the federal judiciary (Feldmeyer & Ulmer, 2011; Ulmer & Johnson, 2017). While they are appointed instead of being elected by citizens, there is still reason to suspect federal judges are influenced by the community in which they work. Federal judges typically work in a local court district for their entire career. Most judges were also born in the region they eventually serve (Neubauer & Fradella, 2015; Richardson & Vines, 1970). By law, they are required to live in the district in which they work (Judiciary and Judicial Procedures, 2012). Furthermore, federal judges join the bench following years of local politicking; therefore, judges already have strong social networks in their district (Richardson & Vines, 1970). Senators within the circuit have an opportunity to provide recommendations to the White House on how to select (Rutkus, 2013). Senators from the state the nominee was selected from could also provide an opinion to the Senate on the fitness of the nominee (Rutkus, 2013). This process was one more opportunity for local influences to shape who makes it to the bench.

Legal decisions could be influenced by local context because, as evidenced by judicial politicking, the president will only nominate attorneys who already reflect the norms of the local district (Richardson & Vines, 1970; Songer & Davis, 1990). Additionally, and perhaps more importantly, residing in a district while on the bench, could shape the normative criteria judge may activate in the court of decisionmaking (Richardson & Vines, 1970). While federal judges have a job for life, and are allegedly immune to political pressures, there is evidence showing that local politics influences federal judges (Baumer & Martin, 2013; Richardson & Vines, 1970).

Prior research in criminal sentencing and employment discrimination research has identified three factors of social context that may be particularly likely to shape decisions in employment discrimination cases: local political orientation (Garnett, 2012; McVeigh & Sobolewski, 2007; Skaggs, 2008, 2009; Songer & Unah, 2006), rurality (Dumas & Haynie, 2012; Eason et al., 2017; Oliver & Mendelberg, 2000; Songer & Unah, 2006), and presence in the south (Glaser, 1994; Oliver & Mendelberg, 2000; Songer & Davis, 1990). All three are relevant to shaping local norms related to discrimination, race, and gender. For examples, the traditional perspective is that the South has more discrimination; nonsoutherners are more likely to believe equality does not yet exist, and southern Whites tend to hold more negative views of Black Americans.

While in many cases, these three conditions, rurality, southerness, and Republican support exist in the same place, they still can be independent. For example, large cities in the south can be left leaning, while cities outside the south, especially prior to the 2016 election, could lean conservative. Rural communities in the south can likewise be majority Black American, and, thus, lean Democratic in voting. Examination of all three allows study of their independent effects, even if there is substantial overlap in some locations.

Additionally, even though there may be some overlap in the existence of racism and sexism, they do however remain different processes. A region or person need not be retrograde or liberal regarding one process simply because they are regarding the other. Therefore, some factors discussed below may apply to both race and sex discrimination cases, while others will only apply to a single one.

How local politics may matter

Previous research suggests that state political climate can affect how courts handle employment discrimination claims and can mediate how discrimination lawsuits change managerial diversity by signaling what kind of behavior is acceptable in the community (Skaggs, 2009). Previous research also shows higher rates of support for Republicans by voters is correlated with higher rates of occupational segregation by gender and race (McVeigh & Sobolewski, 2007). Greater gender segregation at businesses in communities may also be influenced by communities views on gender equality (Garnett, 2012; Skaggs, 2008). Thus, the normative environment in Republican counties may be such that gender and race discrimination is less likely to be seen as illegal discrimination and more likely to be seen as the usual practice.

Conservative voters are also more likely to reject the existence of racial inequality (Pew Research Center, 2016). Republicans are more likely to believe that racial equality now exists, and thus, may be less likely to support claims of discrimination. Republicans are also more likely than Democrats to believe that the government has done too much to try to achieve racial equality (Pew Research Center, 2004). This opinion may increase rejection of government-backed allegation of discrimination. Somewhat less apparent, but equally important, are ideologies connected to the role of government on business decisions with Republicans expressing less support for government intervention in business decisions (Gallup, 2013). Therefore, even if they believe racial discrimination is wrong, they may be more likely to be opposed to the government creating an enforcement system.

There is, however, no research directly examining whether local voting patterns affects judicial decisions in employment discrimination cases. However, a substantial body of research has examined whether judges sentence Black criminal defendants differently based on levels of Republican voting (Helms & Jacobs, 2002; Jacobs & Carmichael, 2001, 2004; Ulmer et al., 2008). Given the established link between political views and views on race and gender, and evidence from criminology that Republican voting patterns influence judicial decisionmaking it follows that courts in Republican divisions may be less supportive of claims of racial discrimination.

Hypothesis 1 Plaintiffs alleging race or gender discrimination are less likely to win in Republican districts compared to Democratic districts.

Rurality as a potentially influential context

Research has also suggested that local rural embeddedness affects legal decisions. The majority of the research has examined severity of criminal punishment and racial disparities in punishment (Britt, 2000; Steffensmeier & Britt, 2001; Ulmer & Kramer, 1996). Prosecutors are five times more likely to seek the death penalty in rural areas than urban. This effect may be due in part to how support for capital punishment and racial prejudice interact (Songer & Unah, 2006). Research shows that urban and rural counties exhibit different racial attitudes (Fennelly & Federico, 2008; Tolbert & Grummel, 2003). A 2013 Pew study shows that rural, suburban, and urban residents have vastly different opinions on whether Black Americans are treated fairly at work, by the police, by the courts, in elections, at schools, in businesses, and by the healthcare industry (Pew Research Center, 2013). Previous research looking at all cases in Alabama found variations in civil cases proceeded, based on whether they occurred in a rural or urban county (Dumas & Haynie, 2012). Despite evidence that rural residents tend to have different racial opinions than White Americans, no research has examined if this influences decisions in employment discrimination cases. The author is unaware of any research studying this in the context of employment discrimination. However, as both criminal punishment and discrimination are often intertwined with conceptions of race, it would be fruitful to examine rurality in the context of employment discrimination. Given the previous findings on the relationship between race and rurality in criminal punishment contexts, Hypothesis 2 predicts a similar effect in race discrimination cases only.

Hypothesis 2 Plaintiffs are less likely to win their race discrimination claims in rural divisions compared to nonrural divisions.

The South and potential effects

Finally, previous research demonstrates that the South continues to exhibit its own unique social context, generating its own behavioral norms. Therefore, we can expect variations in case outcomes based on whether the case is adjudicated in the South.¹ This is important because traditionally speaking; the south is less progressive on race (Glaser, 1994; Orey et al., 2011) The traditional view is still supported by contemporary research (Mas & Moretti, 2009; Oliver & Mendelberg, 2000). Researchers frequently examine differences in cases in the South versus the rest of the country. For example, research on death penalty cases and voting rights consistently demonstrates variations in case handling by region (Alesina & La Ferrara, 2014; Songer & Unah, 2006). Research already highlights that southern courts are less favorable to Black plaintiffs and defendants (Alesina & La Ferrara, 2014; Collins & Moyer, 2008; Vines, 1949). Given the previous findings on the relationship between race and southerness in legal decisions, Hypothesis 3 predicts a similar effect in race discrimination cases only.

Hypothesis 3 Plaintiffs in the South alleging race discrimination are less likely to win than plaintiffs in the North.

It should be noted, that while there is substantial association between southern, rural, and GOP dominant regions, these three do not always co-exist. For example, in the Black belt, there are many rural, southern, but non-GOP dominated counties. Outside the south, cities like Phoenix and Salt Lake City showed republican leaning in voting.

Judicial identities

A judge's attitudes can shape their construction and interpretation of the law (Chew & Kelley, 2008; George, 2007; Johnson et al., 2008; Sisk et al., 1998). The evidence is mixed as to whether a judge's race, gender, and politics are relevant to how s/he decides cases in general (Ashenfelter et al., 1995; Johnson, 2014; Johnson et al., 2008; Steffensmeier & Britt, 2001) Instead, there are theoretical reasons to believe that a judge's identity might only matter for particular types of cases. For example, evidence suggests that a judge's race, gender, and political affiliation are important predictors of how a judge will rule in employment discrimination cases along with death penalty and criminal cases (Boyd, 2016; Boyd et al., 2010; Chew & Kelley, 2008; Collins et al., 2010; Collins & Moyer, 2008; Johnson, 2014; Steffensmeier & Britt, 2001).

A simplistic account would argue that judges of different races or genders simply hold wildly divergent attitudes on almost all legal matters, which in turn results in reaching different opinions. However, for the most part, the research demonstrates that judges, regardless of race or gender, have been socialized into the judiciary and limited by organizational constraints such many matters of law show no significant difference between the decisions of judges of different races or genders (Boyd, 2016; Johnson, 2014; Morin, 2013; Steffensmeier & Britt, 2001). However, a more finely tuned account would turn to the different lived experiences of Black versus White and female versus male judges to suggest that such differences would lead them to be more or less sympathetic to or quicker to recognize discrimination (Boyd, 2016; Boyd et al., 2010; Weinberg & Nielsen, 2012). A quote from Black American Federal Circuit Judge Harry Edwards supports this theory stating “Because of the long history of racial discrimination and segregation in American society, it is safe to assume that a disproportionate number of Blacks grow up with a heightened awareness of the problems that pertain to these areas of the law” (Edwards, 2002).

Research increasingly suggests that it is only in particular types of cases, which activate race or gender identity, that may demonstrate an effect based on judicial race (Boyd, 2016; Boyd et al., 2010; Davis et al., 1993; Songer & Crews-Meyer, 2000; Songer & Lindquist, 1996). Employment discrimination law is one field in which a judge's personal identity may directly relate to the case (Boyd, 2016; Chew & Kelley, 2008; Davis et al., 1993; Weinberg & Nielsen, 2012). For example, Black and female judges may be more to have been personally discriminated against, especially in the subtle ways that have become more common. White or male judges meanwhile may be more likely to have been socialized to believe that discrimination is a thing of the past, and when it exists, it does so in open and obvious way.

This is not a one way effect, as White Americans, are themselves socialized in racially relevant ways, such as being less attuned to discrimination (Priest et al., 2014). Surveys in fact indicate that White Americans are less likely to believe racial discrimination in the workplace remains a major concern (Hertz et al., 2020). Meanwhile, men are less likely to report discrimination at work or believe that gender discrimination will hamper their career when compared to women (Parker & Funk, 2017; Sipe et al., 2016). Men are also more likely than women to oppose workplace gender equity measures (Scarborough et al., 2019). While judges do undergo an additional socialization process in law school and in their legal practice, research does suggest that judges still hold biases, often implicit ones (Levinson et al., 2017; Miller, 2019; Rachlinski et al., 2009).

Existing research suggests that while Black judges and female judges may view the world differently from their counterparts, this does not mean they will thus reach the same legal conclusions. Some research demonstrates that while female judges are more likely to rule in favor of plaintiffs alleging gender discrimination, Black judges may be more likely to rule in favor of plaintiffs alleging race or sex discrimination, suggesting Black judges may have a broader sympathy for plaintiffs, and not merely towards Black plaintiffs (Boyd, 2016; Chew & Kelley, 2008). Though not previously demonstrated, it is possible that female judges similarly show sympathy for all plaintiffs, and not just those alleging sex discrimination.

Several theories exist for why Black and female judges rule differently than their White and/or male counterparts. They may see themselves as group representatives, they may have greater empathy for plaintiffs and/or past personal experience with discrimination, or they may have different ideological/political views than White and male judges (Boyd, 2016; Collins et al., 2010; Edwards, 2002; Farhang & Wawro, 2004; Johnson et al., 2008; Weinberg & Nielsen, 2012, 2017). I put forth two hypotheses regarding judicial identity.

Hypothesis 4 Black judges will be more likely than White judges to rule in favor of plaintiffs in race and sex cases.

Hypothesis 5 Female judges will be more likely than male judges to rule in favor of plaintiffs in sex and race cases.

The intersection of the personal and local context

A constructionist approach must go beyond merely looking at identities and instead consider how embeddedness in a community might interact with these identities. Unfortunately, to date, there is limited research studying how local community norms may impact judges differently depending on the judge's background. The research that exists does suggest that gender can mediate the effects of the southern variable (Davis et al., 1993). More generally, research supports the idea that lay people's perception of discrimination is affected by their race/gender and the local context (Avery et al., 2008). Furthermore, evidence suggests that judge's biases may not radically differ from those of lay people (Miller, 2019).

It is plausible that judges of color in defendant friendly, south, Republican, or rural districts, may be more conservative than their White counterparts. First, judges of color may experience more normative pressure to demonstrate that they reflect community values. Second, for a Black attorney to be considered for a judgeship in these districts, they may have to be clearly friendly to local values (i.e., a judge selection effect). If this is, in fact, the case, more moderate or even progressive Black judges may be selected out of the process or, at the very least, experience more pressure to conform. There is evidence for a similar effect in the criminal sentencing literature as Black judges may sentence more harshly (Steffensmeier & Britt, 2001). Alternatively, the identity and experience of Black and female judges may be most salient in conservative areas, such that they would be even more plaintiff friendly than comparable judges outside those areas. This leads to two competing hypotheses.

Hypothesis 6 Black and female judges in conservative areas will be more plaintiff friendly than comparable judges outside those areas.

Hypothesis 7 Black and female judges in conservative areas will be less plaintiff friendly than comparable judges outside those areas.

Data and Methods

Data

The primary data come from the EEOC Litigation Project (Kim et al., 2013) which is composed of case and judge details from a random stratified sampling of employment discrimination lawsuits filed in federal courts between 1996 and 2006 by the Equal Employment Opportunity Commission (“EEOC”) (Boyd, 2016). The population consisted of all federal cases filed by the EEOC between fiscal years 1997 and 2006. The sample includes all class actions, cases concluded by court order, and cases with a set trial date, along with a random sample of the remaining cases (Kim et al., 2013). I limit my cases analysis to cases alleging race or sex discrimination. In line with previous research, I exclude race claims brought by White plaintiffs and sex claims brought by males (Boyd, 2016). This result in six race cases dropped and 60 sex cases dropped. These claims can present questions of power dynamics that are incongruous with cases filed by women and people of color. This is particularly important in the context of this article, as the importance of social context largely relates to how traditionally, disadvantaged individuals are treated or viewed within particular communities. Including such cases could mute the effects of local community context. In fact, it could be that White plaintiffs alleging race discrimination perpetrated by Black Americans may be best received by courts located in communities exhibiting bias against Black Americans.

The value of these cases for my research emerges from the EEOC investigatory process that occurs before these cases ended up in court. The process screens out cases with no or limited merit to focus attention on strong cases to file. In a mere 3% of cases does the EEOC find merit and file a lawsuit on behalf of the employee (Nielsen et al., 2008). This eliminates the concern that a high volume of frivolous cases, which judges uniformly decide, could hide variations in decisionmaking (Boyd, 2016). Though limited in the sense of being not necessarily reflective of all cases filed each year at state of Federal levels, such data are important for my purposes owing to the fact that: (1) they provide significant detail on case proceedings and decisionmaking, and (2) offer necessary and important variations on my two outcomes of interest (i.e., decisionmaking on motions during the case proceedings, and final decisionmaking on the case itself).

In order to understand the construction of the locality data, the organization of the Federal courts has to be understood. Federal court districts, the lowest level of the system, are further divided into divisions via judicial order. These divisions are where the decisions I examine are made, and they are the smallest way to physically divide the federal court system. Therefore, examining cases at this level allows the normative environment to be measured in the finest way possible. Each of these divisions is made up of multiple counties. Therefore, after linking each of the over 3000 counties to their respective division, I then aggregate county level data up to the division. I used data from 2000 census to weight the counties appropriately during the aggregation (Census, 2013).

County population density data comes from The National Center for Health Statistics Urban–Rural Classification Scheme for Counties (Ingram & Sheila, 2012). Political data consist of county presidential voting data for 1996, 2000, 2004, and 2008. Data for 1996 comes from Dave Liep's Atlas of US Presidential Elections (Leip, 2004). Data for 2000 is from David Lublin and D. Stephen Voss's “Federal Elections Project” (Lublin & Voss, 2001). I obtained data for 2004 and 2008 from the US Geological Survey, Department of the Interior (National Atlas of the United States, 2004). I then combined each county's raw vote count for Democratic, Republican, and Independent candidates for 1996, 2000, 2004, and 2008, and determined the total average support for Republicans in that county.

Measurement

Two key dependent variables are the focus of my analyses. Both center on whether the plaintiff or defendant won, which is fundamentally connected to the question of what the definition of discrimination is. The first looks at judicial decisions on motions within each case. After reductions for incomplete data, the analysis is left with 584 race motions and 963 sex motions. A motion is a filing by the plaintiff or defendant that asks the judge to make a decision that can impact the case. They can address a variety of topics, such as requesting dismissal of the case, a verdict in favor of the plaintiff, instructions to a party to produce a piece of evidence, or permission to avoid providing that evidence. The majority of the motions included address if discrimination occurred or whether a particular piece of evidence is relevant to the question of whether discrimination occurred and needs to be produced. Therefore, these motions directly address the theoretical question of whether social context influences the construction of what constitutes discrimination. Many of these motions, known as motions for summary judgment ask the judge to answer the question: Even if you assume everything alleged is true, is it even illegal discrimination? If a defendant wins a motion for summary judgment, the entire case could be dismissed in the defendant's favor (Boyd, 2016). Therefore, if in a counterfactual scenario two judges made different rulings on the same motion for summary judgment, it follows that each judge has a different construction of discrimination.

The second dependent variable looks at the final case outcome (i.e., did the plaintiff or defendant win the case). With the exception of cases that were still ongoing at the time of data collection, every motion belongs to a case that concluded for or against the plaintiff. I coded this variable based on the winner of decision rendered, with those indicating a form of plaintiff victory, including monetary settlements of any size, and summary judgment for the plaintiff, coded as plaintiff wins, while others, including dismissals of the case and verdicts for the defendant, were coded as defendant wins. Any case lacking a final decision was excluded from this level of analysis. From the initial sample, there are 226 race cases and 375 sex cases. While initially included, jury verdicts were excluded, as they capture the effect social context has on parties other than the judge, which is deserving of its own separate article.

My core independent variables include the race and gender of the judge and three variables connected to location: population density, politics, and south v not south. All variables, except the political measure, are coded as dummy variables. Judicial race uses White as the reference group. Due to their still limited numbers in the federal judiciary, Latino, Native American, and Asian American judges were omitted from my analysis leaving the comparison as Black judges versus White Judges.² Judge sex uses male as the reference category. Population density uses with nonrural as the reference category. The southern variable uses nonsouthern as the reference category. Following previous research, I define the south as the Confederacy, which includes Alabama, Georgia, Florida, Louisiana, Mississippi, North Carolina, South Carolina, Tennessee, Texas, and Virginia, plus Kentucky, which is defined as Southern in various literatures (Glaser, 1994; Jacobs et al., 2005; King et al., 2009). Finally, the political variable is a mean-centered continuous measure of total percentage of Republican votes in presidential elections in the division between 1996 and 2008.

I also include in my analyses controls for whether the motion/case has a private attorney on it. There is evidence suggesting private attorneys have more success than EEOC attorneys (Selmi, 1996). This variable also helps control for overall case quality as private attorneys evaluate the likelihood of case success before agreeing to take it (Goodman-Delahunty et al., 2010). It also accounts for concerns about unequal access to private attorneys depending on location.

I control for the party of the president who appointed the judge, as voting patterns and political party are correlated (Boyd et al., 2010; Pinello, 1999; Sisk et al., 1998; Songer & Davis, 1990). Additionally previous research demonstrates that some of the perceived effects of race and gender could be related to the judge's party, because judges of color and female judges are more likely to be Democrats (Collins et al., 2010; Davis et al., 1993). As a result of the need to control for the party of appointment, my analysis omits magistrate judges, who, unlike other federal judges, are not politically appointed.

I control for class actions, as cases with multiple plaintiffs are likely to be stronger cases and previous research demonstrates that class actions tend to be more successful (Nielsen et al., 2008; Nielsen et al., 2010). Class action lawsuits are also more common in the south and so controlling for this helps to avoid conflating a class action effect with a southern effect.

I control for the circuit court environment, because similar judges under different circuits can be in very different institutional environments, due to controlling precedent, which can constrain their decisions. Controlling for this allows local effects to remain visible. This variable looks at whether the case occurred in the second or third circuit, or one of the remaining ten. This definition has been used in previous research, as the second and third circuit as seen as more plaintiff oriented in employment discrimination cases (Guthrie & Roth, 1999; Hirsh, 2009).³ Omitting this variable could alter the findings of the local influences (Scott, 2006)

When examining individual motions, I also include controls for whether the motion addresses a discovery issues as discovery motions are far more likely to be decided in the plaintiff's favor. Discovery motions address whether a particular piece of evidence is relevant to the question of whether discrimination occurred and needs to be produced. I also control for the number of motions filed in the case, as a greater number of motions may indicate a party is filing more motions that are frivolous. When examining my case level outcome, I also control for the percentage of motions in the case won by the plaintiff as winning motions along the way would seemingly play a role in winning cases.

Finally, I control for whether the case alleges only a single form of discrimination. This primarily controls for the intersectional and/or double disadvantage previous research suggests that Black women face when alleging discrimination (Best et al., 2011). Potentially such an effect could exist for other claimants alleging other multiple forms of discrimination, such as race and disability or gender and religion. As data on the race of plaintiffs only exists for those alleging race discrimination, there is unfortunately no way to control for this potential intersectionality when the plaintiff does not themselves allege multiple forms of discrimination.

Methods and descriptive data

A two-outcome logistic regression forms the backbone of my analysis. Logistic regression is common for analysis of legal decisions (Best et al., 2011; Boyd, 2016; Steffensmeier & Britt, 2001). In line with previous research, which examined race and sex discrimination claims separately, this article does the same. (Boyd, 2016; Skaggs, 2008, 2009). Theoretically, both the regional and judicial identity effects may express themselves in different ways depending on the type of discrimination alleged (Boyd, 2016; McVeigh & Sobolewski, 2007; Songer & Davis, 1990; Stainback & Tomaskovic-Devey, 2012). Given that individual judge can have their own dispositions and the same judge may have worked on multiple motions, clustered errors are used in all regression models of motion outcomes. As my analysis occurs at two levels, motions and final case outcomes, I report descriptives for both outcomes, separated between the race cases and sex cases. Table 1 contains the values at the motion level. Table 2 contains the values at the case level. Since some cases have multiple motions, the overall sample size changes depending on the outcome being examined. Additionally, because each case can have multiple judges, the judicial identity variables do not reflect individual judges, but are instead dummy variables for whether the majority of the judges handling the case were White, male, or Republican.

Table 1. Motion level descriptive statistics

Table 2. Case level descriptive statistics

There is little variation in my dependent and independent variables by case type. For example, plaintiffs win about 72% of the motions and around 86% of the cases, regardless of whether it is a race or sex discrimination case. This success rate does not reflect employment discrimination cases generally, but is a function of the EEOC case screening and selection process. There is, nevertheless, significant enough variation at the motion and case determination levels to pick up systematic differences and/or biases.

Cases are randomly assigned to judges, so we should not expect any substantial differences in judge race, sex, and party based on case type (Abrams et al., 2012; Ashenfelter et al., 1995). We do see that Black and female judges are slightly more likely to handle race cases than other types. We see that race cases are overrepresented in the south, while sex discrimination cases are overrepresented in in rural areas. The mean percentage-voting Republican is slightly lower in cases alleging race discrimination versus those alleging sex discrimination. Sex cases are also slightly more likely to have a private attorney on the case than race cases.

Approximately 17.5% of cases do not report the type of discrimination alleged. Therefore, these cases are excluded. In contrast to cases alleged discrimination based on something like age, these cases are excluded from the analysis without knowledge of whether they may in fact allege race or sex discrimination. As the presence of knowledge about the case type is critical to determining if the case is included, and in which category, this missing data about case type can be conceived as a selection effect problem, or a missing data problem. Therefore, two steps were taken to ensure that missing data or selection effects were not biasing the results. First, to account for potential missing data bias, separate probit models were created to determine the probability of being a race or sex case. This probability was then used to weight the logistic regression on motion and case outcomes. To account for potential selection effects, a Heckman correction was used. The Heckman correction is an identical two-step process, in which first, the probability for selection into the analysis is generated and then in the second step these probabilities are used to weight the regression. The difference between the two methods is that one generates predicted probabilities of being a race or sex case, while the other generates predicted probabilities of not missing data on the case type. The results provided are based on the baseline model without Heckman or missing data analysis, as all three models are largely consistent.

Results and Discussion

This section will start with a discussion of all of the hypotheses relevant to race discrimination cases, before proceeding to discussion of the hypotheses in sex discrimination cases. It will conclude with a discussion of the differences between the race and sex discrimination results.

The influence of judicial attributes and social context in race discrimination claims

Table 3 contains three models, all at the motion level of analysis. The first includes judge characteristics and context indicators, along with the control variables. The second model adds an interaction between judge race and percentage of Republican voters in the division. The third model removes the judge race and percentage of Republican voters in the division interaction and replaces it with an interaction between the judge race and the case being in the south.

Table 3. Logistic regression models of effects of judicial and regional variables on race motions standard errors adjusted for 202 clustered judges

^a Reference category is White judges.

^b Reference category is male judges.

^c Reference category is judges appointed by Democratic presidents.

^d Reference category is cases outside the south.

^e Reference category is nonrural areas.

^f Reference category is cases alleging sex discrimination in addition to race and other forms of discrimination.

^g Reference category is motions that do not address any discovery question.

^h Reference category is cases with only an EEOC attorney representing the plaintiff.

ⁱ Reference category is cases with a single plaintiff.

^j Reference category is cases from the second and third circuit.

^* p < 0.05; ^**p < 0.01;

¹ p < 0.1.

Table 4 displays Model 4, which reports the final case outcome for cases alleging race discrimination and includes the same personal identity and local context variables as model one. Because multiple judges can handle a single case, I base the judge variables on the race, sex and party of a majority of the judges on the case. Due to the shrinking sample size at the case outcome level, interactions are not modeled.

Table 4. Logistic regression models of effects of judicial and regional variables on race cases

^a Reference Category is majority White Judges.

^b Reference category is majority Male judges.

^c Reference category is judges appointed by Democratic presidents.

^d Reference category is cases outside the south.

^e Reference category is nonrural areas.

^f Reference category is cases alleging sex discrimination in addition to race and other forms of discrimination.

^g Reference category is cases with only an EEOC attorney representing the plaintiff.

^h Reference category is cases with a single plaintiff.

ⁱ Reference category is cases from the second and third circuit.

^* p < 0.05

^*** p < 0.001.

Models 1 and 4 allows for testing of Hypotheses 1–5. Hypothesis 1 had no support in either model. This hypothesis predicted that as a division became more politically conservative, plaintiffs would be less likely to win. Hypothesis 2, which predicted that plaintiffs would fair less well in rural divisions, was supported at the case level as seen in Model 4, but not the motion level Model 1. The effect is such that plaintiffs achieve some measure of victory in 94% of cases outside rural areas and 80% within them. Hypothesis 3, which predicted that plaintiffs in the south would have less success, was not supported by at the case or motion level.

In line with previous research, Hypothesis 4, which held that Black judges would be more plaintiff friendly was supported at the motion level. Motions before Black judges have an 84% chance of being decided in favor of the plaintiff, while before White judges the chances are at 72%. At the case level, the effect could not be tested, as there was perfect collinearity between cases handled by majority Black judges and the plaintiff winning at least something. Thus, while it could not be formally tested, this serves as some evidence that the effect may also exist at the case level.⁴ Hypothesis 5, which held that female judges would show more support for plaintiffs in race cases was not supported.

Models 2 and 3 test alternative Hypotheses 6 and 7 at the motion level, which tested whether Black judges before more or less plaintiff friendly as the division becomes more conservative. In the second model, I add an interaction between the judge's race and the percentage of Republican voters in the division. This model reveals an important result, which is that as divisions become more conservative, Black and White judges go in divergent directions. As seen in figure one below White judges become more likely to rule in favor of the defendant, while Black judges become more likely to rule in favor of the plaintiff. This is a crucially important finding, as it demonstrates that personal identity and local environmental context interact to produce multiple constructions. That level of Republican support was not relevant on its own, lends greater support for the idea that a constructionist framework must take into account how the identity of the parties impacts the influence of local context (Figure 1).

Figure 1. Influence of division politics and judge race on Plaintiff's probability of winning a motion in a race case

The third model add an interaction between judge race and whether the case is in the south. These results are marginally significant (<0.064) and show that while White judges are equally likely to rule in a plaintiff's favor in the south or outside, Black judges are more likely to do so in the south. Holding all other variables at their means, a motion before a Black judge in the south has a 97% chance of being decided in the plaintiff's favor, versus an 80% chance if it is before a Black judge in the north. For White judges, the chances are 71%, regardless of location. Analysis using standard in lieu of clustered errors shows this interaction to be significant at the <0.05 level. Given the low number of Black judges in general and especially in the south, there is reason to suspect that the change when using clustered errors reflects the limited computational powers with such low numbers of individual judges. Furthermore, for plaintiffs and attorneys, this low number still reflects their reality of the legal landscape. Thus, while it makes it more difficult to isolate the impact of race on this interaction versus individual judge dispositions, it is still of relevance to those in the legal practice (Figure 2).

Figure 2. Influence of locale and judge race on Plaintiff's probability of winning a motion in a race case

A few important findings emerge from the data. We can see that context alone does not appear to be a powerful influence on legal decisionmaking. The only exception is the finding that at the case level, rurality demonstrates some limited influence. This is not necessarily surprising, as the constructionist model emphasizes that the construction of legality occurs via interactions of involved parties. This is why it is of crucial importance to consider how those parties interact in a specific context. The findings that judicial race interacts with level of local Republican support and possibly with presence in the south is strong evidence of how local conditions mediate the construction. This is congruent with previous research of lay-persons, which demonstrated that perceptions of discrimination were shaped by how the local environment and personal identity shaped the influence of each other. This lends further support to the body of research showing that judges are often subject to the same influences as lay-persons.

Additionally, the final model, looking at the case level, finds that the percentage of motions won by the plaintiff is a significant predictor of whether the plaintiff wins the case. This is important because it demonstrates that the effects that race and local context have on decisionmaking is not merely of abstract intellectual concern. It has discernable effects on the plaintiffs ability to win their case and goes even beyond that. Given the previous research showing how lawsuits can serve to deter discrimination and increase levels of diversity, it stands to reason that these favorable decisions have a cumulative effect on local levels of discrimination and equality (Table 5).

Table 5. Logistic regression models of effects of judicial and regional variables on sex motions standard errors adjusted for 296 clustered judges

^a Reference category is White judges.

^b Reference category is male judges.

^c Reference category is judges appointed by Democratic presidents.

^d Reference category is cases outside the south.

^e Reference category is nonrural areas.

^f Reference category is cases alleging race discrimination in addition to sex and other forms of discrimination.

^g Reference category is motions that do not address any discovery question.

^h Reference category is cases with only an EEOC attorney representing the plaintiff.

ⁱ Reference category is cases with a single plaintiff.

^j Reference category is cases from the 2nd and 3rd circuit.

^* p < 0.05

^** p < 0.01.

The influence of judge attributes and social context in sex discrimination claims

The analysis of sex discrimination involves mirrors that of race cases. Table 5 reports the first three models. Model 1 consists of the personal judge and social context, plus the control variables. Model 2 includes an interaction between judge gender and local politics. Model 3 includes an interaction between judge gender and whether the case is in the south. The second and third models mirror the significant interactions from my race models. Table 6 reports Model 4, which is the judicial identity and social context variables at the case level. Similar to race cases, the judicial identity variables are based on a dummy variable of whether the majority of judges political affiliation, race, and sex.

Table 6. Logistic regression models of effects of judicial and regional variables on sex cases

^a Reference category is White judges.

^b Reference category is male judges.

^c Reference category is judges appointed by Democratic presidents.

^d Reference category is cases outside the south.

^e Reference category is nonrural areas.

^f Reference category is cases alleging race discrimination in addition to sex and other forms of discrimination.

^g Reference category is cases with only an EEOC attorney representing the plaintiff.

^h Reference category is cases with a single plaintiff.

ⁱ Reference category is cases from the second and third circuit.

^** p < 0.01.

Models 1 and 4 test the first five hypotheses. Hypotheses 1–3, examining whether judges are less likely to rule on behalf of plaintiffs in more republican areas, rural areas, and the south are not supported at either the motion or case level. Hypothesis 4, which studied whether Black judges were more likely to rule in favor of plaintiffs was not supported at the motion or case level. This contrasts with previous research that found some support. Supplemental analysis demonstrates that this resulted from the decision to include a greater variety of motions than used in previous research. Supplemental analysis demonstrates that when motions addressing a discovery issue are removed, Black judges are more likely to rule in favor of the plaintiff. Further study could shed some light on why this effect varies based on the types of motions included.

Hypothesis 5, which predicted that female judges would be more plaintiff friendly than male judges was supported at the motion level, but not the case level. The results show that motions before a female judge have an 84% chance of success compared to 75% for motions before a male judge. This is consistent with previous research.

Models 2 and 3 test alternative Hypotheses 6 and 7, to examine whether there is an interactive effect between gender and southern context or level of Republican support. Neither of these hypotheses finds any support.

Similar to the findings in race cases, Model 4 shows that the percentage of motions won by the plaintiff has a significant effect on the final case outcome, once again demonstrating the important effects these motions can have on a cumulative level. As stated above, given the impact successful lawsuits against one company can have on levels of discrimination locally, the importance of influences on how wins these motions cannot be understated.

Differences between race and sex cases

It is notable that, in sex discrimination cases, neither social context nor interaction effects are observed at either motion or case levels. Thus, neither hypothesis six and seven, that female judges in conservative divisions would be more or less plaintiff friendly than their male counterparts, was supported. The lack of a locational effect deserves additional attention in the future. While it is possible that gender related issues do not vary by region, a more likely explanation is that the variation between regions is being balanced in some way. For example, conservative judges may be more likely to see their role as requiring them to “protect” women from lewd behavior in the workplace, while simultaneously being less receptive to claims of more subtle harassment or discrimination in pay, promotion, or hiring. There is no reason to believe there would be a comparable effect in race cases, wherein judges would be protective of plaintiffs for some types of allegations, but less receptive of others.

Conclusion

This research demonstrates that the process of defining employment discrimination in the courtroom is shaped by judicial and social context. While prior research already suggested that the background attributes of judges influences decision, there has been less attention to local conditions. Furthermore, insufficient attention has been paid to determining how local context interacts with judicial identity to shape case outcomes. Doing so, as my analyses have, contributes to our understanding of the law as a socially constructed phenomenon, with decisionmaking being shaped by the context within which judge's reside. This contributes to our understanding our how the definition of employment discrimination is continually redefined, particularly in ways that reflect local normative conditions.

Additionally, this research shows that social context affects judge's handling race discrimination cases differently than is does judges in gender cases. This makes initial sense, as normative standards regarding race and sex do not have to be identical in one space. Furthermore, a conservative environment regarding sex needed not be as defendant friendly as a conservative environment regarding race. Planned future research will examine race and gender cases separately, introducing more refined analysis of social contexts relevant to the type of case. For example, racial density measures would be relevant for race cases, while the percentage of women head of households might be relevant to sex discrimination cases.

My research raises additional questions for future research. The foregoing assumes that the strength of cases is consistent across regions. However, it is reasonable to ask if the cases the EEOC deem worthy of filing varies by region. Government attorneys want to spend public resources on winning cases. Therefore, if EEOC attorneys are aware that race cases in the south need to be stronger to win, they may only file strong cases. Objectively, the cases filed in the south may be stronger than those filed in other regions. Therefore, a southern effect may exist, but its influence is exerted at the case filing level. The presence of the southern and judge race interaction effect lends some support to this speculation, as the EEOC cannot know the judge's race before a case is filed, but they can know the region. If the south as a whole were less plaintiff friendly in race cases, owing to the White judges alone, we would see no effect of the southern variable. However, the judge race and southern interaction could indicate that Black judges are reacting to the strong races cases filed in the south, while biased White judges in the south do not see the same cases as strong.

Ultimately, an in-depth qualitative study of the inner workings of various EEOC offices could lend insight into this question. Beyond the matter of case strength in the south versus outside, regional EEOC offices have different leaders, who may otherwise utilize different methods for determining which cases to file. For example, offices may prioritize particular types of allegations or may perceive greater support based on different types of evidence presented. Quantitative study alone would be insufficient to sort out the complicated decisionmaking that leads from a complaint being filed with the EEOC to the EEOC filing a lawsuit in federal court.

Future research should also examine whether unequal application of federal law across regions results in different levels of racial and gender inequality across those regions. By doing so, researchers will help identify why inequality and discrimination in employment still exist in varied levels throughout the country. The effect of local social context is especially important in the context of EEOC law, as Congress was attempting to remedy local conditions via federal laws. If the law's power to enact social change and reduce racial and gender inequality in the workplace is dependent on plaintiffs winning cases, then inconsistent application of the law means inconsistent change.

It should be noted that absent qualitative analysis, it is difficult to discern if the effect measured is one of more liberal v conservative constructions of anti-discrimination law or if it is instead a generalized bias in favor of or against the parties. Furthermore, as has been noted above, the computational power of these models is severely limited by the underrepresentation of people of color on the federal bench.