Abstract
The authors examine intersectional earnings inequalities in U.S. state and local government workplaces during the Great Recession of 2007 to 2011. Corresponding to closure and exploitation mechanisms as proposed in Relational Inequality Theory, the authors decompose pay gaps into between-workplace and within-workplace segregation components and within-job disparities. Between-workplace closure mechanisms tend to be absent or weak for all comparisons, but within-workplace occupational closure and within-job pay disparities are present for all and quite large for most groups. Within-job earnings inequalities tend to be largest for Black, Hispanic, and Native American women and smallest for Asian and Native American men. During the Great Recession, organizational resources to make claims on shrank, as low-wage job layoffs surged and resources contracted. This resulted in a shrinking of within-workplace and within-job, but rising between-workplace, inequalities.
The United States needs to confront racial injustices in education, policing, neighborhoods and workplaces. Challenging gendered violence, supporting care work, and achieving gender equity at work remain pressing social goals. We focus on one key aspect of these larger inequalities: employment pay disparities associated with race/ethnicity and gender. Our primary contribution is to identify where in employment processes workplace inequalities are generated. We ask, To what extent are racial and gender inequalities a function of closure processes that sort groups into high and low wage workplaces and jobs? Are there processes which further generate race and gender inequalities between groups sharing the same job in the same workplace? Although political challenges to racial and gender pay disparities continue, the ability to locate the source of those disparities in real workplaces is only now being developed. We contribute to this effort.
The employment context we examine are U.S. state and local government workplaces, often described as low inequality workplaces (e.g., Mandel and Semyonov 2014, 2021). The U.S. Equal Employment Opportunity Commission (EEOC) routinely collects earnings data from state and local governments. Unlike conventional survey data, EEOC workplace data help illuminate not only the extent of disparities between groups but also whether wage inequalities are produced by hiring-based segregation between workplaces, job assignment segregation within workplaces, or disparate pay for people in the same job in shared workplaces. This decomposition of total earnings into between-workplace and between-job segregation components and within-job residual inequalities is the main goal of this study.
State sector employment has been of keen interest to social scientists. Some have concluded that state sector race and gender inequalities are lower than in the private sector, reflecting more formalized job and pay practices and stronger commitments to equal opportunity, which are argued to narrow the scope of bias in hiring and pay practices (e.g., Mandel and Semyonov 2014, 2021). Conversely, Wilson, Roscigno, and Huffman (2013, 2015) have argued that since the 1980s, racial inequalities in the state sector have converged with the private sector as goals of equal opportunity were displaced by an embrace of increased managerial discretion in the labor process.
Relational Inequality Theory (RIT; Tomaskovic-Devey and Avent-Holt 2019) provides the theoretical framework for this study. RIT is a general sociological model for understanding inequalities as products of organizational practices. RIT describes organizations as inequality regimes in which closure and exploitation mechanisms install categorical inequalities (Acker 2006; Tilly 1998). These mechanisms are activated within a process of relational claims making over organizational resources (Sauer et al. 2021). The closure and exploitation mechanisms in RIT are observable in terms of inequality outcomes, a clear advantage over more commonly theorized mechanisms, such as “the market,” individual productivity, or discrimination. In RIT, the claims-making mechanism is the interactional process through which closure and exploitation are instituted. In this way, claims making entails any interactional resource actors may have at their disposal, including market demand for skill, individual productivity, or status linked bias and discrimination processes.
We analyze three years (2007, 2009, 2011) of EEOC collected EEO-4 reports covering the period of the Great Recession. Although we cannot comment on long-term trends, we have an ideal window to observe earnings dynamics during a period of severe contraction in organizational resources in state and local government. As claims making over organizational resources is the key distributional process in RIT, this is an optimal period to examine what happens in a moment when resources shrink dramatically.
This study makes three major contributions. The first is to focus on organizational mechanisms, as a demand-side complement to supply-side human capital models. In doing so we offer an operationalization of RIT’s closure and exploitation mechanisms that is consistent with past research on workplace segregation and within-job pay inequality mechanisms (e.g., King et al. 2023; Petersen and Morgan 1995; Smith-Doerr et al. 2019). Second, we advance the notion that there is a dynamic opportunity structure for claims making over organizational resources, and when that structure changes so do the closure and exploitation mechanisms that install organizational inequalities. Finally, for 10 distinct race-gender groups, we produce new insights into the state and local government mechanisms that create group specific inequalities as well as an initial exploration of contextual variation within state and local governments in race-gender hierarchies.
Theoretical Framing
RIT is a synthetic explanatory theory, combing insights from past scholarship on organizational, categorical, and intersectional inequalities (Acker 2006; Collins 2002; Glenn 1992; Tilly 1998). Identifying closure and exploitation as generic inequality installing mechanisms, RIT theorizes the interactional processes within organizations that reflect and enact intersectional categorical status distinctions (Tomaskovic-Devey and Avent-Holt 2019). Closure refers to including or excluding categories of actors from organizational resources, such as hiring or access to desirable jobs, and are observed here as the between-organization (hiring) and within-organization (job assignment) impact of segregation on earnings. Exploitation refers to transferring resources between actors or jobs, observed, albeit imperfectly, in this article as earnings disparities between groups sharing the same job in the same workplace.
RIT describes the relational processes through which closure and exploitation are enacted as claims making over organizational resources. High-status actors’ claims are more likely to be treated as legitimate by gatekeepers and peers. Conversely, the claims on organizational resources from the uncredentialed, racialized minorities, women, and new employees are less likely to be successful. Claims can be active, passive, silenced, or institutionalized. Thus, a claim can be applying for a job (active), being offered a job one did not apply for (passive), not applying for a job for fear of rejection or discrimination (silenced), or embedded in the wage schedule of gendered or racialized jobs themselves (institutionalized). In this study, the claims-making mechanism is not observed but is theorized to be the key underlying process producing group differences in earnings. The acceptance or rejection of a claim can reflect active categorical discrimination, but it often reflects subtle status-inflected interactions, implicit biases, in-group favoritism, self-censoring on the basis of identities and experiences, and the intersection of categories with other salient status characteristics (e.g., education). We also assume that the final adjudicator of most claims are workplace managers, thus they are the primary causal agent in decisions around hiring, job sorting, and earnings levels.
Conventional Pay Gap Research
The vast majority of prior research on pay gaps use an individual supply-side human capital framework. Typically, gross pay gaps are disaggregated into components associated with individual education and labor market experience conceptualized as sources of individual productivity, occupation conceptualized as an individual choice, and a statistical residual. The residual is interpreted as an estimate of potential employer bias as well as unmeasured individual choices and capacities (e.g., Blau and Kahn 2017). RIT criticizes the focus on individual job seekers rather than employers as the primary decision makers. Rather, RIT sees market demand, individual productivity and status traits, and occupational skill as all potential claims-making resources for actors as they interact with decision makers in employing organizations.
Lewis, Boyd, and Pathak’s (2018) research is an example of a conventional individual human capital model for state sector earnings focusing on the intersection of gender and four race categories (Asian, Black, Hispanic, and White). They found substantial earnings pay gaps relative to White men for Black and Hispanic men and women, as well as White women. Asian women had equivalent mean earnings and Asian men out-earned White men on average by a substantial 18.6 percent. All gaps shrank once adjusted for individual characteristics (age, hours worked, education, citizenship, English proficiency) and state of residence (see also Wang, Takei, and Sakamoto 2017 for the particular importance of state of residence for Asian earnings). Earnings gaps dropped further once Lewis et al. adjusted for occupational segregation, but residual gaps remained. Relative to men, all women had large residual earnings gaps, ranging from a low of 10.2 percent for White women to a high of 14.1 percent for Black women. Black men also had relatively large residual gaps at 6.4 percent, while Hispanic and Asian men had salaries nearly equivalent to those of White men after adjusting for occupation, state, and individual traits. The large gender residuals suggest that gender bias at the job level in state governments may be nearly as large as in the private sector (cf. Blau and Kahn 2017). Because they used conventional survey data, they were unable to locate the sources of these wage gaps in workplace processes, and so we cannot tell if these large total and residual gaps are produced by segregation between government agencies, job segregation within government agencies, or disparate pay within jobs in the same workplace. We produce such estimates.
It is well documented that White women as well as Black men and women are more likely to be employed in the public than private sector (Mandel and Semyonov 2021). Most prior research has concluded that gender and race pay gaps are lower in the public than the private sector as are the residual gender-race components (see Mandel and Semyonov 2021 for both current estimates and a thorough review of past evidence). Within the public sector, occupational segregation explains the majority of pay gaps relative to White men for Black men and Black women. For White women, the impact of occupational segregation and the residual “discrimination” component are nearly equivalent. Mandel and Semyonov (2014) concluded that occupational segregation and particularly hours worked are now the dominant source of the gender and racial pay gaps in the public sector. Wilson et al. (2013, 2014), however, reached a different conclusion. Focusing only on Black-White comparisons, they showed for a variety of employment outcomes that the state sector was more egalitarian in the 1980s but racial inequalities have since grown. They pointed toward increased managerial discretion and mimicry of private sector human resource practices as the mechanisms behind these rising inequalities.
Studies Comparing Pay Gaps across and within Organizations
We conceptualize observed earnings distributions as primarily a result of employer decision making over access to and the distribution of organizational resources. Employers hire, make job and promotion assignments, and have control over earnings at hire and pay raises over the course of an employee’s tenure. If high-wage employers are more likely to hire White men than other groups, then some portion of the total White male earnings advantage will be generated by a between-workplace segregation process. If employers tend to hire or promote White men into higher paying jobs, this within-workplace segregation will contribute to White male earnings advantages. Finally, if employers make further earnings distinctions within jobs that favor White men, then a third source of advantage may exist, this one paying similarly skilled workers differently.
There is only a small literature that has had access to linked employer-employee data and so was able to investigate the relative impact on inequalities of between- and within-workplace segregation and within-job disparities. The classic study by Petersen and Morgan (1995) of U.S. private sector workplaces decomposed gender pay gaps in contractual wages into their firm, occupation, and job (occupations within firm) components, finding an average within-job gender pay gap of only 3 percent. Using similar data, a slightly earlier study revealed essentially no within-job pay gap (Groshen 1991). Both studies demonstrated firm and job segregation to be large contributors to the gender pay gap and so pointed to closure mechanisms in hiring and within-firm job segregation as the primary sources of gender pay disparities. Other studies suggest that within-job disparities across the labor market may be quite a bit larger (Bayard et al. 2003; Carrington and Troske 1998). King et al. (2023) analyzed linked firm-employee data using Internal Revenue Service W-2 reports linked to U.S. Census Bureau American Community Survey occupation and human capital indicators and found within-job gender pay gaps of about 10 percent. Although most research has focused on gender, the two studies that have taken a more intersectional lens (Bayard et al. 2003; Carrington and Troske 1998) suggest that within-job pay gaps may be larger for African American men and women than for White women. Using federal sector personnel data, Smith-Doerr et al. (2019) and Brummond (2022) found much smaller gross pay gaps in federal agencies than are observed in national and private sector individual-level analyses as well as quite small within-job pay disparities.
King et al. (2023), focusing on only the gender earnings gap, found for prime-age workers’ main job that 20 percent of the education-, hours-, and age-adjusted pay gap is between firms, 30 percent is from occupational sorting within firms, and about half is within jobs in the same firm. Interestingly, they found a workplace- and job-adjusted gender pay gap almost identical to the human capital–, industry-, and occupation-adjusted individual-level estimate by Blau and Kahn (2017).
Summary of Past Research
The organization-based literature leaves us with an expectation that the between- and within-workplace segregation components of job disparities will be larger than within-job components. The supply-side individual-level pay gap literature suggests that segregation actually favors Asian men and women and that Black men and women will be particularly disadvantaged by within job bias processes. There is no past research to make similar predictions for Hispanic and Native American men and women. A partial exception is Hunt et al. (2010) who, also using the EEO-4 reports, document high levels of Native American job segregation in state and local government. That study does not make comparisons with other groups, but does document a general pattern of substantial Native American exclusion from managerial and professional occupations. The conventional individual-level literature also suggests a national status hierarchy with White men at the top, followed by Asian men, with Native American and Hispanic men, White and Asian women, and Black men in the middle, and Native, Hispanic, and Black women at the bottom (e.g., Alonso-Villar and del Río 2022).
As no prior study has taken as complete an intersectional lens to the problem of pay gaps nor decomposed pay disparities into segregation and within-job components for state and local governments the relative importance of these mechanisms remains speculative. Hypotheses 1 and 2 are based on what we think we know from individual-level and organizational-level studies.
Hypothesis 1: There will be a race/sex hierarchy in pay gaps, with White men at the top, followed by Asian men, Native American, and Hispanic men, with White and Asian women and Black men in the middle, and Black, Hispanic, and Native American women at the bottom.
Hypothesis 2: Segregation between and within workplaces will tend to be larger contributors to pay gaps than within-job disparities.
Hypothesis 1 is based primarily on the residual “discrimination” component in individual-level studies, none of which have information on between organization segregation. Prior organizational-level analyses are largely silent for other groups but suggest that Black men and women tend to be paid less in the same job than White men and women. There is very little prior research on Native American people’s earnings. Hypothesis 2 is based on the few organizational-level studies that we have reviewed, none of which focuses on state and local governments specifically.
Hypothesis 3: Asian men and women will benefit from segregation into higher paying workplaces and jobs but will experience within-job pay penalties.
Hypothesis 3 has considerable prior empirical support in the work of Kim, Sakamoto and colleagues, although their work is also based entirely on individual-level studies (Kim and Sakamoto 2010; Kim and Zhao 2014; Wang et al. 2017).
The Great Recession and the Opportunity Structure for Discrimination
RIT emphasizes the primary role of organizational resources (e.g., jobs, revenue) as the target of claims making and resulting distributional inequalities (Tomaskovic-Devey and Avent-Holt 2019). Petersen and Saporta (2004) advanced the complementary notion that there is an opportunity structure for discrimination. In their opportunity structure framework segregation mechanisms should be stronger than direct within-job discrimination. The discrimination and bias processes in Petersen and Saporta’s framework correspond to relational claims making in the RIT model, while between- and within-firm segregation are closure mechanisms at hiring and in job assignment. Petersen and Saporta made a further distinction between valuative discrimination, in which jobs typically filled by minorities or women are paid less and direct discrimination in which groups in the same job receive different earnings. In RIT, these two are both seen as instances of exploitation, the former being a transfer of income between jobs at comparable skill levels and the latter between incumbents of the same job (Tomaskovic-Devey and Avent-Holt 2019). However, RIT does not define the claims-making process narrowly in terms of discrimination under the law, but rather sees them as the product of multiple status-linked resources, expectations, and reactions in a relational claims-making process.
There is clear evidence that organizational resources, the opportunity structure for claims making, shifted during the Great Recession. Chattopadhyay and Bianchi (2021) found that racial earnings inequalities rise during recessions, with job loss the dominant mechanism. Wilson et al. (2012, 2015) described racial discrimination in the state sector as increasing in this period as well. Thus, it is likely that recession induced layoffs targeted jobs disproportionally filled by racialized minorities. In agreement, Laird (2017) documented high levels of job loss among Black state and local government workers during the Great Recession. There is also evidence that municipalities with more fiscal stress during the great recession increased outsourcing and privatization (Yunji and Warner 2016). Thus, we suspect that during the Great Recession, increased layoffs and outsourcing of low-wage jobs held by women and racialized minorities would result in increased between-workplace segregation as low-wage gendered and racialized jobs disappeared. We see in Table 1 that the Great Recession led to low-wage job losses for all groups, although these were most dramatic for Hispanic, Black, and Native American women, followed by Hispanic men, White women, and Black men.
Percentage Employment in Low-Wage (<$15), Medium-Wage ($15–$33), and High-Wage (>$33) Jobs, State and Local Governments, EEO-4 Records.
Sources: EEO-4 surveys (2007 and 2011) and author’s calculations.
This discussion leads to three predictions about change in the opportunity structure for claims making.
Hypothesis 4: During the Great Recession, between-workplace segregation components of pay gaps relative to white men will increase.
Conversely, because the opportunity structure for within-workplace segregation into low-wage jobs declined, earnings inequalities associated with within-workplace closure processes during the Great Recession should decline as well.
Hypothesis 5: During the Great Recession, within-workplace segregation components of pay gaps relative to white men will decrease.
Finally, the opportunity structure for claims making within jobs we expect will be reduced by Great Recession–induced fiscal austerity. This period was marked not only by job losses but also by the absence of revenue to cover pay raises of all types. Fiscal austerity will reduce the space for claims making over promotions and pay raises within jobs.
Hypothesis 6: During the Great Recession, within-job pay disparities relative to White men will decrease.
Finally, we note that variation in organizational context within the state sector may matter as well. RIT sees the claims-making process as a local within-organization process and so leads to an expectation of contextual variation when organizations inhabit different institutional fields. Although we do not have formal hypotheses, we examine differences in the sources and levels of inequalities in different state agency functions and in different regions of the country. Our reasoning to expect organizational heterogeneity in the basic processes outlined above is developed later in the article.
Data and Methods
EEO-4 Reports
EEO-4 reports cover the population of U.S. state and local government workplaces’ earnings levels by race, gender, and occupation. Elected officials are excluded from these reports, as are all schools and education linked administrative jobs. Thus, we have data on the population of state and local government workplaces, with those two exceptions.
These reports are collected from city, state, and county governments for multiple functions: financial administration, streets and highways, public welfare, police, fire, natural resources, hospitals and sanatoriums, health, housing, community development, corrections, utilities and transportation, sanitation and sewage, employment security, other, and small jurisdictions. A government with all 14 functions will return 15 reports, one for each function and a residual form for all other employees. Smaller governments report functions with at least 100 employees. Those with no functions having 100 or more employees, return a single aggregate report. We refer to functions within governments as agencies; this is our operationalization of workplaces.
For each agency the EEO-4 survey collects wage and employment count data for full-time employees in eight occupational groups: officials and administrators, professionals, technicians, protective services, administrative support, paraprofessionals, skilled craft, and service and maintenance. Within each occupation, within each workplace, they further collect the sex (male or female) by race/ethnic (White, Black, Hispanic, Asian, Native American) composition of eight earnings bins.
We reshape these organizational-level data to units of observation that cross-classify pay bin by occupation by sex by racial/ethnic cells. When these cells are weighted by the number of incumbents in an observational unit (e.g., the White women earning <$16,020 a year in the service and maintenance occupation in the Greenfield, Massachusetts, highway department), we create a data structure almost equivalent to individual-level data. In 2011, we observe 5,447,737 individuals, in 99,216 jobs nested in 12,402 workplaces, down from pre–Great Recession numbers of 6,154,515 individuals, in 105,432 jobs nested in 13,179 workplaces in 2007.
There is limited past research using EEO-4 pay data. Hunt, Rucker, and Kerr (2020) found gender disparities in managerial access to high pay across organizational functions, but these narrow over time and are smallest in the redistributive functions (e.g., welfare, housing, hospitals, employment security). Two additional studies (Reese 2019; Reese and Warner 2012) used state aggregate EEO-4 pay gap data. They found that gender pay gaps were smaller when states had done major gender pay equity adjustments to salaries and that these adjustments were particularly efficacious for minority women. These studies, in addition to those on Asian pay advantages’ being a function of state of residence (Kim and Zhao 2014; Wang et al. 2017), lead us to treat state as an exogenous organizational context for pay gap analyses. Thus, we make within-state pay gaps relative to White men the baseline for our pay gap decompositions into between-organization, between-job, and within-job pay disparities.
Measures
The EEOC asks employers to report data on earnings and occupation separately by gender for five racial/ethnic groups (White non-Hispanic, Black non-Hispanic, Hispanic, Asian, and Native American). Employers report these categorizations on the basis of employee self-identification.
Earnings
Earnings refer to full-time employees and are observed as eight pay bands: less than $16,020, $16,021 to $20,025, $20,026 to $25,034, $24,035 to $33,043, $33,044 to $43,055, $43,056 to $55,070, $55,071 to $69,999, and $70,000 or more. We convert the eight pay bands in the EEO-4 surveys to continuous earnings estimates. For all pay bands, except the top and bottom, we impute the midpoint of the pay band as our earnings indicator. We use American Community Survey data for full-time workers to produce imputation estimates of median earnings for the top and bottom bins for each combination of state, state versus local government employer, and occupational groups.
Binned earnings may lead to errors in pay gap estimation. Binning obscures all within bin inequalities. For low paid jobs with tight earnings distributions this will obscure internal pay inequalities that may exist. Black and Hispanic men and women are most likely to be in these jobs and White men the least likely. It is possible that we are underestimating population level pay gaps for Black and Hispanics men and women with White men for this part of the distribution. Asian men and women and White men are all more likely to be found in the top bin, and the likelihood of bias in estimates is higher for these comparisons.
Occupation
The EEOC uses eight occupational groups, comprising jobs with roughly similar skill and authority levels. Appendix B lists the more detailed occupations that are included within these occupational groups. The most common occupational groups are professionals at 27.3 percent of all jobs and protective services at 20.8 percent. The other occupational groups are smaller, ranging between officials and administrators (6 percent) and administrative support (14.5 percent).
Between 2007 and 2011, all occupations showed declining employment, but these shifts were most dramatic for administrative support (−17.5 percent), service and maintenance (−14.5 percent), and technicians (−12.6 percent). In contrast, professional (−8.6 percent), paraprofessional (−9.9 percent), official (−10. 1 percent), protective services (−10.1 percent), and skilled craft (−10.5 percent) occupations had substantial but smaller losses. There is a great deal of occupational segregation, most notably along gender lines (see Appendix C for data and detailed discussion).
Jobs
Following Petersen and Morgan (1995), we define jobs as the intersection of workplace (agency) and occupation. As our measure of occupation is less precise than job titles, we interpret our measure as capturing the set of jobs in identical functions (e.g., highways and roads) with similar skill levels (occupation groups). As a result, our within workplace analyses of job-level pay disparities averages different pay within job titles and differences in pay between job titles that share roughly the same skill level. In Petersen and Saporta’s (2004) framework these two components could be interpreted as valuative and direct discrimination. In RIT, these estimates can result from two different processes: income shifts between jobs and between coworkers with the same job title. RIT describes both as different realizations of the more general exploitation mechanism (Tomaskovic-Devey and Avent-Holt 2019). Under legal theories of discrimination measuring job groups rather than job titles introduces measurement error. Under a comparable worth theory that sees different pay for similarly skilled jobs as a form of valuative bias, there is less concern with measurement error (England 1992; Tomaskovic-Devey and Hong 2024).
Tomaskovic-Devey and Hong (2024), using federal government personnel data, find within workplaces 80 percent of EEOC job groups include three of fewer job titles, suggesting that job-title distinctions within job groups are typically not numerous. That same study, however, showed that estimates of within-job pay gaps relative to White men were lower when job titles, rather than EEOC job groups, were used. The magnitudes of these estimate differences averaged 2 percent for men of color and 4 percent for women of color and White women. Under a legal theory of bias this strongly suggests that within–job group pay gaps are overestimates relative to job titles but also that the magnitude of the bias may be relatively small. We discuss our estimates in the discussion section taking this finding into account.
Unobserved factors
Mandel and Semyonov (2014) identified differences in hours of work as particularly consequential for public sector pay gaps. Because the EEO-4 collects pay data only for full-time workers, this source of omitted variable bias is controlled by design.
As we do not observe any individual-level characteristics beyond full-time employment and race/gender, our estimates of pay gaps may be too high (or low) if at the aggregate level a particular group averages lower (or higher) tenure or other job relevant characteristics within jobs. It is difficult to know the extent of this problem, as most past research focuses on between-person, rather than within-job, estimates. In an exception focused on within-job gender pay gaps, Penner et al. (2023) reported for the entire U.S. labor force an average within-job earnings gap of 14.1 percent in a model that controls for individual human capital and a 14.7 percent gap in a model that omits these individual covariates (compare Table 1 with Supplemental Table 7), or only a .006 difference. We believe that omitting human capital variables is unlikely to lead to large shifts in estimates or inferences.
Estimation Strategy
Consistent with past practice our analyses focus on a nested set of ordinary least squares regression estimates (Penner et al. 2023; Petersen and Morgan 1995; Smith-Doerr et al. 2019). The first model reports the gross earnings gap relative to White men. In subsequent models we produce pay gap estimates within state (model 2), within workplace (agency) (model 3), and within job (i.e., occupation-workplace unit; model 4). All models are estimated only on integrated units. Comparing the results of these four models enables us to see the degree to which gross gender differences in pay are explained by difference in local pay rates (states) and sorting across agencies and jobs within states. We estimate these models separately for 2007 and 2011.
The equations all follow the same general form, using four different specifications:
and
where subscripts represent i for each status category, s for state, w for workplace, and j for job (occupation-workplace cells). The dependent variable is yearly full-time earnings (Earnings ji ) for each job for each racial/ethnic-sex group. Regressions are unweighted, reflecting a job-cell unit of analysis.
The independent variables of interest are a series of dummy variables for nine status groups (i.e., White women, Black men, Black women, Hispanic men, Hispanic women, Asian men, Asian women, Native American men, and Native American women, with White men as the reference category). The coefficients βXi1–i9 are estimated earnings gaps relative to White men for each racial/ethnic/sex group. Models 2 to 4 produce estimates of βi1–i9 after fixed-effects adjustments for state, workplace, and job, respectively. Each model has a residual error term, ε j . Standard errors for all models are clustered at the workplace level.
These regression models yield estimates of the within context (state, workplace, job) pay gaps relative to White men. We see these decompositions as producing population estimates of the average causal effect of claims making over organizational resources as adjudicated by employers. Clearly, at the workplace level, the causal complexities are quite large, and we expect substantial workplace variation in status hierarchies, claims making, and employer decision making. We explore two potential contextual sources of such variation in a later section.
In our interpretation of estimates we follow the methodological framework introduced by Lundberg, Johnson, and Stewart (2021). They advised researchers to first choose one or more theoretical estimands and make clear the relationship to a general theory. Next, choose an empirical estimand that is derived from the theoretical estimand. Finally, choose an estimation strategy to learn that empirical estimand from the data.
We have five theoretical estimands for nine sex-race groups, each of which compares earnings with White men. The first is the total effect of the claims-making process on organizational resources. The second is the share of that total effect produced by between firm closure processes. The third is the share of the total effect produced by within-workplace job closure. Both closure processes we see as primarily about employer hiring and job allocation decisions in response to employee claims making. These claims can be active, as in applying for a job; passive, as in being offered an unsolicited job; or silenced, as in avoiding or being unaware of particular jobs and employers. These closure processes are often referred to in the literature as sorting or segregation processes, the former between workplaces and the latter within. The fourth is the residual inequality within job groups. This one lacks a singular tie to a theorized mechanism. Rather we understand it as produced by three mechanisms. One mechanism is further closure at the job-title level, which would occur if similarly skilled race-sex groups tend to be allocated by employers into lower paid job titles (Huffman and Cohen 2004). We also theorize two distinct exploitation mechanisms. The first of which is the within-workplace transfer of earnings between similarly skilled jobs on the basis of the typical sex-race of incumbents, often referred to in the literature in terms of the devaluation of typically women or minority jobs (England 1992). The second is the transfer of income between individuals who do the same work in the same job, sometimes referred to in the literature as within-job discrimination (Petersen and Morgan 1995).
Our five theoretical estimands correspond to five empirical estimates that we will learn from the data. These empirical estimands are the descriptive earnings gaps between each categorical group and White men, estimated as described in equations 1 to 4.
The first empirical estimand (E1; βXi1–i9 in equation 2) is the total earnings gap relative to White men conditional only upon state. We treat sex-race categories as exogenous to employment, and this quantity is the population average consequence of not being a White man. The most interesting information in equation 1 is the size of the average earnings gap and its variation across sex-race groups, which we see as reflecting total differences in claims-making access to organizational resources relative to White men. We condition on state because of the relatively strong association between racial groups and residential location. E1 estimates are used to evaluate hypothesis 1.
The second empirical estimand (E2), again for all sex-race groups, is the share of the E1 total effect that is produced by closure processes that sort people by sex-race category across workplaces. This quantity reflects differential access to higher and lower wage workplaces and is observed as the differences between βXi1–i9 from equation 2 from the parallel coefficients in equation 3 divided by βXi1–i9 from equation 2.
The third empirical estimand (E3) is the share of E1 produced by the sorting of people by sex-race categories into distinct job groups within workplaces. This quantity is observed as the differences between βXi1–i9 from equation 4 from the parallel coefficients in equation 3, again divided by βXi1–i9 from equation 2. Because we have job groups, rather than job titles and controls for within-job experience, E3 is likely to be underestimated under the legal theory that discrimination occurs only within narrowly defined job titles.
The fourth empirical estimand (E4) is the share of E1 that remains within jobs, calculated as the ratio of equation 4 βXi1–i9 coefficients to equation 2 βXi1–i9 coefficients. For the same reasons that E3 is likely to be an underestimate under a legal conceptualization, E4 is likely to be an overestimate to the extent any particular group is further segregated from White men at the job titles or differs at the population level in human capital composition. E2 to E4 are used to evaluate hypothesis 3.
We are also interested in a fifth empirical estimand (E5), the average level of pay gaps within job groups (e.g., equation 4, βXi1–i9). E5 is useful in two ways. First, because it is estimated for people in the same workplace and job group, we can assume that most of the unobserved covariates with race-sex category (e.g., education, age, experience, neighborhood) are effectively controlled as are any workplace level closure process (e.g., employer taste for discrimination or sex-race group reluctance to apply for jobs at particular workplaces). Thus, comparing the levels of these coefficients across race-sex groups produces a narrow version of E1, responding to the expectation in hypothesis 1 of the relative ranking of each group, but now net of skill levels, employer taste for discrimination, labor supply to employers, and job group sorting. Second, one can also interpret these levels as upper limit estimates of the degree of exploitation in the average employment relationship. We do so cautiously, as these estimates remain vulnerable to omitted variable bias (e.g., education, age) and misattribution to within-job inequality between job closure produced by fine-grained job-title distinctions within job groups. Prior literature leads us to believe that the former is quite small (Penner et al. 2023), but the latter may be much larger (Tomaskovic-Devey and Hong 2024). We return to these issues in the discussion.
In Lundberg et al.’s (2021) framework, estimand choice involves being explicit about theory-estimation linkages and population. In the RIT framework, the importance of categorical hierarchies, claims-making processes and causal and exploitation mechanisms are treated as universal, but the actual claims-making resources of categories and the presence, absence, and magnitude of closure and exploitation are deeply local. Thus, under an RIT theoretical model, empirical generalizations from these analyses are limited precisely to average effects in U.S. state and local governments from 2007 to 2011.
Results
Table 2 reports the results of the basic regression models for 2007 and 2011. In 2007, Asian men and women have gross pay advantages over White men of 23.7 percent and 10.2 percent, respectively. Echoing Wang et al. (2017), model 2 shows that all of the Asian advantage is a function of state of residence. Within states, Asian men earn 1.3 percent and Asian women 15.4 percent less than White men. This pattern of pay gaps being larger within states holds, if less dramatically, for Hispanic and Native American people as well. These patterns are essentially the same in 2011. Focusing on within state estimates (E1, model 2), White men have significant wage advantages over all other groups, and Native American, Black, and Hispanic women have by far the lowest relative earnings. This same pattern is repeated within agencies (model 3) and within jobs (model 4).
Ordinary Least Squares Regression Models of Group Earnings Gaps Relative to White Men, Full-Time Employees, State and Local Government Agencies.
Note: Estimates represent model based comparisons with White men with sequential fixed effects for state, occupation, workplace (agency or function), and job (the cross-classification of occupation and agency or function). Coefficients reflect average within-context pay disparities across all state and local government workplaces. The gross and state models are estimated for all agencies. The workplace and job models are estimated only for workplaces and jobs that include both White men and the row status group. All estimates are statistically significant below a .05 probability, unless in italic type.
On the basis of the estimates from models 2, 3, and 4, we decompose the within-state pay gaps into components associated with between-workplace segregation, within-workplace job group segregation, and within–job group pay disparities. We see the two segregation components as different dimensions of the closure process and within-job pay gaps as indicators of finer job title closure and exploitation, defined broadly to include differential pay in the same job and differential pay for similarly skilled jobs in the same workplace. Figure 1 presents this decomposition for 2007. Groups are arrayed within gender from highest to lowest total pay gap relative to White men, patterns are similar for 2011.
Dollar deficits (2007) relative to White men associated with workplace and job segregation and within-job disparities.
Every group experiences a substantial within-job pay penalty relative to White men coworkers, and consistent with hypothesis 1, the total gaps are smallest for Asian and Native American men and largest for Hispanic, Black, and Native American women. All other groups are in intermediate positions, also consistent with hypothesis 1.
Between-workplace segregation components are quite small. In fact, for Hispanic, Black, and especially Asian workers, between-workplace segregation actually narrows pay gaps with White men. For example, Hispanic women’s average $21,844 pay deficit is slightly reduced by between-workplace segregation (−$949) but generated by high penalties associated with within-workplace job segregation ($10,542) and within-job pay disparities ($12,291). Black and Native American women share a similar pattern of small between-workplace components and roughly equivalent and very high within-workplace segregation and within-job pay deficits.
All groups, except Asian men, experience strong within-workplace job segregation penalties. White and Asian women’s pay gaps are produced by both within-workplace segregation and within-job pay disparities, with the latter being greater. For Black, Hispanic, and Native American men, pay penalties associated with within-workplace job segregation are larger than within–job group pay deficits. Asian men stand out for suffering pay disparities only within jobs but only small ($370) job group segregation–generated wage deficits relative to White men.
In Figure 2, we see further support for hypothesis 1, which predicted a race/sex hierarchy with White men at the top, followed closely by Asian and Native American men, White and Asian women and Hispanic and Black men in the middle, and Black, Hispanic, and Native American women at the bottom. Because Figure 2 is within job, it is least likely to suffer from omitted variable bias. Figure 2 is arrayed from the largest to smallest within–job group pay gap relative to White men. The first finding is that all women have larger within-job pay gaps than all men. Among women, the rank order from higher to lower gap is Hispanic, Black, Asian, Native American, and then White women. Among men, the within–job group gaps relative to White men are smaller, and the rank order for average gaps is the same as for women: Hispanic, Black, Asian, and Native American. The racial hierarchy in these state sector organizations displays smaller penalties for Native American men and higher ones for Asian women than in past research using individual survey data. They also suggest that the average gender hierarchy within jobs take precedence over the race hierarchy.
Within-job dollar deficits (2007 and 2011) relative to White men.
Hypothesis 2, that segregation between and within organizations will tend to be the larger contributors to pay gaps relative to White men, is supported only among men. For all minority men, it is job, rather than between-workplace sorting, that is most influential. Job segregation within workplaces is a major source of pay gaps relative to White men for all groups, except Asian men. All women have larger within-job pay gaps relative to White men than gaps produced by segregation mechanisms. Thus, hypothesis 2 is supported only for men of color. For all women, pay gaps within job groups are larger than those between job groups.
Hypothesis 3, that Asian men and women will benefit from segregation into higher paying organizations and jobs but will experience within-job pay penalties, is also only partially supported. Both Asian men and women benefit from workplace sorting and at comparable magnitudes (see Figure 1). In contrast, there is no net benefit for Asian men from occupational sorting within workplaces relative to White men. Asian women experience pay penalties relative to White men from within-workplace segregation. Although the penalty is about half as large as it is for other minority women, it is considerably larger than the job segregation component for White women.
The Great Recession
Table 3 reports 2007 to 2011 changes in the total pay gaps relative to White men and the contributions of between-workplace and within-workplace segregation and within-job disparities to these changes. Groups are arrayed from the largest widening of the pay gap (Black men) to the largest narrowing (Native American women). Total pay gaps declined for all groups, with the exception of small growth for Black men ($410) and Hispanic women ($200). Black women (−$127) and Asian men (−$164) experienced small declines in total pay gaps. White women (−$467) and Native American men (−$740) experienced somewhat larger pay convergence with White men, but the big gains during the recession period went to Hispanic men (−$1,053), Asian women (−$1,188), and Native American women (−$1,404).
Dollar Change in Earnings Gap Relative to White Men, 2007 to 2011, by Mechanism.
Between-workplace segregation produced rising earnings gaps relative to White men for all groups, except for Native American women. Whereas for Hispanic men and White women, the between-workplace contributions to changing inequalities were relatively small, for the other groups, they were quite large, in the range of $750 to $2,000 lower earnings. This result is strongly consistent with hypothesis 4, which predicted rising between-workplace inequality because of the elimination of low-wage jobs more likely to be occupied by women and minorities.
In contrast, most groups saw declines in the earnings consequences of within-workplace job segregation, most dramatically for Asian women (−$1,136), followed by Native American men (−$789), Asian men (−$474), Hispanic men (−$427), and Black women (−$312). In contrast, workplace segregation between high- and low-paying jobs rose for Hispanic women ($683) and marginally for White women ($120). These results are largely consistent with hypothesis 5, which predicted declining within-workplace segregation, also because of the disproportionate elimination of lower wage jobs.
Consistent with hypothesis 6, every group saw declining within-job pay gaps relative to White men. This was most striking for Asian women ($2,012) followed by Hispanic ($1,138), White ($1,028), and Black ($966) women. Black men saw the smallest within-job progress ($661), but all men saw convergences with White men in within-job pay gaps. Native American women’s gains ($842) were relatively small and more comparable with minority men than to other women.
We see these estimates as indicators of a declining opportunity structure for pay raises and promotions, suggesting that discrimination and more subtle bias processes declined during the Great Recession for all groups, but most dramatically for most women. But as this happened primarily because of the absence of money to pay for raises and promotions during the Great Recession, it is probably more appropriate to interpret this as declining managerial opportunities to exhibit pay, hiring, and promotion preferences for White men.
Contextual Heterogeneity
Of course, these estimates are national averages (net of state of residence). If the status ranking of groups reflects a general national race-gender hierarchy, we would expect that the basic status rankings would be constant across contexts. The idea that the institutionalization of status hierarchies is a national phenomenon is at least partly at odds with both intersectional and RIT accounts. In RIT, the claims-making process is relational and embedded in local institutional contexts. Similarly, in most intersectional accounts, the institutional and interactional valence of status characteristics is spatially and historically produced (Glenn 1992; Misra, Curington, and Green 2021). Thus, when organizations have different local cultures and practices we should expect different local status hierarchies. Past research on state and local governments has identified redistributive government functions, which include social services of various types, as producing lower gender job segregation than other government functions (Hunt et al. 2020). The interpretation has been that redistributive functions tend to be both women dominated and more culturally committed to equal opportunity. In contrast, distributive functions (highway, natural resource, and community development) have specialized professional cultures, with no natural affinity with equal opportunity goals. Regulatory functions (police, fire, corrections, and utilities/transportation) have strongly hierarchical male cultures and may even be actively hostile to such goals.
Intersectional analyses have always stressed the place based historical origins and continuities of race and gender hierarchies (Collins 2002; Glenn 1992). Consistently, Chattopadhyay and Bianchi (2021) suggested that bias processes are geographically variable. Stainback and Tomaskovic-Devey 2012 reported on regional differences in Black men and women’s access to better quality jobs and found somewhat better opportunities relative to White men in the West and worse in New England.
These expectations of status hierarchy heterogeneity led us to look at within-job pay gaps for different government functions and regions of the country. Here we focus on within-job gaps, as they are the best comparison of otherwise equivalent workers available in these data. It seems most likely that it will be for cross-cutting status characteristics that we find organizational and geographic variation in within job status hierarchies, but for strongly institutionalized distinctions (e.g., White men, Black and Hispanic women), advantages and disadvantages may be ubiquitous across contexts.
Table D1 in Appendix D makes the function comparisons. Hispanic, Black, and Native American women experience large within-job pay disparities relative to White men in all three government functions, both before and after the Great Recession. Similarly, White men’s within-job earnings are higher than all other groups. However, the gaps with Asian men are quite small in redistributive and distributive functions, not even reaching statistical significance in 2007. They are similarly small for Native American men in redistributive and regulatory functions.
We do see some switching of status orders among cross-cutting race-gender categories. In the redistributive function, Black and Hispanic men’s relative earnings are much lower than in the other functions, almost as low as Hispanic, Black and Native American women’s earnings. These results contradict expectation of lower inequalities in redistributive functions on the basis of Hunt et al. (2020). This suggests to us that in the redistributive function, it may not be that biases against women are lower, but that biases toward minority men are higher, thus reducing observed gender inequalities.
The other notable deviation from the national pattern is in the regulatory function, where Asian men suffer large within-job pay penalties, roughly equivalent to White women. In regulatory functions Asian women’s pay deficit is depressed to the level of other minority women. In the regulatory function, Asian disadvantages are quite high, moving Asian men into the same rank as White women and Asian women into the bottom class. Branch (2011) concluded that in the United States, Black women are consistently the “bottom class.” Our results suggest that in state and local governments they are often joined by Hispanic women and in regulatory functions, such as policing, by all women of color.
Regional comparisons (Table D2) are less dramatic. The basic status hierarchies are quite similar across regions. One notable exception is for Native American men and women in the Northeast, where their disadvantages relative to White men are considerably larger. Native American men, who in other regions have nearly equivalent within-job earnings to White men, have a 12 percent deficit in the Northeast in 2007, closest to what is experienced by Black women. This deficit is largely gone by 2011. Native American women, who in other regions tend to be closest to Black women and better off than Hispanic women, are the most disadvantaged in the Northeast, with a 17 percent in 2007 and 15 percent in 2011 within-job pay deficit relative to White men.
The other notable regional pattern is that Asians, both men and women, in the South have no or small within job group deficits relative to White men. In the South in 2011, Asian men actually out-earned on average White men in the same job by 2 percent. As this was not statistically significant, we might say that Asian American and White men have equivalent claims-making status in the South in state and local governments. In the South, Asian American women’s pay deficits relative to White men are smaller than those of White women in both years and shrink by 2011 to a nonsignificant 3 percent.
These heterogeneity analyses confirm that the basic status hierarchy, with Black, Hispanic, and Native American women at the bottom and White men at the top, is consistent across functions and regions. Cross-cutting demographic intersections are more fluid across contexts, presumably responsive to local cultural histories. This is particularly the case for Asians, who in some contexts are indistinguishable from same-sex Whites and in others approach the position of other racialized minorities. Finally, both function and regional temporal comparisons confirm that the declining within-job pay gaps during the Great Recession that we found in national estimates, are replicated within each of the functions and regions examined.
Discussion
Using workplace-level data on state and local government employment, we estimate pay gaps relative to White men for nine race-sex groups and decompose them into between- and within-workplace segregation components and within–job group pay gaps. After accounting for state differences in pay rates, all groups earn less on average than White men, although gaps are smallest for Asian and Native American men. Black, Hispanic, and Native American women have the largest gross pay deficits relative to White men. This is one of the very few studies, and the first with workplace data, to examine the position of Native American men and women. We replicate survey-based comparisons, which reveal small pay gaps for Native American men relative to White men and much larger deficits for Native American women.
These patterns are largely replicated in comparisons across agency function and region. The one prominent exception is that in the U.S. South, Asian Americans in the public sector actually tend to out-earn same-sex Whites when in the same job, although the differences are not statistically significant. If one was going to assert that Asian Americans could become honorary Whites, these data would suggest that this is only true in the South in certain circumstances. In contrast, in regulatory agencies, such as police, fire, and jails, Asian Americans are as or more subordinated within jobs as their same-sex Hispanic and Black coworkers. In regulatory agencies, such as in police work, Asian Americans appear closer to forever foreign than to honorary Whites (see Tuan 1998 for a discussion of this tension in the U.S. racialization of Asians).
Between-workplace segregation generally had small effects on pay gaps and in some cases actually reduces the pay deficit relative to White men. In contrast, within-workplace segregation is a strong source of total pay gaps for all groups, with the exception of Asian men who tend to work in higher paying jobs than White men. Strikingly, all groups experience within-job pay penalties relative to their White male coworkers. Within-workplace segregation components are larger than within-job disparities for Black, Hispanic, and Native American men, while the within-job components dominate for Asian women, White women, and especially Asian men. Black, Hispanic, and Native American women endure similarly high pay gaps generated by both job segregation and within-job deficits. To translate our findings into the theoretical language of RIT, the strongest evidence for closure-generated inequalities in state and local government happens via job segregation within workplaces. Closure in access to high-wage agencies is not a dominant mechanism in U.S. state and local governments.
Within-job pay gaps relative to white men are quite large for all groups, suggesting that the exploitation mechanism is relatively widespread. It is possible that this reflects both favoritism toward White men in pay and promotion decisions and gendered and raced valuation of more finely segregated job titles within similarly skilled occupational groups. Theoretically we think that our observed within–job group pay gaps are produced by both skill linked job title closure as well as exploitation mechanisms associated with both the devaluation of women and minority jobs and the devaluation of individual minority and women employees within jobs shared with White men. The former two would be associated with job title segregation within job groups. Tomaskovic-Devey and Hong (2024) found that using job groups instead of job titles for federal employees led to inflated with job wage gaps of 4 percent for all women and 2 percent for minority men. If we adjust our 2011 within-job pay gaps downward on the basis of these estimates, then our conclusions would shift somewhat, suggesting larger between-job segregation and lower within-job inequality components for all groups and particularly for all women.
Within-job pay gaps are not a constant across groups. They are typically quite large for Black, Hispanic, and Native American women; intermediate for Black and Hispanic men and Asian and White women; and fairly small for Native American and Asian men. Within-job pay inequalities can also be context dependent: absent in some contexts for Native American and Asian men, exaggerated for Asian men and women in regulatory functions, exaggerated for Hispanic and Black men in redistributive functions, and larger for Native American men and women in the Northeast.
Examining the degree to which pay gaps shifted and earnings mechanisms changed during the period of organizational resource contraction during the Great Recession, the general pattern for within-workplace inequalities is one of declining pay gaps relative to White men. Layoff patterns during the Great Recession added to White men’s between workplace advantaged access to high-wage agencies. In contrast, during the Great Recession, within-workplace job assignments and within-job pay shifts tended to erode White men’s earnings advantage inside their workplaces. We see this package—increased between-workplace inequalities and reduced within-workplace inequalities—as reflecting a change in the opportunity structure for claims making. Government agencies tended to protect jobs held by White men, leading to increased between-workplace inequalities. But at the same time, the elimination of low-wage jobs reduced within-workplace inequalities in access to higher wage jobs. Austerity prevented agencies from offering raises, and so the normal within-job process that advantaged White men was weakened considerably, and within-job inequalities declined.
There are, of course, limitations to these analyses and so caveats to any conclusions. The low top pay band probably conceals additional pay variation. To the extent that White and Asian men are more likely to be in that pay band we may be misestimating their relative earnings. If we had finer detail in occupational coding, distinguishing, for example, between police officers and police sergeants, we would estimate a larger withing workplace segregation component and a smaller within-job pay gap. Nonetheless we see these estimates as meaningful, representing a combination of within-job disparities and pay disparities between similarly skilled people in the same organization.
One weakness of these data are that we cannot adjust for group differences in education, tenure, and experience. Prior research has shown that individual traits have only small impacts on earnings within jobs (Penner et al. 2023). We do not think that there are likely to be large differences in education levels by race and gender in the same job groups in the same workplaces, but it is possible that tenure differences associated with the age structure of groups may influence our results. This source of omitted variable bias would be most likely to influence estimates for the Hispanic populations as they are marginally younger on average than other groups (see further discussion in Appendix A).
Conclusion
Our analysis of organizational data permitted a decomposition of pay gaps into between- and within-workplace segregation, as well as within–job group pay gaps. Conventional estimates have emphasized the importance of segregation, a result we confirm, but distinguishing between types of segregation allows us to identify the much more important role of within-workplace segregation for the production of inequalities in the U.S. state and local public sector. This is in contrast to general population (Bayard et al. 2003; King et al. (2023) and private sector (Ferguson and Koning 2018; Petersen and Morgan 1995) estimates, which point to a larger role of between-workplace segregation. State and local governments appear to be more equal opportunity at the hiring stage than the private sector.
Although we find between-workplace segregation for state and local governments to be a weak and sometimes even offsetting source of pay inequities in cross-section, it was a strong contributor to rising pay inequality during the Great Recession. This more dynamic finding is reminiscent of Ferguson and Koning (2018), who found for private sector firms rising between-firm racial segregation. It is different, however, in that they document that rising private sector between-firm segregation resulted from the death of more integrated and birth of more segregated private sector workplaces. Governments are not so dynamic, rarely being founded or shuttered. Our results suggest that during the Great Recession, White men tended to suffer less from job elimination in high-wage government agencies. This is consistent with the work of Wilson et al. (2012, 2015), who argued that the public sector is marked by increased racial discrimination, and Chattopadhyay and Bianchi (2021), who saw recessions as conducive to increased layoffs targeting African Americans.
The state sector is often described as a low-inequality employment context (e.g., Mandel and Semyonov 2014, 2021). Mandel and Semyonov (2014, 2021) showed that as recently as 2013, gender and racial earnings inequalities tended to be lower in the private than public sectors. They explain this in terms of more formalized personal practices and stronger commitments to equal opportunity. In partial contrast, Wilson and Roscigno (2010, 2017), and Wilson, Roscigno, and Huffman (2013, 2015) showed that black-white inequalities in pay, firing, promotions, and authority were all lower in the public sector in the 1980s but have since converged to the level of racial inequalities in the private sector. They invoked an explanation of the copying of private sector logics in producing widen inequalities. These included the rise of precarious employment, increased managerial discretion, and the reduction of earlier equal opportunity protections. Our results may shed some light on these two narratives. It may be that state and local governments are more equal opportunity at the hiring stage, but not in their internal human resources practices. This would be consistent with the very weak effect of between-workplace segregation and the relatively strong effects of job segregation and within-job inequalities.
Mandel and Semyonov (2014) showed declining gender pay gaps in the public sector through 2010, but comparable estimates for Black-White trends are not available. It is possible that rising racial inequality in the public sector is limited to Black-White comparisons, a trend in the economy more generally (Mandel and Semyonov 2016). We see the potential for research using a longer time series of EEO-4 reports on state and local governments to clarify both the trends for different race-sex groups and the mechanisms producing them. Mandel and Semyonov are clear that the unexplained component of their models is growing over time for both the state and private sectors. That component could be being produced be segregation at the firm and job levels as well as within-job inequalities, which is only discoverable with workplace-level data similar to what we use here. Additionally, Wilson et al. referred to increased precarity, outsourcing, and internal discrimination as potential sources of rising black-white inequalities in the state sector. These might manifest in rising between-workplace inequality and declining within-workplace job segregation effects as the public sector outsourced low-wage jobs, in a pattern similar to what we observe during the Great Recession. Rising within-workplace discrimination, however, would be observed in data similar to ours as rising within-job inequalities. We do not think that the Great Recession pattern of resource contraction and declining within-job inequalities is generalizable to their much longer temporal trend. Clearly much more could be done with these data to understand what and how inequalities are produced in the public sector.
Although we cannot directly compare these state and local government estimates with either the federal government or the private sector with these data, it is clear that despite its reputation as more equitable in the contemporary era, race and gender pay inequities exist in the state and local government public sector. We can, however, make some indirect comparisons. Bayard et al. (2003) made race-gender comparisons for all jobs circa 1990 and found residual pay gaps comparable with those we find for Black men and women and White women. A recent study using the same decomposition approach as ours, but limited to gender comparisons and for the general population of workers but with more precise job measures, demonstrated yearly earnings pay gaps between men and women to be strongly tied to between- and within-workplace segregation and within-job pay gaps of about 11 percent, about the same size as our result for White women (King et al. 2023). These two comparisons suggest that pay disparities within state and local government agencies are of comparable magnitude as those in the labor force overall. Conversely, Brummond (2022), using high-quality personnel records, found very small between-workplace and within-job gender pay gaps in federal government agencies and a dominant role of job title segregation to explain a much smaller baseline inequality. It is quite possible that past research describing the state sector as low inequality was based on averaging low-inequality federal workplaces with high-inequality state and local governments. This possibility could be examined fairly easily by adopting Mandel and Semyonov’s approach to private-public comparisons but extending it to private-federal-state-local comparisons, something that is possible with census and American Community Survey data.
Potentially good news from our analysis is that within-workplace segregation and job-level pay disparities declined for most groups during the Great Recession. We suspect, however, that the Great Recession did not reduce interactional bias against minorities and women or for White men but rather that the pay raises and promotions that typically propel White men higher in the workplace wage distribution were largely absent during this period. Because some of the reductions in inequality were produced by the elimination of low-wage jobs, the hardship of unemployment must be weighed against the reduction in workplace inequalities.
During the Great Recession, within-workplace occupational segregation produced rising pay inequalities for Hispanic and White women, but for all other groups, within-workplace segregation was associated with declining pay inequality. Job sorting weakened most dramatically for Asian women, followed by Native American and Asian men. There clearly are egalitarian tendencies across the study period in state and local governments, but the benefits of reduced segregation and discrimination favor Asians of any gender most strongly and African Americans the least.
These results point toward two paths toward more equality. In the first, reduce pay disparities within workplaces by increasing wages for low-wage jobs. This approach would be functionally equivalent to eliminating low-wage jobs, without the pain of job loss. In the second, eliminate pay disparities within comparably skilled jobs by eliminating individual claims making and performance pay, without the pain of no-raise austerity. It is not necessary to have another Great Recession to reduce race and gender pay disparities; the same process can be accomplished by reducing organizational inequalities in the division of labor and between similarly situated people.
A focus on average pay gaps is only a first step. There is organizational variation in pay gaps, particularly for groups with cross-cutting status characteristics. Future analyses of organizational inequalities should analyze this variation. Under what conditions do racial minority men and women get access to good-paying jobs and workplaces? What characteristics of workplaces expand and contract status-based inequalities? These are the exciting questions we can turn to next as we develop theoretical and empirical models for examining workplace inequalities.
This study makes major contributions to RIT. We offer an operationalization of closure and exploitation mechanisms that is suitable for linked employer-employee organizational data within a theoretically coherent framework. Our empirical strategy follows Petersen and Morgan (1985) and Smith-Doerr et al. (2019) in suggesting that the basic inequality installing mechanisms of closure and exploitation can be observed via pay disparities associated with segregation and dissimilar treatment of similarly situated actors.
The second contribution is to advance the notion that there is a dynamic opportunity structure for claims making. When structure changes, as it did during the Great Recession, so to do the closure and exploitation mechanisms that install organizational inequalities.
Third, we demonstrate that group level inequalities are created by distinct organizational processes. Within-job inequalities are consistently higher for women than men, Asian men and women’s apparent earnings advantages disappear when we move from a national human capital model to the actual workplace context, between-workplace inequalities are weak in the public sector jobs we examine compared with past research on the private sector, and race-gender hierarchies vary across functions and regions. Most strikingly, Asian American men and women’s workplace disadvantages relative to White men disappear in the U.S. South but are as large as those of other minorities in policing, fire, and jails.
Social scientists urgently need more and better data that link employees to their employers. Unlike many European countries, the United States does not routinely make such data available to researchers. There was an initiative to collect private sector workplace pay data by the EEOC in 2017 and 2018, but it is unclear if they will be made available to researchers or if the pay data collection will resume in the future. Some U.S. states, including California and Illinois, have recently passed legislation to collect private sector pay data. Other states interested in race and gender pay equity may follow suit. Until we have good pay data at the workplace and firm levels, strong workplace and sectoral comparisons will not be possible.
Footnotes
Appendix A: Comparison with Individual Survey Results
It is useful to compare EEO-4 organizational to conventional individual survey estimates. We calculated estimates limited to state government workplaces and compare within-job pay gap estimates from our EEO-4 2011 analysis with Lewis et al.’s (2018) net pay gap estimates for state government employees for the period from 2011 to 2015. We see their pay gaps adjusted for human capital, state, and detailed occupation as roughly comparable with our within-job pay gap estimates. The studies differ in approach: the EEOC incorporates employer information and jobs defined in terms of occupational groups, whereas Lewis et al. (2018) controlled for individual-level education, age and detailed occupation but lack workplace and job information. In addition the EEO-4 compares each race-gender group with White men in the same job in the same workplace; Lewis et al. compared mean wages by group using American Community Survey data. Lewis et al. provided no estimates for Native American men or women.
The general rank order of groups is similar with marginal exceptions. In the EEO-4 analysis it is Hispanic women who suffer the largest within-job pay gap. In Lewis et al.’s (2018) analysis it is Black women. Although the rank order between Asian women and White women is flipped in the two analyses, both produce estimates that are similar in magnitude. The EEO-4 analysis point toward Asian men as having the smallest within-job pay gap in state government workplaces, while Lewis et al. point toward Hispanic men as having the smallest “discrimination” component. But both analyses agree that they are small. The largest contrast between the EEO-4 and Lewis et al. in estimates is for Hispanic women and men, both of whom have much higher within-job pay gaps in the EEO-4 estimates. This may reflect the lower educational attainment and younger average age of this population.
Appendix B: EEOC Occupational Group Descriptions
Appendix C: Occupational Segregation
White men’s modal occupation is protective services (31.2 percent), but they are also overrepresented among officials and administrators (8.4 percent) and skilled craft (12.6 percent) positions. (We discuss only 2007 distributions, but the 2011 distributions are substantively equivalent. See Table C1.)
Black men’s modal occupation is also protective services (27.7 percent), followed by service and maintenance (23.9 percent), and skilled craft (12.3 percent). Hispanic men’s model position is protective services (32.3 percent), in which they are represented at higher rates than either White or Black men. Hispanic men are also overrepresented in service and maintenance (18.6 percent) and skilled craft (12.6 percent). Asian men are overrepresented only in professional jobs (38.8 percent). Compared with other men, Asian men are quite a bit less likely to be in protective service (16.3 percent) or skilled craft (8.7 percent) positions. Native American men’s modal job is also protective service (28.4 percent), and they are the most overrepresented among skilled craft jobs (16.2 percent vs. 6 percent overall).
White women’s modal occupational jobs are professional (35.0 percent), but they are also overrepresented among officials and administrators (6.5 percent) and administrative support (30.0 percent). Black women’s modal occupation is also professional (27.2 percent), with strong representation in administrative support (25.8 percent) and paraprofessionals (13.3 percent). Hispanic women’s modal position is in administrative support (35.1 percent). They are underrepresented in all other occupations, with the exception of paraprofessionals (11.3 percent vs. 6.5 percent overall). Asian women are by far most likely to be in professional occupations (49.1 percent), followed by administrative support (22.8 percent). Like most other women, Native American women’s modal job is professional (31.8 percent) and second most common is administrative support (28.9 percent).
Appendix D: Heterogeneity Analyses
Regional Comparisons of the Within-Job Pay Gap in Percentage Relative to White Men.
| 2007 | 2011 | |||||||
|---|---|---|---|---|---|---|---|---|
| Northeast | Midwest | South | West | Northeast | Midwest | South | West | |
| Hispanic women | .16 | .15 | .17 | .15 | .15 | .13 | .16 | .12 |
| Black women | .13 | .12 | .16 | .13 | .12 | .09 | .13 | .10 |
| Native American women | .17 | .11 | .12 | .12 | .15 | .09 | .13 | .09 |
| Asian women | .08 | .10 | .08 | .11 | .07 | .04 | .03 | .08 |
| White women | .08 | .09 | .10 | .08 | .07 | .07 | .08 | .06 |
| Black Men | .08 | .06 | .09 | .07 | .07 | .06 | .08 | .06 |
| Hispanic Men | .08 | .07 | .09 | .08 | .05 | .06 | .08 | .06 |
| Asian Men | .06 | .02 | .00 | .05 | .06 | −.02 | −.02 | .04 |
| Native American Men | .12 | .03 | .03 | .04 | .03 | .03 | .04 | .04 |
Note: Italic type denotes not statistically significant.
Acknowledgements
We acknowledge helpful close readings of the manuscript by Socius reviewers, Eiko Strader, Laurel Smith-Doerr, and Karen Brummond. Underlying data are confidential, and code available upon request.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the National Science Foundation (SES-1851349).
