Combating Inequality: The Between- and Within-Group Effects of Unionization on Earnings for People with Different Disabilities

This paper addresses whether and how unions help to dismantle workplace inequality experienced by people with different types of disabilities. Using pooled 2009-2018 CPS MORG data of 630,799 respondents covering almost a decade, we find that union membership is especially beneficial for people with disabilities compared to the larger population, as well as other status groups. Furthermore, people with the severest disabilities benefit the most from being in unionized work, increasing weekly earnings by 36% for people with self-care and independent living-related disabilities. Because union membership increases disabled workers’ weekly earnings by more than double the increase experienced by people without disabilities, it brings unionized disabled workers closer to overall average earnings with important implications for inequality. Unionized work reduces earnings inequality between disabled and non-disabled workers, but earnings boosts associated with union membership generate more pronounced inequality within groups of workers with disabilities depending on whether individuals have access to unionized employment. We find that gaps among employed unionized and non-unionized disabled workers are significantly larger than those experienced by unionized and non-unionized female, Black, and Hispanic workers.


Introduction
The plight of workers with disabilities presents an acute example of how broader policies and workplace structures maintain and reproduce inequality. Despite protections provided by the Americans with Disabilities Act (ADA) -a law intended to improve labor market outcomespeople with disabilities continue to experience particularly low employment levels (Blanck et al. 2007;Bruyère et al. 2016;Erickson, Lee and von Schrader 2014;Maroto and Pettinicchio 2014a; U.S. Bureau of Labor Statistics 2020). Workers with disabilities earn considerably less than those without disabilities, and disparities tend to be largest for people with more severe disabilities, like those reporting multiple, cognitive, or functional limitations that diminish selfcare and independent living (Brucker, Houtenville, and Lauer 2016;Jones 2011;Maroto andPettinicchio 2014b, 2015).
Although individual factors like education and training shape labor market outcomes, workplace structures like wage policies and power dynamics governing employer-employee relations tend to have larger and more lasting effects on worker outcomes (Acker 2006; Tomaskovic-Devey and Avent-Holt 2019). However, the kinds of inequalities generated by these organizational and institutional dynamics are difficult to observe. Unions are thought to mitigate some of this inequality and improve workers' economic wellbeing through collective bargaining.
Presumably, unions play an important part in improving disabled workers' earnings as much as they have other marginalized groups of workers (Farber et al. 2018). Yet, disability is underrepresented in research on unions and earnings, reflecting a broader neglect of disability in the sociology of work, inequality and stratification. 3 Through its purported universalizing effects, unionism is an important countervailing societal mechanism that challenges workplace inequality and shields workers from negative policy contexts such as right-to-work-laws (Hacker 2006;VanHeuvelen 2018). By creating a socalled "moral economy" (Swensen 1989; see also Rosenfeld 2006), unions improve the economic wellbeing of workers. They do so directly through union wage premiums and indirectly when employers in non-unionized environments increase wages to preempt unionization of their workforce (Corneo and Lucifora 1997;Lewis 1986). Unions further mitigate labor market inequality by increasing baseline wages among marginalized groups of workers, moving them closer to an overall average wage (Mishel 2012;Schmitt 2008). This makes unionized employment a particularly important mechanism for addressing earnings inequality.
Despite the benefits of unionized employment, not all workers see similar earnings boosts from union membership (Bowser 1985;McCall 2001). Part of this has to do with whether collective bargaining agreements actually emphasize efforts to "democratize" the workplace for all groups (Wright 2009). In the case of African Americans and women, union organizations were not always inclusive (Kalleberg, Reskin, and Hudson 2000;Kilbourne, England, and Beron 1994;Reid 1998;Western and Rosenfeld 2011). With disability, unions have not necessarily been pro-active in integrating people with different types of disabilities or promoting certain workplace accommodations within their mandates even when they count disabled workers as members (Lurie 2017).
Consequently, unionization has the potential to reduce between-group inequality -inequality between unionized minority and majority group members, e.g., people with different disabilities and people without disabilities. However, because all groups do not have equal Electronic copy available at: https://ssrn.com/abstract=3690958 access to unions, this often occurs at the expense of within-group inequality --inequality within or among minority groups based on union membership, e.g., unionized and non-unionized workers with disabilities. 4 This means that the more salient locus of inequality after considering the moderating effects of unionization is not between workers with and without disabilities but, rather, within disabled worker groups based on whether workers are in unionized employment or not.
This within-group inequality results in part from workers' unequal access to unions that is shaped by structural, organizational, and individual-level factors. These factors include broad deunionization trends, how firms and unions define jobs as temporary or part-time precluding union membership, and the way some members of marginalized groups are clustered in industry sectors where union strength is relatively low. That is, while some workers with disabilities benefit from unionized work, disabled workers left out of unionized employment remain among the most marginalized in the labor market.
In light of the potential varying effects of union membership on economic wellbeing, we address the following research questions: Does unionized employment provide additional earnings benefits for people with disabilities? In turn, how does unionized employment influence inequality between workers with and without disabilities? What does this mean for inequality within groups of workers with disabilities who do and do not have access to union jobs? And, finally, how do these relationships vary across people with different types of disabilities, i.e., cognitive, physical, sensory, and independent living disabilities? These questions are especially salient as we expect uneven positive effects of union membership given varying degrees of access based on group status embedded in the broader economic and political challenges unions face today.
We use data from one of the largest on-going labor market surveys in the United States, the Current Population Survey Merged Outgoing Rotation Groups (CPS-MORG), pooled over nine years (2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) with a series of interaction models to examine the relationship between disability, unionized employment, and earnings. We find that unions directly improve economic wellbeing for all workers, but people with disabilities, especially those with severe disabilities, benefit the most from union membership. Being in unionized work increased weekly earnings for workers with disabilities by 28%, almost twice the increase seen by people without disabilities, and unionized employment was associated with a 35% increase in earnings for people with independent living (IDL) limitations. We show that being in unionized work brings disabled workers' already low earnings closer to a higher overall average reducing inequality between unionized people with and without disabilities. At the same time, because unionized workers with disabilities earn considerably more than non-unionized workers with disabilities, considering union membership demonstrates more pronounced earnings gaps within disability groups compared to gaps between disabled unionized workers and non-disabled unionized workers.
Our paper highlights the importance of studying mechanisms that both reproduce and mitigate stratification where wage inequality within disability, race, and gender groups is today more salient than wage inequality between these status groups (Reskin 2003;Leicht 2008Leicht , 2016Brady and Leicht 2008;Morris and Western 1999). By focusing on disability, we are able to shine light on the power dynamics underlying structures labor unions confront when seeking to Electronic copy available at: https://ssrn.com/abstract=3690958 mitigate labor market inequalities. Namely, we address how the effects of union membership manifest across status characteristics like disability, on the one hand, and within status groups based on their access to unionized employment, on the other. This latter point should not be lost.
Unions benefit members of certain marginalized groups when they have access to union jobs.

The Mitigating Effect of Unionization on Inequality
The formalization of the worker-employer contract through the 1935 National Labor Relations Act (NLRA) resulted in increased employment security and upward mobility for many American workers through the 1970s. Unions and collective bargaining agreements (CBAs) have since come to represent one of the primary ways in which groups gain access to important resources, opportunities, and rewards (Reskin 2003). Unions seek to challenge everything from unfair organizational requirements of work to informal interactions among employees to wage setting and supervisory practices that directly impact labor market inequality.
As a "historical guarantor of good wages and benefits" (Hudson 2007, 287), unions require employers to establish set pay for certain jobs, especially in sectors with weak internal consensus about hiring, firing, and promotion -so-called weak internal labor markets (ILMs). 5 They also set rules across occupations and sectors, particularly where covered by the same CBAs, such that wages across firms consistently reduce wage gaps (Blau and Kahn 2000;Elvira and Saporta 2001). Unions help stabilize and bureaucratize decision-making surrounding promotions and wages, especially because they seek to include employees in the decision-5 Typically, well-developed ILMs are more common in occupations that require a specific or technological skill set, are core rather than peripheral occupations, and in larger firms with personnel departments (Pfeffer and Cohen 1984) -jobs in which members of historically disadvantaged groups tend to be underrepresented. The public sector is generally thought to have strong ILMs, as do professional/technical and craft occupations, while tourism/hospitality and management and sales have weak internal consensus (Baum 2006;Cullen 1978;Bridges and Villemez 1994;Tolbert 1996). making process (Cornfield 1991). This can also mitigate labor market inequality because codification (see Bridges and Villimez 1991) reduces arbitrariness in decisions affecting earnings.
Because of their "inclusive strategies" and provision of tangible resources, such as education, apprenticeships, and job training, unions also play a critical role in improving workers' positions in the labor market (Knoke and Kalleberg 1994;Knoke and Ishio 1994;Pfeffer and Cohen 1984;Rosenfeld and Kalleberg 1990). Unions are thought to reduce exposure to nonstandard work arrangements and encourage stable, well-paying jobs (see Dickens and Lang 1992). For instance, Finnigan and Hale (2018) found that union membership in states with higher union density reduced the erratic and inconsistent assignment of work hours -a situation common among low-wage service workers --and improved wages among workers regardless of scheduling standards (Alexander and Haley-Lock 2015; Halpin 2015; Henly and Lambert 2014).
Indeed, unions historically worked to prevent the rise of precarious and insecure employment (Kalleberg, Reskin, and Hudson 2000) -the kinds of jobs characterized by low and volatile earnings prevalent among people with disabilities (Maroto andPettinicchio 2014b, 2015). As a result, unions have, in varying degrees, become especially important forces for improving earnings among different status groups.

The Varying Effects of Unionization Between and Within Marginalized Groups
Labor market studies have consistently found disparities in earnings and rates of promotion across race, gender, and disability groups that span occupations and industries (Blau 2012;Blau and Kahn 2017;Brown and Misra 2003;Maroto andPettinicchio 2014a, 2014b).
Although female, Black, and Hispanic workers still tend to earn less than male and non-Hispanic Electronic copy available at: https://ssrn.com/abstract=3690958 white workers, respectively, these between-group gaps have declined over time (Leicht 2008(Leicht , 2016. Scholars have attributed this observed decline in earnings disparities between different status groups to various intrapsychic, interpersonal, societal, and organizational mechanisms, including access to unionized employment (Acker 2006;Reskin 2003;Tilly 1998). As a societal mechanism primarily instituted through CBAs, unions counter the negative effects of inequality by making power relations more egalitarian (Card, Lemieux, and Riddell 2004;Freeman 1980Freeman , 1984. When low-wage workers, including many from marginalized status groups, gain access to unionized workplaces, they tend to benefit considerably (Gautie and Schmitt 2009). Unions increase earnings for low-wage and low-skilled workers (Applebaum et al. 2003;Card, Lemieux, and Riddell 2004;Farber et al. 2018) while reducing rates of working poverty across states (Brady, Baker, and Finnigan 2013). For instance, Schmitt (2008) found that unionization increased the wages of Hispanics by 18%, and Mishel (2012) reported union premiums of 17.3% and 23.1% for Hispanic and Black men, respectively, compared to 10.9% for whites. Minority women also tend to benefit from higher union premiums compared to white women (Wunnava and Peled 1999). 6 By codifying and standardizing workplace policies around wage setting, promotions, and the provision of job skills training and workplace accommodations, unions are partly responsible for declining wage gaps between men and women (Blau and Khan 2017), between African American and white workers (Rosenfeld and Kleykamp 2012), and we presume, workers with and without disabilities.
Although a well-developed sociological literature has documented wage inequality between white, Black and Hispanic workers and male and female workers, scholars like Leicht precarious, nonstandard and non-union jobs where workers are more likely to experience unmet needs due to a lack of necessary accommodations in turn leading to work interruptions and low earnings (Shuey and Jovic 2013).
Labor market outcomes also vary considerably by type of disability. Individuals with more severe disabilities, which tend to include cognitive, IDL, and multiple disabilities, experience even greater barriers with lower employment rates and earnings (Brucker et al. 2016;Maroto andPettinicchio 2014b, 2015;Pettinicchio and Maroto 2017). Employer misperceptions regarding productivity and performance among people with more severe disabilities and, relatedly, the perceived cost and alleged undue hardship of accommodations also shape job assignments (Houtenville and Kalargyrou 2015;Schur et al. 2009). Although unionized employment might undermine irrational job assignments based on these misperceptions, contributing to better working conditions and higher earnings, the limited research in the area points to significant variation between disability type and union representation where people with more "prejudiced disabilities" are less likely to be in unionized employment (Baldwin and Johnson 1994;Bowe et al. 2005;McMahon et al. 2005;O'Hara 2004).
Additionally, people with disabilities are sometimes confronted with potentially contradictory norms and values generated by the interplay between employer preferences, the nature of the work, disability type, and workplace accommodations mandated by law.
Legislation meant to increase employment and earnings among people with disabilities requires that employers provide reasonable accommodations, such as shift changes, transfers, or reassignments. 7 As the chief legislative remedy to improve labor market conditions for Americans with disabilities, reasonable accommodations increase retention and, in the long term, overall earnings (Burkhauser, Butler, and Kim 1995;Chow, Cichocki, and Croft 2014;Solovieva, Dowler, and Walls 2011;Schur, Kruse, and Blanck 2005). Unions play an integral role by providing information and training to supervisors who make decisions about accommodations (Bruyere, Erickson, and VanLooy 2000), as well as support for those who must be absent from work (Baldwin and Johnson 1994;Johnson and Ondrich 1990;Daly and Bound 1996).
Unions can therefore be instrumental in institutionalizing reasonable accommodations that help people with disabilities remain in the workplace, but only if they are able to integrate an individual's right to reasonable accommodations with the principles of collective bargaining.
However, neither the ADA nor the Equal Employment Opportunity Commission (EEOC) has effectively specified how collective agreements should be considered when determining an individual's right to a reasonable accommodation (Browne and Gianpietro-Meyer 1989;Rottenberg 1993;Stalhut 1993). 8 Scholars have suggested that these institutional arrangements pitted workers' collective rights against disability rights, pointing to "tensions" (Roberts and Lundy 1995) and "irreconcilable conflicts" (O'Melveny 1993;Schoen 1997) between disability rights legislation and collective bargaining. This has generated "competing expectations" for employers in addressing accommodations by employees with disabilities (Balser 2000). 9 Reaching common ground between collective agreements, policy, and employer practices is complicated because accommodations can vary considerably across disability type and 8 The ADA's precursor, the 1973 Rehabilitation Act, made no mention of collective bargaining or seniority systems. 9 Courts have for the most part ruled that any reasonable accommodation request seen as violating a collective agreement is an "undue hardship" (Lee 1992;; see also Daly-Rooney 1993 on the 1984 Daubert v. USPS case). In turn, the courts in some cases have used collective bargaining agreements to limit the options regarding accommodations (Daly-Rooney 1993), pitting the interests of disabled workers against the interests of workers without disabilities (O'Melveny 1993). occupational demands. There is no universal accommodation (Lurie 2017). This situation is further problematized by accompanying (mis)perceptions about the alleged unreasonableness associated with different accommodations, leading many people with different disabilities to leave their jobs or the labor market altogether.
Despite research showing significant variation in disability-based labor market outcomes, surprisingly little is known about whether people with disabilities have access to unionized employment and how unionized employment shapes earnings among workers with different types of disabilities. To shed light on the disability union effect, we bring disability into the growing conversation about the effect of unionized employment on earnings gaps and what this means for inequality. In that vein, we include sets of expectations regarding how unions affect inequality when looking at union and non-union workers between and within status groups, which we address through a series of interaction models.
With lower earnings than the rest of the population, we expect that people with disabilities will experience greater wage benefits associated with unionized employment than people without disabilities (hypothesis 1). This will result in a positive interaction term between disability and unionized employment within our models. The greater effects of unionized employment on earnings among people with disabilities will bring disabled union workers' earnings closer to other workers. This will result in smaller between-group differences in earnings among unionized workers with and without disabilities compared to non-unionized workers (hypothesis 2). However, workers with disabilities left out of unionized employment will not experience such benefits. Therefore, we expect larger within-group earnings differences when comparing unionized and non-unionized workers with disabilities than we do between unionized and non-unionized workers without disabilities (hypothesis 3).
Furthermore, we expect the relationship between unionized employment and earnings to vary by disability type. We expect that the benefits of unionized work will be greatest among workers with more severe disabilities, specifically disabilities that include cognitive limitations or barriers to independent living, (hypothesis 4). These workers tend to experience the greatest disadvantages within the labor market. This means that the interaction term between IDL disability and unionized employment will be largest in magnitude. As a result, we expect smaller between-group differences when comparing union workers with cognitive and IDL disabilities to unionized workers without disabilities (hypothesis 5). We also expect larger within-group differences between union and non-union workers with cognitive and IDL disabilities than between union and non-union workers with other disabilities (hypothesis 6). Finally, although differences in demographics, education, and occupation likely explain part of these disparities, we expect to find continuing gaps net of these controls.

Data
We analyze pooled Current Population Survey Merged Outgoing Rotation Group (CPS-MORG) data, also referred to as the Earner Study, for 2009-2018. The Current Population Survey is a national monthly household survey conducted by the U.S. Census Bureau and the U.S. Bureau of Labor Statistics (BLS). The CPS uses a household rotation pattern where households are included in the survey for four consecutive months, left out of the survey for eight months, and then re-interviewed for four months (U.S. Census Bureau 2001). In addition to the monthly CPS questions, employed CPS respondents age 15 and older also answer specific earnings-and work-related questions (including union affiliation) when rotating out of the survey at months 4 and 8. Thus, although other surveys, such as the American Community Survey (ACS) and the National Health Interview Survey (NHIS), contain information on disability status, the CPS is one of the few datasets that also contains detailed information on employment and union membership important to our study.
We include 2009 as the first year in our analysis because this is when the CPS revised their set of disability-related questions to provide better information regarding different types of potential limitations. 10 To ensure that we only observe each individual once, we restrict our sample to respondents exiting in month 4 of their rotation (MISH = 4). We also limit our sample to working-age employed adults between 25 and 61 years of age in order to account for continued schooling and early retirement. After restricting our sample to employed respondents with weekly earnings of $1 or greater and at least some work hours in the previous week, we obtained a sample of 630,799 respondents for 2009-2018. Finally, we weight all descriptive statistics and model estimates using survey-provided weights divided by 12 to account for the inclusion of all monthly samples per year (EARNWT/12).

Methods and Measures
We use a series of linear regression models to estimate a respondent's logged weekly earnings at their current job. We log this variable in order to account for the skewed earnings distribution and to satisfy model assumptions. All monetary amounts are adjusted for inflation and appear in 2018 U.S. dollars. As shown in Table 1, average weekly wage and salary income between 2009 and 2018 amounted to $1,023.

[Table 1]
Our primary predictor variables relate to the respondent's membership in unionized employment and their disability status and disability type. Unionized employment is a dichotomous variable, indicating whether the respondent is covered by a union or employment contract in their current job. Disability status includes limitations based on six questions used by the CPS to identify the population with disabilities (Livermore et al. 2011). In addition to disability status, we also include five variables to measure the type of disability. These variables indicate whether the respondent reported a cognitive, physical (ambulatory), hearing, vision, or independent living or self-care disability. 11 Cognitive difficulties include those related to learning, remembering, concentrating, or making decisions. Ambulatory difficulties include anything that limits a respondent in one or more basic physical activities. Vision difficulties indicate whether the respondent was blind or had serious difficulty seeing even with corrective lenses. Hearing difficulties indicate whether the respondent was deaf or had serious difficulty hearing. Independent living difficulties (IDL) indicate the presence of any condition lasting six months or more that makes it "difficult or impossible to perform basic activities outside the home alone," and self-care difficulties include personal needs, such as bathing and dressing.
Approximately three percent of the employed working-age population reported one or more disabilities, but these rates varied across type of disability (Table 1). 12 The relationship between disability, unionized employment, and earnings is also likely influenced by groups' varying demographics, education, work hours, and occupation -factors that are associated with both earnings and unionization (Blanck et al. 2007;Jones 2008;Maroto 11 Due to the small percentage of individuals reporting an IDL or self-care disability and the overlap in reporting of these disabilities, we combine independent living and self-care into a single measure. 12 We also include a series of supplemental models with additional measures for the number of reported disabilities and the severity of the disability within the appendix. and Pettinicchio 2014b). For example, men continue to have higher unionization rates than women, and despite African American overrepresentation in unionized employment, unionized African American workers are still disadvantaged compared to unionized white workers (Denice and Rosenfeld 2018; Rosenfeld and Kleykamp 2012). We therefore include a series of demographic and human capital control variables to account for these relationships.
Gender is a categorical variable of male (the referent) or female. We indicate race/ethnicity with a categorical variable measured as Non-Hispanic white (the referent), Non-Hispanic Black, Hispanic, or Non-Hispanic other. 13 We measure age in years and include a quadratic age-squared term to account for its non-linear relationship with employment and earnings. We measure educational attainment with a categorical variable that indicates whether the respondent obtained a high school diploma or equivalent degree (the referent), did not complete high school; attended some college without obtaining a degree; obtained an Associate's (two-year) degree; completed college with a Bachelor's (four-year) degree; or obtained a Master's, professional, or Doctorate degree. We measure marital status with a categorical variable that indicates whether the respondent was currently married (the referent), never married, or separated, divorced, or widowed.
In addition to demographics, the labor market position theory of unionization (Rosenfeld and Kleykamp 2012) posits that where status groups are located in the labor market determines their respective access to union jobs and therefore, higher earnings. Unionization levels range from less than 5% unionized in the agricultural sector to over 40% unionized in transportation manufacturing (Western and Rosenfeld 2011). Given that prior research shows that people with disabilities are located in sectors with low unionization levels -jobs characterized by irregular scheduling and nonstandard arrangements (Jones 2008;Shuey and Jovic 2013;Maroto and Pettinicchio 2014b) -controlling for occupation should account for some of the union and earnings differences between and within groups. We therefore control for detailed occupation, which includes 569 occupational categories. We also include measures to indicate whether the respondent was a government employee and whether they worked part time, as defined by their work status in their current job. These sub-national institutional contexts may shape how unions work to alleviate working poverty across states. To better account for differing legal and policy environments, we also include indicator variables for the respondent's state of residence and for the survey year.
In two sets of linear regression models, we interact disability and disability type with union membership to estimate the varying effects of unionized employment on logged earnings across groups. We present models with and without control variables to broadly address our expectations in hypotheses 1 and 5 with a discussion of unionization rates and earnings by disability. We then discuss predicted percent differences in earnings between union workers with and without any disability and different disabilities to address hypotheses 2 and 5 and within groups of unionized and non-unionized workers with disabilities to address hypotheses 3 and 6. 14

Results
Although union strength is far weaker today than it was decades ago, our results show the importance of union membership for workers' earnings, especially among workers with disabilities. We begin with a discussion of rates of unionized employment among workers with disabilities to provide context for our main findings related to unionization and earnings. We then discuss how the benefits of unionized employment across groups present varying consequences for between-and within-group inequality.

Unionized Employment among Workers with Disabilities
Access to unions varies according to disability status and in turn partially explains earnings differences within and between groups.

[Figure 1]
To expand on the descriptive information provided in Figure 1, Table 2 regresses logged weekly earnings on disability status, incorporating a set of interactions with union membership. On average, union membership was associated with a 23% increase in weekly earnings in models without controls, and a 15% increase in weekly earnings after accounting for demographic, human capital, and work variables in Model 2. The positive and significant interaction terms in Table 2 show that the effects of unionized employment varied by disability status, as indicated in hypothesis 1. Prior to incorporating control variables in Model 1, union membership was associated with a 54% (e (0.204+0.230) -1) increase in weekly earnings among workers with disabilities. After incorporating control variables in Model 2, however, union membership still resulted in an earnings boost of 30% for employed people with disabilities.
Workers with cognitive, physical, hearing, and IDL disabilities also received greater benefits from unionized employment, as indicated by the positive and significant interaction terms in Table 3 and supporting hypothesis 4.
This first set of results shows that even though union membership may have positive universalizing effects for all groups, union jobs especially benefitted workers with disabilities, supporting our first and fourth hypotheses. Importantly, these findings held even after accounting for human capital differences, work situation, occupation, and state context. This shows that although unionized employment varies across occupations where occupational segregation can negatively affect workers' wages, the positive effects of union membership transcend workers' location in the labor market.
To better illustrate union membership's varying wage effects across different groups, Tables 2 and 3 as a predicted percent change in weekly earnings associated with unionized employment and disability. Figure   2 highlights between-group differences and Figure 3 illustrates within-group differences. 15 As shown by these figures, the greater effects of unionized employment on earnings for workers with disabilities resulted in smaller disparities between unionized workers with and without disabilities when compared to non-unionized workers, as predicted by our second and fifth hypotheses. However, it also led to larger disparities within groups of disabled workers when comparing workers with and without access to unionized employment, as expected by hypotheses 3 and 6.

Figures 2 and 3 present the results of the interaction effects in
[ Because union membership had stronger effects on weekly earnings for people with disabilities, as indicated by the positive interaction term in Table 2, earnings inequality was also smaller among disabled union members (hypothesis 2). Non-unionized workers with disabilities earned approximately 13% less than other non-unionized workers without disabilities, but unionized workers with disabilities earned only 3% less than unionized workers without disabilities, net of other individual characteristics and labor market factors. 16 So, while disabled unionized workers still earned less than non-disabled workers, their earnings were much closer to non-disabled 15 Predicted values and standard errors are also presented in the appendix. 16 Effects are presented as percent changes in weekly earnings and calculated using the formula (e b -1)*100. workers' earnings than non-unionized workers with disabilities were to non-disabled workers' earnings .
Inequality between workers with most types of disabilities was also smaller among unionized workers, as shown in Figure 2 and posited in hypothesis 4. Because the benefits of unionization were greater for people with cognitive and IDL disabilities, members of these groups also saw smaller gaps in inequality. Unionized workers with cognitive disabilities earned 9% less than other similar unionized workers compared to non-unionized workers who earned 17% less, and unionized workers with IDL disabilities earned 7% less than other workers compared to non-unionized workers who earned 22% less. For workers with physical, hearing, and visual disabilities, where the smallest disparities already existed, being in unionized work removed earnings disparities entirely, once other factors were controlled for.
Unionized employment clearly reduces between-group inequality but the effects are not the same for all disabilities, as we expected in hypothesis 3. Compared to other disabilities, workers with cognitive or IDL disabilities benefit the most from union membership. This may be because baseline wages and earnings among these groups are already relatively low compared to other historically disadvantaged groups; even the small premiums union membership provides do a great deal to improve labor market conditions and earnings for disabled workers. Given where people with disabilities are situated in the labor market, collective bargaining agreements, increased transparency, and wage setting structures supported by unions may indeed be helping to combat workplace attitudes and norms that limit opportunities for workers with disabilities.
Finally, selection may also be a factor where more productive workers with disabilities could be more likely than other workers with disabilities to gain entrance into unions. Whatever the case, our findings cast some doubt on the notion that potential conflicts generated by unionism lead to negative labor market outcomes among disabled workers.

Unionized Employment and Within-Group Inequality
Because unionized workers with disabilities earned a greater premium than non-union workers, union membership actually resulted in widening inequality among people with disabilities. As shown in Figure 3, unionized workers without disabilities earned, on average, 15% more per week. However, unionized workers with disabilities earned 28% more than otherwise similar non-unionized workers. These greater effects mean larger gaps by union membership among workers with disabilities, making access to union jobs an important component of wage inequality.
In line with our fourth hypothesis, disaggregating disability into different types showed that union membership was most beneficial to workers with the severest disabilities, but this resulted in greater within-group inequality among disabled workers, as predicted in hypothesis 6.
For instance, after incorporating control variables, weekly earnings were approximately 36% higher for unionized workers with IDL disabilities and 26% higher for those with cognitive disabilities compared to non-unionized workers with these same disabilities. The results in Figure 3 confirm that union membership greatly benefited workers with disabilities while increasing inequality within disability groups based on whether workers were unionized.
Our analyses situate the effects of union membership on earnings within a context of unions' universalizing positive effects across historically disadvantaged workers. We find that unionized employment reduces inequalities between workers with and without disabilities, acting as a countervailing force on the earnings penalties associated with disability. We also find that the effects of unions vary among disabled workers where unionized employment significantly increases within-group inequality, much like with other groups. Finally, even after accounting for the varying effects of unionized employment, workers with different disabilities continue to experience earnings disparities suggesting that unions only go so far in combatting labor market inequality.

Discussion
Twenty-five years ago, Cornfield asked whether union membership has differing effects across segments of the working population. Our results provide a resounding "yes" to this question. This is especially relevant given declining union membership in the United States and mounting political resistance constraining the ability of unions to protect already marginalized workers. We make three general conclusions based on our analyses. First, due to the positive effects on wages, union membership reduces earnings inequality between workers with and without disabilities while simultaneously increasing inequality within groups of workers with disabilities. The latter is attributable to workers in union jobs earning more than non-unionized workers even within the same occupations. Workers with disabilities experienced additional earnings boosts that reduced inequality among those in unionized employment but contributed to inequality across union and non-union disabled workers. While especially acute, these reflect broader patterns found between union and non-union Black and white workers and male and female workers. Uneven access to unionized employment means that only some workers from marginalized groups are able to experience the important benefits of unionization.
Second, ours is one of the few studies thus far to test how unionized employment is associated with labor market outcomes for people with different types of disabilities. We found that even in light of de-unionization and potential legal conflicts, the effects of union membership on weekly earnings were greatest for people with more severe disabilities who started out with much lower earnings. Among workers with physical, hearing, or visual disabilities, who experienced smaller wages gaps, union membership completely closed the disability wage gap.
Third, our analyses point to the critical role of unionized employment in combatting inequality and working poverty among marginalized groups in the labor market. The interplay between disability rights and collective bargaining can serve as an opportunity for unions to institutionalize the provision of reasonable accommodations and norms around antidiscrimination, especially in sectors where disabled workers experience significant structural, organizational, and attitudinal barriers. Our paper, therefore, represents an additional step in situating disability more centrally within sociological theories of stratification, work, and organized labor.
The ability of unions to combat inequality involves understanding how members of disadvantaged groups are organized within the labor market. Even though all occupations are governed by some form of inequality regime (Acker 2006), the level of inequality and distance between groups in terms of pay, promotions, and authority can vary substantially across work organizations. In a so-called dual market of "good" and "bad" jobs, unions may not be able to successfully contend with occupational segregation (Blau, Brummand, and Yung-Hsu Liu 2013;Charles and Grusky 2004;England 2005;Kmec 2003;2005;Maroto and Pettinicchio 2014b;Reskin, McBrier, and Kmec 1999). They may also be ineffective in overcoming occupational preferences about job assignments that can limit unions' ability to increase workers' earnings contributing to dramatic earnings inequalities between groups across occupations (Cornfield 1991;Card 2001;Elvira and Saporta 2001;Maume 1999;Weeden et al. 2007). 17 While challenged by these limitations, unions still offer one of the few means for workers to collectively improve their economic wellbeing, This is why net of occupation, we find broad benefits to unionized employment particularly for people with disabilities.
Despite the contributions of this work, our study does have some limitations. First, because we examine union membership and earnings outcomes, our study includes only employed people with disabilities, which is a small subset of the disabled population. With an employment rate of 19% in 2019, this means we omit the 80% of people with disabilities without employment. However, understanding barriers among employed workers with disabilities can still provide insight into how the labor market functions to exclude those without jobs, including unemployed persons and discouraged workers.
Second, we could not directly test the effects of potential conflicts between collective bargaining agreements and accommodations or how these might vary by the nature of the disability. Nonetheless, if policy challenges indeed diminish the ability of unions to improve economic wellbeing, we would not see as great of a wage benefit for unionized workers with disabilities. In turn, between-group differences with disability would be larger because people with disabilities would not be experiencing the same positive effects of unionization. Conversely, within-group differences would be smaller because we would see less of an earnings difference between union and non-union workers with disabilities.
Qualitative and quantitative studies combining cross-sectional or longitudinal data with state-level legislation would help illustrate the connections between collective bargaining and 17 It may also be the case that occupational segregation and isolation determine whether workers are able to unionize, rather than the reverse. Bridges (1982) proposed that unionization is difficult in sectors where existing occupational segregation is high. However, his findings do not support this hypothesis. disability rights policies in shaping earnings. In the case of disability, much of the relationship between unions and a firm's ability to negotiate accommodations depends on the content of CBAs (see Balser 2007). It is not always clear if CBAs promote institutional discrimination by supporting corporate culture and certain managerial practices (Schur, Kruse, and Blanck 2005 Third, although incorporating variables for disability type allowed us to assess how the relationship between disability, union membership, and earnings varied with the severity of a disability, we were still limited by the available data. Other measures, such as the number of disabilities and the combination of different types of disabilities, can also help to assess disability severity. Results from supplemental models (provided within the appendix) show that persons with more severe disabilities, measured as the number of disabilities and the presence of a concurrent IDL disability, experienced the largest earnings declines but the greatest benefits from unionization, supporting the results presented in the paper. Additional studies examining other measures of disability, as well as those related to health status, would be helpful to further elaborate on these findings. Fourth, as with most cross-sectional studies, we were unable to fully control for selection into employment, unionized jobs, and occupations within our models. We include a variety of controls for education, work hours, and prior history; however, it is still possible that better and more productive workers might end up in unions, thereby increasing wages. Nevertheless, our controls for detailed occupation further help to address issues of selection, allowing us to compare union and non-union workers in the same occupations, which require similar skill sets.
Using longitudinal data to assess within-person variation in earnings before and after  (Drew, Flood, and Warren 2014), this is often too short a period to examine how changes in CBAs, specific union efforts, and, importantly, changes in disability status, shape outcomes in the long-term.

Conclusion
The relationship between economic inequality and unions depends a great deal on how different groups gain access to unionized work in a context of de-unionization. Even though rates of unionization remain low for workers today, union membership increases wages for all unionized workers. Unions' purported universalizing effects have reduced differences in earnings between people with and without disabilities, between men and women, and between white, Black, and Hispanic workers. But, gaps based on union/non-union status within race, gender, and disability groups have widened given the stronger effects of union membership on earnings within these historically marginalized groups. Consequently, more earnings inequality exists within groups than between groups today.
As the "chief effort" to reduce social inequalities, unions counteract inequalities that work norms and cultures help maintain and often exacerbate (Cornfield 1991). They improve wages and job security, advocate for educational and training programs, limit precarious employment, and increase worker say within organizations. When unions engage with precarious workers and those in non-standard work arrangements (arrangements that often contain a disproportionate share of individuals from historically marginalized groups), they significantly improve those workers' earnings (Gomez and Lamb 2019).
Electronic copy available at: https://ssrn.com/abstract=3690958 Unions are especially important for workers from vulnerable groups who often lack voice in the labor market (Freeman and Medoff 1984). Yet, their efforts are filtered through specific institutional arrangements, including policies working against them (Kollmeyer and Peters 2019). Variability in unions' positive influences must be understood against a backdrop of political efforts threatening organized labor (Bound and Dresser 1999;Blanchflower 2007;Freeman 1993;Milkman 2007;Mishel 2012;Western and Rosenfeld 2011;Rosenfeld and Kleykamp 2012;Telles and Ortiz 2008). With the recent Supreme Court ruling in Janus vs. AFSCME, which limited unions' abilities to collect fees from non-union members, it is likely that union strength will continue to decline, particularly within the public sector, compounding already rising inequality. But, even with their declining economic and political influence, our research shows that unions are substantially beneficial for workers with disabilities and their positive efforts would be even more so if not for the current climate of de-unionization.
The debate about how and when unionized employment helps people with disabilities continues as legislative, judicial, labor market, and union contexts remain in flux (Pettinicchio 2017(Pettinicchio , 2019. Labor market trends raise questions about whether and how unions combat inequality given an undesirable political and legislative climate. Limited work in the area points to situations where unions have not been proactive in promoting rights to reasonable accommodations, treating flexible scheduling and working from home as undermining the collective interests of other workers (Balser 2002(Balser , 2007O'Hara 2004;). As Stahlhut (1993, 74) argued, "unions are worried about whether the obligation of reasonable accommodation will allow an employer to circumvent the seniority provisions of a collective bargaining agreement.
Management on the other hand, is perplexed because it needs to know its obligations under the ADA so it can avoid legal liability." It may very well be, as Lurie (2017) suggested, that unions increase job security for all employees indirectly, but they may not be advocating for reasonable job accommodations if these are seen as challenging collective agreements.
Although we do not deny the existence of such situations and experiences, our findings suggest that accommodations are not inherently problematic for unionized workers with disabilities. Unions can, as Campolieti (2005) found, provide accommodations to disabled employees so that they remain in their jobs thereby reducing the duration of worker compensation claims. They may even help turn low-paying, non-standard jobs, prevalent among disabled workers, into pathways to full-time stable work (Schur 2002(Schur , 2003. In other words, collective agreements can be used as tools to mitigate conflicts rather than treated as challenges to disability rights legislation (O'Melveny 1993).
Our findings and discussion raise broader concerns about how the legitimacy and visibility of certain inequalities might influence outcomes among minority groups. Inequalities that are seen as legitimate (or not seen at all) are more likely to persist (Acker 2006). Most employers and workers view gender and race inequality as illegitimate -as having no reasonable basis (Cotter, Hermsen, and Vanneman 2011;Marsden 2012). In the case of disability, however, unequal outcomes are still predominately viewed as legitimate. Many assume that people with disabilities are simply unable to work and that a disability will limit productivity, despite data that directly contradicts these sentiments. This is compounded by a lack of awareness of the barriers that people with disabilities face within the labor market, despite antidiscrimination and equal rights legislation meant to rectify these wrongs. Given these broader concerns, it is important to ask how unions interact with employers to change widespread negative attitudes that continue to act as barriers to employment and promote the kinds of mechanisms outlined by legislation (like reasonable accommodations) to help people with disabilities retain gainful employment. Corneo, Giacomo, and Claudio Lucifora. 1997   171234 NOTES: Continuous variables are mean centered. "eb-1" refers to the percent change in weekly earnings associated with the predictor variable. All models control for detailed occupation, state of residence, and survey year. Models also incorporate survey-provided weights. SOURCE: Current Population Survey, MORG, 2009-2018, working age adults 25-61 years old with employment and earnings, N=630,799 *** p<.001, ** p<.01, * p<.05 171234 NOTES: Continuous variables are mean centered. "e b -1" refers to the percent change in weekly earnings associated with the predictor variable. All models control for detailed occupation, state of residence, and survey year. Models also incorporate survey-provided weights. Electronic copy available at: https://ssrn.com/abstract=3690958 Tables 2 and 3. Betweengroup effects refer to differences between each disability group and respondents without disabilities. Results are further broken down by union membership. Estimates can be interpreted as the percent difference in earnings associated with the presence of each specified disability.

NOTES: Estimates based on interaction models with all covariates in
Error bars indicate 95% confidence intervals.   Tables 2 and 3. Withingroup effects refer to differences for unionized employment among workers with different types of disabilities. Estimates can be interpreted as the percent difference in earnings associated with being in a union among workers with each specified disability.  Table  S2 includes a continuous variable measuring the number of disabilities, along with a squared term. Model 2 incorporates a categorical variable for the number of disabilities with the following categories: no disability (referent), one disability, two disability, three disability, and four or more disabilities. Model 3 incorporates a categorical variable for the number of disabilities with the following categories: no disability (referent), single disability, and multiple disabilities. Model 1 in Table S3 incorporates a categorical variable that combines different disabilities with IDL disabilities to assess the severity of disability. This variable includes the following categories: no disability (referent), cognitive disability, physical disability, hearing disability, vision disability, IDL disability, cognitive and IDL disability, physical and IDL disability, hearing and IDL disability, and vision and IDL disability. All models include the same covariates as those presented in the full paper and interaction terms between disability type and union status.
Results from the supplemental models show that persons with more severe disabilities, measured as the number of disabilities and the presence of a concurrent IDL disability, experienced the largest earnings declines but the greatest benefits from unionization. In Table S2 Model 1, for non-unionized workers, each additional disability was associated with a 20% decrease in earnings but the strength of the negative association decreased with the number of disabilities. Unionization, however, was associated with increased earnings for workers with disabilities but the effects of this variable also decreased with the number of disabilities. Models 2 and 3 in Table S2 further support these findings with categorical variables. As shown in Model 3, having a single disability was associated with an earnings decrease of 9% and having multiple disabilities was associated with a decrease of 25%. Among workers with a single disability, unionization increased earnings by approximately 23%, but among workers with multiple disabilities the increase was approximately 38%. Finally, Model 1 in Table S3 shows that across all disability types, having an additional IDL disability was associated with larger earnings declines, but unionization was also more beneficial for these workers in most situations. Table S4 includes supplemental models from OLS models estimating logged weekly earnings with interaction terms for gender, race/ethnicity, and union membership. The positive and significant interaction terms in Models 1 and 3 show that the effects of unionization varied across groups. Union membership was associated with larger increases in weekly earnings among almost all minority groups prior to controlling for other covariates. However, for individuals in the non-Hispanic other category, unionization was actually associated with smaller increases in earnings. Many of these relationships did not hold up, however, once we accounted for differences in demographics, education, and work characteristics. After incorporating control variables in Models 2 and 4, union membership resulted in a greater boost in earnings for only Hispanic workers.
Finally, Table S5 presents values for the predicted percent difference in earnings by union status and minority group membership before and after controlling for covariates. This table provides values for people with different types of present in Table 3 and Figures 1 and 2 in the full paper.
It also provides added results based on gender and race interactions in Table S4.
For women, non-Hispanic Black workers, and workers in the non-Hispanic other category however, unionization was not very helpful in decreasing inequality after accounting for other covariates. Unionized women earned 16% less than unionized men, but non-union women earned 17% less. Non-Hispanic black workers earned 9% less than otherwise similar non-Hispanic white workers, regardless of whether or not they were unionized. Unionized workers in the non-Hispanic other category earned 7% less, and non-union workers earned 5% less, on average, Finally, unionized Hispanic workers earned 5% less than unionized non-Hispanic white workers, and non-unionized workers earned 8% less.
Within-group inequalities based on union membership among other disadvantaged groups were less pronounced. For instance, unionization increased hourly earnings by 15% for both men and women. Unionized non-Hispanic white workers earned 15% more than otherwise similar nonunion workers, unionized non-Hispanic black workers earned 14% more than other workers, unionized Hispanic workers earned 19% more per week, and unionized non-Hispanic workers from other racial categories earned 13% more per week.    .501 AIC 171191 NOTES: Continuous variables are mean centered. "e b -1" refers to the percent change in weekly earnings associated with the predictor variable when multiplied by 100. All models control for detailed occupation, state of residence, and survey year. Models also incorporate survey-provided weights. SOURCE: Current Population Survey, MORG, 2009-2018, working age adults 25-61 years old with employment and earnings, N=630,799 *** p<.001, ** p<.01, * p<.05 .500 AIC 324581 171493 NOTES: Continuous variables are mean centered. "eb-1" refers to the percent change in weekly earnings associated with the predictor variable. All models control for detailed occupation, state of residence, and survey year. Models also incorporate survey-provided weights. SOURCE: Current Population Survey, MORG, 2009-2018, working age adults 25-61 years old with employment and earnings, N=630,799 *** p<.001, ** p<.01, * p<.05 .500 AIC 326397 171477 NOTES: Continuous variables are mean centered. "e b -1" refers to the percent change in weekly earnings associated with the predictor variable. All models control for detailed occupation, state of residence, and survey year. Models also incorporate survey-provided weights. SOURCE: Current Population Survey, MORG, 2009-2018, working age adults 25-61 years old with employment and earnings, N=630,799 *** p<.001, ** p<.01, * p<.05  Tables 3 and 4. The columns for within-group differences compare union and non-union workers within disability, gender, and race groups. The columns for between-group differences in earnings compare results to the specified referent category for each group (no disability, male, and NH white) for union and non-union members. SOURCE: Current Population Survey, MORG, 2009-2018, working age adults 25-61 years old with employment and earnings, N=630,799