Abstract
Starting in the 1980s, US employers revived aggressive action against unions. Employers’ public opposition to unions yielded a scholarly consensus that US employers actively and consistently discriminate against union supporters. However, evidence for widespread employer anti-union discrimination is based mainly on employer reactions to rare union organizing campaigns. To measure baseline or preventive anti-union discrimination, the authors field the first ever US-based résumé correspondence study of employer responses to union supporter applicants. Focus is on entry-level, non-college degree jobs and findings show no difference in employer callback rates for union supporter applicants relative to non-union applicants. Drawing on interviews and survey data, the authors suggest that union weakness itself may have hollowed preventive employer discrimination against union supporters.
The long decline in US union representation has transformed the labor market, spurring wage stagnation and growing earnings inequality. The sources of this decline in union influence are various, but a central proximate cause is renewed employer aggression against unions. In the 1980s, employers stiffened resistance to union organizing drives, demanded concessions in collective bargaining, and began hiring permanent replacement workers during strikes. Outside the drama of active organizing drives and strikes, employers also experimented with covert, preventive anti-union practices designed to undermine an organizing drive before it even begins. The most prominent of these preventive practices is to screen out union supporters at the point of hire. In this article, we present the first ever US study on the prevalence of preventive anti-union hiring discrimination.
We focus on anti-union hiring discrimination for two reasons. First, it is important in its own right, as widespread employer anti-union hiring discrimination (or even workers’ perception of such discrimination) can deter workers from active union support. Second, anti-union hiring discrimination is a measurable example of preventive anti-union practices more broadly, which are typically covert and challenging to measure systematically. Screening out union supporter job applicants, while illegal, is difficult for regulators to monitor and thereby offers employers a low-cost opportunity to decrease the risk of union organizing. Historically, this has led some employers to embrace coded interview questions or even hiring attitude questionnaires to identify potential union supporters. But, as with other forms of preventive union avoidance (such as surveillance of union interest among workers or anti-union messaging), researchers face distinctive challenges in trying to measure the prevalence of this practice.
Indeed, unlike the well-publicized instances of open employer anti-union campaigning, employers engaging in preventive union avoidance may seek to conceal their activities. Concealment is particularly likely in the case of anti-union hiring discrimination, which is illegal under the National Labor Relations Act. As such, opinion surveys and interviews (which we nonetheless use as supplementary evidence in this article) about these topics are subject to social desirability bias. Studying reports of actual employer anti-union behavior during organizing drives mitigates this problem. But that research setting selects on situations of very high union salience and threat. Organizing drives are rare and unlikely representative of employer prioritization and focus on unions more generally. The best prior US study of preventive anti-union hiring discrimination (discussed below) uses a careful case study of a single workplace, where employee surveys and hiring outcomes were available, but is unable to rule out unmeasured differences between union and non-union supporter applicants.
To solve the problems involved in identifying preventive anti-union hiring discrimination, we conduct a résumé correspondence audit study. The correspondence audit method allows us to isolate employers’ reaction to an applicant’s union affiliation and support, net of work experience, education, race, and other confounders. Audits have been used to study employer discrimination on applicant characteristics ranging from race and sexual orientation to employment histories marked by nonstandard employment or family leave. We also build on one Belgian audit study (discussed below), the closest to our article, which measured employer discrimination based on applicant involvement in youth trade unions.
To measure employer responses to union-affiliated applicants, we compare callback rates between applications with and without disclosure of labor union affiliation and involvement at a prior job. We submitted 1,025 résumés and cover letters to 514 entry-level, high school graduate jobs in Chicago. We find that employers are no less likely to call back union supporter applicants than applicants with no union affiliation. Specifically, we can rule out levels of employer anti-union discrimination at magnitudes comparable to the discrimination suffered by workers who are Black, gay, unemployed, or from a low socioeconomic class background. This precisely estimated null effect is not due to a weak signal of union affiliation in the résumés: In a separate, online survey, 80% of respondents identify our application materials as those of a union supporter. Moreover, the null discrimination effect holds even for establishments more exposed to union activity; during the period following a high-profile teachers’ strike; and across stronger and weaker signals of union affiliation. To understand this result, surprising in the context of prior consensus about employers’ widespread anti-union discrimination, we interviewed 20 Chicago employers. These conversations made clear that labor unions are typically not a focus of employers of non-college workers, either during the hiring process or beyond. We also discuss survey data showing that managers’ distrust of unions has declined markedly since its early 1990s high point.
Our study makes several contributions. First, we introduce correspondence audit methods to US labor relations researchers. Using this method to determine whether employers discriminate against union-affiliated applicants, and under what conditions, could help researchers, labor unions, and National Labor Relations Board investigators track anti-union discrimination. To that end, we make all our study materials and code publicly available. Our baseline study offers a point of comparison for future research in different cities, different job types, and concerning worker applicants with different ascriptive characteristics. Second, our findings suggest that employer discrimination against union supporters at the point of hire, at least among the low-wage, Chicago-based employers we study, is rare. If this is the case, it offers further evidence for the very low influence unions have in the labor market during our era of long-term union decline. If employers do not discriminate against union supporters when given an easy opportunity to do so, they are unlikely to invest in costly pro-worker human resources management practices to stave off the increasingly remote threat of union organizing. Third, our study suggests that opponent weakness can erode seemingly fundamental class priorities. After a long period of pitched anti-union activism, US employers’ focus may have turned elsewhere.
Employers’ Anti-union Discrimination
In an earlier era of union decline, much was made of employer efforts to lock union supporters out of jobs. In the early 1900s, the Employers’ Association of Detroit kept records on tens of thousands of industrial workers, blacklisting “agitators” and union supporters (Jacoby 2004: 14). One manufacturer filtered out applicants with a “special notation after the names of all those suspected of union activity . . . ROBAL, which is LABOR spelled backwards” (Clawson 1980: 111). In the 1920s, shipowners fought union organizers in the longshore industry by demanding registration of all dockworkers to identify “Radicals, Soap Box Orators, and Wobblies” (Kimeldorf 1988: 33). The 1935 Wagner Act testified to the perceived importance of these practices by banning them: Any “discrimination in regard to hire . . . to discourage membership in any labor organization” was made an unfair labor practice (U.S. Congress 1934).
The Wagner Act itself, together with an upsurge of industrial union organizing, ultimately contributed to the erosion of employer resistance to labor unions. Yet, the vehement anti-union action on display in old blacklisting practices appeared resurgent during the 1980s. A new era of industrial relations was marked by stiff employer resistance to both new union organizing and to once-routine bargaining proposals from increasingly embattled incumbent unions (Kochan, McKersie, and Chalykoff 1986; Kochan, Katz, and McKersie 1994).
The ultimate sources of this revived employer opposition are multiple. Employer associations organized increasingly strident lobbying and legal strategies to chip away at union protections (Hacker and Pierson 2010). Then President Ronald Reagan’s replacement of striking Professional Air Traffic Controllers Organization (PATCO) air traffic controllers signaled high-level political support for strikebreaking (Schnell and Gramm 1994; Rosenfeld 2006; McCartin 2011). The anti-union consulting industry, which executes campaigns of persuasion and intimidation to prevent union recognition, expanded rapidly and became a near-universal presence in new organizing campaigns (Smith 2003; Logan 2006). Deeper economic forces, such as intensified foreign competition and an increasingly dynamic economy (Hirsch 2008), also raised the costs of traditional collective bargaining arrangements. By the 1990s, it was clear that employer anti-union activism had reached a level of intensity that undermined the standard union recognition processes originating in the Wagner Act (Ferguson 2008; Bronfenbrenner 2009).
Did preventive anti-union hiring discrimination return alongside strike replacements and all-out opposition to new union organizing? Commentators have typically emphasized that alongside the drama of public labor conflicts, employers embraced a range of subtler, preventive union-avoidance practices. These practices range from union-substitute human resource management policies (Foulkes 1980; Kochan et al. 1994) to anti-union messaging during onboarding or sophisticated surveillance of employee interest in unions (Greenhouse 2015; Burns 2018). As one human resources textbook put it, companies in this period became “significantly more savvy in union avoidance” (Lussier and Hendon 2017: 376).
Moreover, the meager remedies available under the NLRA were insufficient to entirely deter hiring discrimination against union supporters (Leap, Hendrix, Cantrell, and Taylor 1990; Saltzman 1995). Instances of anti-union hiring screening appear in the literature, particularly in studies of industries with a strong history of union organizing and particularly during the early period of employers’ offensive against unions. One manager of a reopened California steel plant in the 1980s admitted, “Many of the former workers were pro-union, and we don’t hire them. Remember, we’re trying to run it nonunion” (Milkman 1991: 106). In the best prior study of US anti-union hiring discrimination, Saltzman found that in a southern auto plant during the early 1990s, managers screened out union supporters by seeking workers with no prior auto industry employment experience (Saltzman 1995: 91). In 1980, an anti-union consultant offered coded language for this approach, urging employers “to screen applicants closely and to get rid of ‘anybody who is not . . . a team player’” (Smith 2003: 108).
This research focuses on providing rich case study evidence on the ways preventive union avoidance can be carried out. But it does not address how common it is. More broadly, prominent reports of preventive anti-union discrimination in the media frequently come from heavily unionized industries in the early period of the employer anti-union offensive or from the rare large firm that faces persistent union organizing attempts, such as Walmart or Amazon (Greenhouse 2015; Burns 2018). No prior research establishes employer anti-union discrimination (as opposed to surveyed attitudes) in a baseline, preventive setting (outside of rare organizing drives) and that is generalized (across many employers, not just employers most frequently targeted by labor unions). This overall prevalence of preventive anti-union discrimination is what we study in this article.
Indeed, even if generalized, preventive anti-union discrimination characterized employers in the 1980s and 1990s, it may not capture contemporary employer behavior. The very success of the employer offensive against unions may have contributed to the erosion of employer vigilance. The unity and virulence of employers’ anti-union activism was in part a function of the strength and assertiveness of organized labor in the 1970s (Feurer and Pearson 2017). A long line of research on countermovements and countermobilizations finds that as the threat posed by an original social movement fades, countermobilizing loses its urgency (Meyer and Staggenborg 1996; Soule and Olzak 2004).
This countermovement decline is particularly likely in the context of countermovements led by for-profit businesses. Amid a new employment rights regime, focused on individual workers and ascriptive characteristics (Dobbin and Sutton 1998), continual anti-union mobilization by employers could be costly and distracting. Consistent with these scope conditions of countermobilization theory, US labor unions now represent only 7% of private-sector workers, down from a post–WWII peak of more than a third of workers. Even in industries with a legacy of union influence, employers rarely face an active union organizing threat. Within organizations, employers are decreasingly likely to employ human resources personnel who have any experience with labor unions. As a result, at many employers no formal work function is in place to undergird a constituency that pushes preventive union avoidance onto the agenda.
In sum, employer anti-union activism resurged in the 1980s and persists in harsh employer reactions to infrequent union organizing drives in more recent years. But, little evidence addresses whether preventive anti-union discrimination cemented among employers in general during this period. Moreover, even if generalized, preventive anti-union action accurately characterized employers during the early years of anti-union countermobilization, and in the highest union density industries, this posture may have been a victim of its own success. Prior research, which focuses on employer reactions to union organizing drives or develops case studies of hiring discrimination, cannot assess the prevalence of preventive anti-union discrimination. As such, in what follows, we test for employer anti-union discrimination at the point where early 20th-century employers elaborated blacklists and screened out union supporters: the point of hire.
Auditing Anti-union Hiring Discrimination
Assessing employers’ level of preventive anti-union discrimination requires data on employer reactions to unions during periods beyond organizing drives. Unfortunately, it is difficult to measure the level of anti-union discrimination using observational data or social survey experiments. Observational data, for instance on unemployment duration among union and non-union workers, cannot distinguish effects of union affiliation from unobserved worker characteristics—such as the specifics of educational attainment or work history. Surveys that ask employers about their hiring decisions suffer from desirability bias (Phillips and Clancy 1972), wherein employers avoid acknowledging practices they perceive as shameful, let alone illegal.
For these reasons, discrimination research increasingly relies on experiments known as correspondence audit studies. In an audit, researchers submit job materials reflecting hypothetical applicants to real job postings and observe employer responses to an experimentally varied application characteristic. Audits are useful for assessing differential screening practices in hiring because they match hypothetical applicants on every characteristic except for a treatment of interest—in this case union affiliation as an indication of union support—to estimate the causal effect of that treatment on employer responses. The audit method was originally developed as an in-person strategy for regulators to test for racial discrimination in housing in the 1970s. Recent audits have extended the method online to study discrimination by gender, class, sexual orientation, age, disability status, and nontraditional work experience (Tilcsik 2011; Gaddis 2015; Pedulla 2016; Rivera and Tilcsik 2016; Ameri et al. 2018; Neumark, Burn, and Button 2019).
Although no audit research is available on the effect of union affiliation on hiring in the US context, one Belgian paper studied employer responses to student applicants who disclosed they were members of a youth trade union (Baert and Omey 2015). The authors found an average negative effect of such disclosure, but in sectors with low union density no union penalty occurred in hiring. We build on this study to test the effect of union affiliation among working adults and to assess whether this penalty holds in the dissimilar US labor relations context.
Empirical Case: Non-college Jobs in Chicago
We test the effect of union affiliation on entry-level, non-college jobs in Chicago. As the third largest metropolitan statistical area in the United States, Chicago is a substantively important empirical case to estimate the baseline level of discrimination against union applicants. Chicago has a long history of labor organizing, and it remains in the 70th percentile of union density of all metropolitan statistical areas in the United States (Hirsch and Macpherson 2020). As such, we expect employers will be somewhat more likely to notice and understand union affiliation on a résumé than would employers in cities with very low union involvement. We restrict our focus to a single city because it allows us to develop highly realistic, tailored application materials (discussed in more detail below). While some audit studies apply to many cities across the country, simulating an applicant willing to relocate (Weisshaar 2018), this approach is implausible for the kinds of entry-level and non-college jobs we focus on here. In our concluding discussion, we consider how future research might study union discrimination in other cities.
We further restrict our study to entry-level, front-line positions that do not require a college degree. These types of jobs have suffered most from union decline, and they are of substantive interest to research on wage stagnation and upward mobility (Western and Rosenfeld 2011; Wilmers 2018). Within this category, we develop application materials for jobs in warehousing, restaurants, retail and grocery, front desk positions, security, manufacturing, customer service, and truck driving. We selected these specific industries and job types for a few reasons. First, these are common jobs representative of positions that do not require a college degree (Osterman 2020). Second, they do not require specialized prior experience or certification, as do many jobs in the health care, child care, and construction sectors (the main sources of non-college jobs that we exclude). Third, all of these industries face some plausible labor union activism, and employers hiring for these jobs are likely to at least be familiar with labor unions. 1 These selection decisions therefore balance representativeness of jobs against the realism of the experiment we conduct.
Correspondence Audit Methods
In the correspondence audit, we submit matched pairs of résumés to jobs, for a within-subjects design. 2 This approach allows us to estimate models with employer fixed effects, which compare employer responses to union and non-union applicants for the exact same firm and job. The downside of a within-subjects design is that it poses a greater risk that an employer realizes they are being audited. We mitigate this risk by randomizing the style, content, and order of the application materials we submit. In the following, we describe our decisions about the design of application materials and the specific procedures we followed to identify and apply to jobs. In the interest of encouraging labor relations scholars to replicate this research design, we have also posted all application materials, code used to automate portions of this process, anonymized data, and code used to analyze the data at https://osf.io/dqgm8/.
Résumé Design
Consistent with our within-subjects design, we submit two résumés to each job posting: a control—an applicant without any union affiliation—and a treatment—a résumé with signals of union affiliation. Figure 1 gives examples of résumé content, formatting, and union participation signals. Designing these materials raised two challenges: plausibly signaling union affiliation and creating résumés that are closely comparable but clearly distinct. We discuss each of these issues in turn.
Example Résumés
We signal union affiliation on the résumés in two ways. First, the treated résumé includes union affiliation via membership (signaled with either “Local 1, Service Employees International Union” or “Local 881, United Food and Commercial Workers International Union”). This signal appears directly below the employer name of either the applicants’ current or second most recent job. We vary the signal by current or prior job to ensure that employers are not passing on a union applicant because he might have a higher reservation wage for the prospective job. In addition to this main union signal, in a stronger treatment linked to one type of résumé content, we also include “History of active union participation” as a specific skill on the résumé.
We developed these signals of union affiliation as plausible additions to a résumé. Of course, most job applicants would not include explicit information about union affiliation on their job applications. But, these signals allow us to test whether employers, when given the opportunity, screen out union-affiliated applicants. In a regular hiring process, a hiring manager may uncover union affiliation by web searching an applicant’s name; inferring it from experience in a high union density industry or from prior employment at a prominently unionized competitor; or, later in the screening process, through responses to hiring questionnaires or interview questions. We provide a much easier signal by noting union involvement on the résumé, but expect that employers would respond to this disclosure similarly to how they act on union affiliation information obtained through these other means.
Our second challenge was designing résumés that are comparable but distinct. Specifically, we sought to 1) make résumés as comparable as possible and 2) avoid arousing employer suspicion that the résumés were so similar, which meant that résumé formatting and content had to differ enough to seem as though both résumés came from a distinct source and that résumés had to be similar enough not to swamp the potential effects of the union signal. So, we matched the applicants on educational attainment (they both have a high school degree); work experience (they both have three jobs in their job history with positions that steadily increase from back- to front-of-house work); race (both have likely white names); and gender (both have male names).
Because we focused on entry-level jobs that did not require a college degree, the high school each applicant graduated from is a salient credential for prospective employers. We chose two well-known Chicago high schools (Lincoln Park High School and Von Steuben Metropolitan High School) that had similar student demographics, similar numbers of students, were located in similar neighborhoods, and were not selective enrollment schools.
For work experience, we chose past work experience that included both local Chicago companies and national chains active in Chicago. Each hypothetical candidate had previous job positions as a cashier or sales team member, a package handler, and a bus cleaner or maintenance person. We identified job experiences that would broadly prepare applicants for any of the non-college jobs in the industries we targeted. But, in the interest of realism, we also sought to prevent over-qualifying applicants for the entry-level positions we focused on.
We also included on the résumés home addresses in similar neighborhoods and in similar (multi-family apartment) buildings. One address is in West Side and one in North Side Chicago, to both capture the geographical scope of Chicago and to avoid arousing employer suspicion by using the exact same neighborhood. The neighborhoods in which these addresses are located (Ukrainian Village and Logan Square) have similar sizable white populations and a roughly comparable reputation within the city.
Together, each of our two distinct versions of high school, prior work experience, and home address constituted a bundle of résumé content. We randomly assigned one version of résumé content to treatment and one to control for each job application.
Next, we designed two visually distinct résumé formats (see Figure 1), again aiming to avoid employer suspicion. These formats use different fonts, indentation, placement of dates, and header designs. These distinct formats make it unlikely that, despite the similarity of the underlying content of the treatment and control résumés, an employer would identify the two résumés as coming from the same source. We randomly assigned these formats across treatment and control groups, and did so separately from the assignment of résumé content.
To distinguish employer callbacks for each applicant, we permanently assigned names and contact information for the treatment applicant and the control applicant. The choice to assign permanent names and contact information reduced the risk that monitoring callbacks for each condition would suffer from human error. It also reduced the logistical complexity of setting up multiple email addresses, phone lines, and voicemail boxes for each name. We took several steps to ensure that both names carried similar connotations. We matched both treatment and control applicants as white men, to avoid measuring the effects of race or gender-based discrimination.
So we selected two common and predominately white male names: Brian Hartman and Steven Anderson. Both first and last names are among the most common names from the 1990 Census. Both names are predominantly white and relatively neutral in terms of class status (Gaddis 2017). We also checked, using Google, to ensure that searching online for these names in Chicago did not yield any objectionable or suspicious results. We pre-tested the names along with all application materials (discussed further below) and found that respondents perceived no demographic differences between them. As such, we randomly, but permanently, assigned Brian Hartman to the control condition and Steven Anderson to the treatment condition.
We then created email addresses and pre-paid Chicago-area cell phone numbers that we permanently assigned to each applicant. We also created separate usernames for each applicant on Indeed.com, the job-posting website we used to find and apply to postings. To avoid automated detection by Indeed.com, we used a separate, dedicated Chicago-based VPN for each applicant, which allowed each applicant to apply from seemingly distinct locations.
Cover Letter Design
In addition to a résumé, most jobs allowed submission of a cover letter. We wrote 14 generic cover letters. They were specific to each of our eight distinct target industries (we combined warehousing and manufacturing). Each industry had two cover letters, written in distinct voices to alleviate employer suspicion of the applications. These cover letters described general experience common across both résumé versions, so we could randomly assign résumé and cover letter content independently. We targeted different industries with different cover letters by discussing the industry in the cover letter and by emphasizing the aspect of each general entry of work experience potentially relevant for entry-level work in the target industry. Examples of the cover letters, for the retail industry, are included in Figure 2.
Example Cover Letters (Retail Industry)
We randomly assigned one version from the relevant industry to treatment, and one to control, prior to applying for a job. In treatment cover letters, we also included a positive, but reasonable, mention of union participation: “Also, my experience being in a labor union has greatly helped my organization skills.” For each application, we then filled in the relevant company name in the cover letter, derived from the job posting, along with the contact information for the applicant. The cover letters were pasted into the application website and therefore did not need to be distinguished by different formatting (unlike the résumés).
Testing Strength of the Union Signal
A key question for our research design is whether we reliably signal union affiliation with our application materials. This concern is difficult to test during the audit itself, as surveying employers would tip them off to their involvement in our study. Instead, we conducted two rounds of experimental pre-tests, using respondents from Amazon’s Mechanical Turk (MTurk) platform, to ensure that the signal of union affiliation was salient enough to be recognized by employers. 3
In the first round of pre-testing, we signaled union affiliation in several ways to test whether adding more signals (e.g., an explicit statement of union participation in addition to a local number) increased the strength of identification. We randomly assigned to each respondent either a control résumé, a weak-signal résumé (with a local number and union listed under one job experience), or a strong-signal résumé (the weak signal, plus an explicit statement of active union participation). We then asked respondents the extent to which they agreed or disagreed (on a 7-point Likert scale) that the candidate was a union member. The strong-signal résumé was more likely to be recognized than the weak-signal one: For the strong-signal résumé, 84% of respondents agreed (somewhat, agreed, strongly agreed) that the applicant was a union member, compared to 44% of respondents for the weak-signal résumé (and 13% for the control). Based on these results, in the audit we randomly assigned either a weaker- or a stronger-signaled résumé, as well as a résumé with union affiliation listed in the current and penultimate position. We included these four variations to ensure that our results were not sensitive to small differences in signaling approach.
In a second round of testing, we presented our final materials, including both résumé and cover letter, as a full application for respondents to assess. 4 We asked respondents both whether they thought the applicant was a union member and whether they thought the applicant supported unions. Asking about union support in addition to union membership is important to ensure that employers do not read the union-signaled résumé as reflecting someone who wants to leave a union job and therefore likely does not support unions.
Figure 3 shows the results. Panel (a) shows that 80% of respondents assigned the treatment application agreed that the candidate was a union member. By contrast, only 20% of respondents randomly assigned to the control résumé agreed with the same statement. Panel (b) reveals a similar pattern for respondent perceptions of the applicant’s support for unions: 83% of respondents assigned the treatment application agreed that the treatment candidate supports unions. This test establishes that respondents do not read the treatment application as reflecting involuntary or incidental prior union membership, but instead identify it as an application submitted by a union supporter.
MTurk Respondents Recognize Signals of Union Affiliation
We also asked MTurk respondents if they had ever worked in a hiring capacity (40% had). Those who had worked in a hiring capacity were slightly more likely to recognize the treatment résumés as indicating union affiliation: for union membership, 80% without hiring experience compared to 84% with hiring experience; for union support, 80% without compared to 87% with. Although not large differences, these results indicate that respondents with hiring experience are at least as likely and possibly more likely to recognize the union signals in the application materials. Taken together, these results give us confidence that our union treatment applications do indeed signal union affiliation and support. If 80% of MTurk respondents recognize that a treated application indicates union support, we expect that hiring managers who are actually hiring for frontline positions would also categorize the treatment application as a union supporter. 5
Application Procedures and Analysis
From October 2019 to October 2020, we submitted 1,025 résumés to 511 and 514 job postings on a large job posting website in the metropolitan area of Chicago. 6 We chose an online rather than an in-person design because most jobs are advertised online (Carnevale, Jayasundera, and Repnikov 2014). Each week, a research assistant searched Indeed.com (a job-posting website) for jobs in warehousing, restaurants, retail and grocery, front desk positions, security, manufacturing, customer service, and truck driving that were posted within seven days and 25 miles of the city center of Chicago.
We searched for jobs manually rather than relying on an automated script to ensure the job postings that most closely fit our criteria were selected for application. We selected jobs that did not re-direct to a third-party website, that were for full-time employment, and that did not generate from staffing or temporary help agencies. We also excluded positions that required a college degree, more than two years of relevant experience or any special licenses or certifications (except Commercial Driver Licenses for trucking positions).
To avoid raising employer suspicion, we applied to a given employer only once during the course of the audit, even if the same employer posted multiple relevant jobs. To ensure we did not apply to the same job twice, we cross-checked postings each week with the employer names of jobs that had been previously collected. To maintain consistency without compromising our matched within-subjects design, we excluded jobs that required long questionnaires in addition to résumés and cover letters. When jobs required a few short screening questions—such as those about capacity to lift, whether an applicant’s age is 21 or older, or whether the applicant owns a car—we applied and answered these simple questions identically across treatment and control applications.
Once we selected a sample of jobs for the week, we randomly assigned the résumé and cover letter format and content for each job (as discussed above). Each condition was then randomly assigned to apply to a given job either first or second, and we control for this difference in subsequent analyses. We waited one day between job submissions: On Monday and Tuesday, new postings were collected by the research assistant. We then submitted the first application to a job on Wednesday and the second on Friday. As such, in some instances a job posting was removed between one day to the next, causing slight variation in the number of jobs covered by control and treatment applications (511 and 514, respectively).
After submitting the résumés, we tracked whether both, neither, or one of the two candidates received a callback with either an email or a voicemail message. We coded a callback as any favorable response including a job offer or invitation to interview.
We coded no callback as either an outright rejection or no response. In addition, we coded another variable which had a more sensitive threshold for a favorable response, such as a personalized message or which included specific details about the applicant. We did this as a way to track callbacks that indicated higher engagement with the application.
After coding these callback variables, we use a simple regression framework to predict callbacks with the treatment, controlling for application order, résumé substance (the specific job experience listed in the résumé), résumé format, and cover letter version, as well as with and without employer fixed effects. 7
Table 1 shows that the treatment and control résumés were balanced across characteristics. Consistent with our repeated randomization approach, the two applicants each submitted half their application materials in one format; and half the applications submitted by the treatment condition reflected a résumé with the additional strong signal of union participation. They applied evenly on the first or second application day. The treatment and control conditions were also balanced in terms of the occupations to which they applied. The largest share of jobs applied to by each candidate was in restaurants (almost 30%) and as front desk workers (slightly more than 20%), followed by warehouse jobs (slightly more than 15%) and customer services jobs (approximately 11%).
Descriptive Statistics
Notes: Data gathered in original correspondence study. SD, standard deviation.
Table 1 also shows the overall callback rate for each candidate. Both candidates received a callback rate of approximately 15%. This callback rate is somewhat higher than found in previous correspondence audit studies (Gaddis 2015; Pedulla 2016). This outcome could reflect the tight labor market that prevailed during the first half of our study, or the effect of our industry-specific cover letters and locally relevant résumé materials.
Figure 4, panel (a) shows the frequency of applications submitted during the 13-month period in which we conducted the study. Note that our year of data collection reflects ebbs and flows in the labor market during several periods. First, collecting data after the holiday season into the beginning of the new year of 2020 reflects a lower level of job-posting submissions (the vertical bars) than in the fall of 2019. Second, we collected data before and after cities issued stay-at-home orders due to the COVID-19 pandemic in April 2020. Evident in panel (a) is that job postings decreased sharply during this period. Prior to starting the audit, we set no timeline for completion, but instead targeted a sample size of 1,000 jobs based on a power calculation (assuming a 12% control callback and a 7% treatment callback). We ended the audit during the week that we met this predetermined sample size.
Summary of Applications and Callback Rates
The lines in panel (b) of Figure 4 reflect each applicant’s quarterly callback rate over the study period. In the fall and winter of 2019, the union applicant had a slight advantage over the control applicant. After COVID-19, however, the applicant who signals union participation appears to face a small penalty in callback rates. This difference, however, is not statistically significant. The average difference across the time period remains quite small (less than .01%). This descriptive evidence suggests that no hiring penalty is associated with union affiliation in our sample.
Results
Audit Results
Table 2 shows in a regression framework similar results to those in Figure 4. We first estimate the baseline effect of union affiliation on callback rates. Model 1 shows a precisely estimated zero effect: Callbacks for the union-affiliated applicant are no different from the control. Model 2 shows that this null difference is unchanged after conditioning on differences in application substance or résumé format. While the fancier résumé format receives slightly higher callbacks, controlling for application characteristics makes no difference to the estimated union effect. The null effect is also robust to whether the treatment or control condition applied for the job first or second. This insensitivity to controls results from the random assignment of these application characteristics and, as Table 1 shows, their balance across treatment and control.
Effect of Union Affiliation on Callback Rates
Data Source: Union Correspondence Study.
p < .05; **p < .01; ***p < .001 (two-tailed tests).
Model 3 shows that this null effect persists with the inclusion of employer job-posting fixed effects. Even comparing only among jobs that received both applications but responded to only one, the union résumé was not significantly different from the control. Finally, model 4 uses the strict callback indicator as its outcome and shows still no significant difference in callback rates even across this more sensitive threshold.
Table 3 shows a series of interaction effects defining subgroups of applications and jobs that are more likely to yield a union penalty in callback rates. First, model 1 tests whether résumés with the strong union signal showed a more negative effect than the weaker signal. If this were the case, then it would suggest that employers are not parsing the weak union signal, but are reacting to the stronger signal. As Model 1 shows, however, no additional negative effect occurs for the strong-signaled union affiliation résumé. If anything, the strong-signal union résumé has a slightly higher callback rate than the weak-signal union résumé. This finding, along with the MTurk tests discussed above, assuages concerns that our union signal was simply too subtle for employers to respond to. If employers frequently would discriminate against union affiliates, but simply miss the résumé signal, the discrimination effect should become increasingly negative. The null result for the strong signal subset suggests that, in context of our pre-testing, employers notice the résumé signal but are not more likely to discriminate against the stronger-signaled résumé applicant.
Heterogeneity in Union Affiliation Callback Effect
Data Source: Union Correspondence Study.
Notes: CTU, Chicago Teachers Union.
p < .05; **p < .01; ***p < .001 (two-tailed tests).
Next, model 2 in Table 3 assesses whether our main null result is driven by industries and neighborhoods with little union activity. We match National Labor Relations Board (NLRB) union certification elections and Federal Mediation and Conciliation Service (FMCS) collective bargaining announcements to the industry and Census tract in which our audited establishments are located. We use NLRB and FMCS data from the two years prior to our audit (2018 and 2019). The results show that even in industries and neighborhoods with union activity, there is a null result on anti-union discrimination. Similarly, in model 3, we also exploited variation due to a strike by the Chicago Teachers Union (CTU) in October 2019. High-profile teacher strikes can increase salience and support for unions (Hertel-Fernandez, Naidu, and Reich 2019). Yet, the null result holds whether the job was posted before or after the strike announcement.
Finally, in model 4, we test whether the union discrimination result varies by firm size. Smaller firms may be less threatened by union-affiliated applicants than are larger firms that are more likely to be targeted for an organizing drive. Although larger firms are slightly more likely to call back the control than the treatment applicant (0.008 + –0.039), this difference is not statistically significant. As such, larger firms may indeed be somewhat more likely to have union prevention screening in place.
Taken together, however, the main outcome of these subgroup analyses is that they support a consistent null effect of the union affiliation signal.
Null results are difficult to interpret, because they may result from an underpowered experiment, and researchers frequently do not submit them for publication (Franco, Malhotra, and Simonovits 2014). Prior to conducting our audit, we pre-determined a target sample size based on the magnitude of the union penalty we expected to observe. After conducting the audit, we assembled several similar audit studies that found callback penalties for various types of applicants. The majority of studies were identified using Baert’s 2018 paper on all audits conducted within the United States from 2005 to 2016 (Baert 2018). We then searched additional papers from 2016 to 2021, all of which were conducted within the United States and all of which used a signal in résumés or cover letters and followed procedures broadly similar to our audit. These studies provide a useful benchmark for assessing the null result we estimate, so in Figure 5 we compare our union signal estimate to results in these other studies.
Comparing Union Signal to Penalties for Other Résumé Signals
Two key patterns stand out in Figure 5. First, our coefficient is estimated as precisely or more precisely than most of the prior studies we identify. So, in addition to having a near-zero point estimate, our result is relatively precisely estimated. Second, the confidence intervals around our estimate allow us to rule out penalties on the order of the callback penalties associated with other applicants, including those who are Black, gay, from a low socioeconomic class background, unemployed, or having nontraditional prior work experience. This comparison suggests that our finding is a substantively meaningful null result, rather than the product of an underpowered study.
Supplementary Findings
The audit results indicate that Chicago employers of entry-level, non-college workers do not consistently discriminate against union affiliates in the first stage of the hiring process. Although audit studies are useful for isolating the causal effect of union affiliation, they give no context for employers’ decisions. For these reasons, to account for this audit result, we turn to supplementary data from social surveys and informant interviews.
First, the General Social Survey (GSS) has tracked distrust of union leaders over time. Although social surveys have problems with desirability bias, they can be useful for illustrating the general context of union sentiment among employers over a long timeframe. In one question, the GSS asks: “As far as the people running these institutions are concerned, would you say you have a great deal of confidence, only some confidence, or hardly any confidence at all in organized labor?” We tracked the answer to this question among all respondents, union members, and managers from 1970 to 2020. We define managers broadly as any manager, supervisor, or self-employed respondent.
Figure 6 shows that managers’ distrust of union leaders has been declining since 1990. The high point of anti-union sentiment among managers in the late 1980s likely reflects the peak of intense employer anti-union activism. The subsequent decline in anti-union responses (from a peak of 50% to a nadir of 30%) provides suggestive evidence that manager views of unions have shifted. Note that this trend is present for other respondents in the GSS, but shows the starkest shift for managers. Whereas in the 1980s, half of managers were highly anti-union, today that number has nearly halved to one-third.
How Confident Are Managers in Labor Leaders?
Second, from April 2020 to February 2021, we interviewed 20 hiring managers to understand whether and why union affiliation means little in the hiring process. To find informants, we used the Dun & Bradstreet database of all establishments in Chicago, and then we selected a random sample of 800 establishments in the same industries as our correspondence study. We then cold-called 341 of these establishments to ask questions about recruitment and hiring for entry-level workers, as well as experiences with unionization or pro-union applicants. Because the interviews were all conducted after the stay-at-home orders associated with COVID-19, we also asked informants whether COVID-19 changed any hiring practices for entry-level workers, as well as what the hiring practices were like prior to the stay-at-home orders. The vast majority of businesses we called did not answer or declined to participate. We interviewed six informants in hotels, four informants in retail services, four in transportation or security, three in manufacturing, and three in restaurants.
Managers, employers, and human resources managers we spoke with insisted that, whether in large or small establishments, information about labor union participation has little impact on hiring decisions. For small establishments, this is because of a very low likelihood of workers unionizing to begin with. One small café and meat market owner, for example, said that he only has a few people working for him and he has grown to know them personally. For this employer, the likelihood these few employees would organize was small. A human resources manager of a manufacturing firm similarly said, “Our place is so small that we never have any problems or anything. They wouldn’t be able to unionize anyway since they’re temp workers.” Although she acknowledged that unionization could pose a potential “problem” in theory, her small and temporary workforce posed little threat of unionizing.
Yet even in large establishments with a higher likelihood of workers organizing, employers saw hiring union supporters as little threat to their business. One human resources manager of a large manufacturing establishment, the majority of whose workers have long been unionized, expressed the sentiment that “unions are dying, slowly and surely. It’s really a matter of time.” He went on to say that the union representing employees at his establishment had lost money in the most recent recession and suspected similar financial constraints following COVID-19.
In large establishments in which workers were not unionized, employers did not interpret union support as meaningful in the hiring context. One hotel hiring manager explained that she does not “really pay attention to union experience,” while another hotel manager said that because the hotel is non-union, if she were to see “that someone did have union organizing experience, it would not mean much.” These comments suggest that union support means little to employers at the hiring stage. Instead, employers emphasized professionalism and reliability in hiring. In retail and food services, meanwhile, some hiring managers reiterated that unions could pose a “problem,” but that a long non-union history in a company makes subsequent unionization very unlikely. One branch manager in the retail services, for example, said: It wouldn’t mean that much to us because of what it is, you know we’re non-union. So I get what they’re coming from like, in their last job ok, union yes they’re defended, you know, they’re covered, they have all the rights. Us, yeah, I mean we’re retail, we’re non-union, so we give them that deal like basically so it’s discipline or develop.
Even though this manager linked union presence to greater workplace protections, he reiterated that hiring a union employee would pose little threat to his company’s well-established, non-union work culture. This culture was so entrenched in the store branches that he expected hiring workers with prior union experience would have little effect.
In sum, the social survey and interview data provide some context for why employers might not discriminate against union-affiliated applicants at the point of hire. Employers report less anti-union sentiment than in prior decades. When asked directly about how they would react to seeing union affiliation on a job application, our respondents emphasize that union weakness and obsolescence makes prior union experience irrelevant.
Discussion
In this article, we conduct the first-ever US correspondence audit study testing whether employers discriminate against union applicants. We find no detectable penalty associated with prior union affiliation on hiring callbacks. Whether using a broad or a sensitive threshold for defining a callback, the union-signaled candidate faced no decreased likelihood of callback relative to the candidate without that signal. This null result holds even with a stronger union signal, with employers in neighborhoods and industries with direct exposure to union organizing or bargaining, and in the period following a high profile citywide teachers’ strike. Our supplementary data suggest that employer anti-union sentiment has declined over time and that employers see unions as largely irrelevant (rather than threatening) to their businesses. In context of the long decline of US labor unions (Western and Rosenfeld 2011; Wilmers 2019), employers in our sample do not filter out, at the initial stage of application, union-supporter job applicants.
Our analysis has several limitations. First, we submit applications and track callbacks, but observe nothing about the employer-side review and decision-making process. Based on our online survey validation results, we are confident that the union signals used are noticeable and legible to a range of respondents. We cannot be certain, however, that employers are carefully reading and interpreting résumés at the initial review stage that we study. This limitation is part of all audit studies, but it is of particular concern with our topic. While race and gender are salient considerations (whether positively or negatively) in the hiring process, union support is typically harder to infer from a résumé. It is possible that our results reflect reviewers’ lack of attention to the union signal, rather than nondiscrimination. This would still be evidence for our general interpretation that employers do not consistently practice baseline anti-union hiring discrimination. However, it would limit the extent to which these results shed light on types of preventive employer anti-union discrimination outside of the hiring context.
Perhaps more important, we do not know exactly who is reading the applications we submit. Employers using third-party human resources services for application review could face an agency problem in weeding out union-involved applicants, which in turn could account for our null results. We exclude postings by temporary help and other labor intermediaries to mitigate the risk that we would submit applications to third-party sources. In an era of outsourcing and contracting though, it is still possible that some employers in our sample used third-party application reviewers. As an additional check, we coded callbacks according to the position of the caller (if identified) and found that among callers with identifiable job titles (32%), 27 were in human resources and 22 were managers. While initial screening could still be done by third-party contractors, we were reassured that none were explicitly identified in our callbacks. As such, we interpret our results as generated by a mix of human resources and line manager assessments that reflect in-house employer review practices.
A third limitation, of generalizability, suggests exciting research opportunities for labor relations scholars. Our design focuses on young, likely white, male applicants applying to entry-level, non-college jobs in Chicago. Recent movements such as Black Lives Matter and #MeToo demonstrate that organizing often centers around multiple axes of identity. As such, employers’ hiring decisions may depend on how they interpret applicants’ prior union experience alongside their race, gender, sexual orientation, or migration history. Future research could borrow our method to test whether these ascriptive characteristics interact with union affiliation to change outcomes for applicants.
Future research could also consider whether sharper anti-union sentiment exists in more politically conservative cities, or, conversely, in cities with even stronger and more active labor unions than in Chicago. Labor relations researchers could also audit higher-skilled jobs, such as nursing or teaching, that currently face stronger union activism than the low-end jobs we audit in this study. Perhaps most promising for future research, our results suggest that a union penalty in hiring may have emerged during COVID-19 (although this penalty was not statistically significant). We ended the audit once we reached our pre-committed sample size; however, this pattern leads us to speculate that, as with many other worker characteristics, employers may discriminate more against union supporters during periods of high unemployment and labor market weakness. Likewise, large firms may be slightly less likely to call back union-signaled applicants (but again, we found no statistically significant difference in our sample). By varying worker identity, geography, employer and occupation types, and economic conditions, future research can use our study as a baseline point of comparison.
Aside from the specificities of geography, job types, and economic context, we conduct our audit in an era of labor union weakness. This circumstance is likely critical for explaining our results: Unlike previous historical periods, labor unions are not a pressing concern for most employers outside of rare organizing drives. Beyond point of hire discrimination, employers have a variety of other options for preventive union avoidance strategies. Some of these strategies may focus narrowly on heading off union interest, such as showing anti-union videos during onboarding (Greenhouse 2015). But other strategies have meant real gains for workers. The threat of union organizing gave rise to versions of responsive management practices throughout the 20th century, ranging from welfare capitalism (Jacoby 1997) and non-union personnel departments (Foulkes 1980) to the human resources management model (Kochan et al. 1994).
While we cannot compare results directly between our study and those that would have been obtained during the 1980s—at the height of the early employer counteroffensive—we expect that our results show that preventive union avoidance has been deprioritized by employers. If employers fail to take advantage of an easy opportunity to sift out union supporters at the point of hire, it is likely that union power has fallen below the level at which its threat can spur these more costly and wide-reaching pro-worker management practices among non-union employers.
Related to union weakness, labor law currently allows employers a series of choke points at which they can stop organizing drives, prior to signing a first collective bargaining agreement (Ferguson 2008). And unions almost always organize to represent a majority of workers in a bargaining unit. Multiple NLRB rulings have found that employers retaliate against union supporters by saying “one bad apple spoils the basket” (e.g., Rogers Electric 2006). But, our interviews suggest that the requirement that labor unions receive majority support before legally imposed bargaining obligations makes employers less concerned about hiring a single union supporter into an otherwise non-union shop. Some commentators advocate abandoning exclusive representation and pursuing minority unionism (Harcourt, Lam, and Wood 2014). If this strategy takes hold in US labor unions, it may strengthen incentives for the preventive anti-union discrimination we test for here.
Given the nearly half-century-old employer offensive against unions, one could expect (and indeed, we did expect) that employers would practice discrimination against union applicants when given an opportunity to do so. However, maintaining preventive anti-union discrimination, outside of active organizing drives, requires prioritization and attention. Many employers have little contact with unions, and in the case of a rare organizing drive, employers often outsource their response to anti-union consultants, rather than maintaining in-house labor relations specialists. The contemporary weakness of labor unions may have allowed preventive anti-union discrimination to slip off the agenda of many employers. This ebbing of employer anti-union vigilance could be an opportunity for labor union growth and resurgence.
Footnotes
Author order is randomized. Thank you for helpful advice from Virginia Doellgast, J.P. Ferguson, Tom Kochan, Lisa Lynch, David Pedulla, Adam Reich, Adaner Usmani, Kate Weisshaar, Alix Winter, and anonymous LERA reviewers. Elijah Ruiz provided excellent research assistance. This research was funded by MIT Sloan School of Management.
For information regarding the data and/or computer programs used for this study, please address correspondence to
1
In warehousing, Teamster Locals 743 and 703 have long organized in Chicago, and more recently Warehouse Workers for Justice has sought to mobilize previously non-union warehouse workers in the broader Chicagoland region. For restaurants, UNITE HERE Local 1 represents a small number of legacy union restaurants, but the Fight for $15 and the Restaurant Opportunities Center Chicago have also recently led high-profile campaigns against a broader set of food service employers. Among front desk positions, UNITE HERE Local 1 represents many hotels, and other unions represent a smattering of front desk workers (particularly in health care facilities). In retail and grocery, UFCW Local 881 and other unions represent workers across several large chains. A variety of unions, from the autoworkers to the steelworkers, represent and actively organize manufacturing facilities in the Chicago region. For security, SEIU Local 1 represents thousands of security workers across multiple employers and two large collective bargaining agreements. Several Teamsters locals represent local delivery truckers and UPS drivers in Chicago (though there is very little union activity in long-haul trucking positions that may be based out of Chicago). It is least clear whether customer service positions face much union activity, as they are scattered across many industries, with and without union presence. Overall, however, all of these job types face at least some union activism in Chicago.
2
Our research design was reviewed and approved by the MIT Committee on the Use of Humans as Experimental Subjects.
3
In both rounds, we started with 250 responses and dropped those who failed attention-check questions (50 to 75 respondents in each sample).
4
We also tested whether listing the union affiliation in either the current or penultimate job varied signal salience. Although listing the union affiliation in the current position might increase signal identification, it could also confound the results if employers interpreted current membership as raising an applicant’s reservation wage. We found no difference in responses by the order in which the union job was listed.
5
To test the other elements of our application materials, we asked respondents to rate the extent to which they agreed that the résumé belonged to a candidate who had a high school degree; worked at one of the large national chains; was white, Black, or Latino; was native- or foreign-born; was male; and was working, middle, or wealthy class, beyond asking about the union signal, which we describe in a separate subsection below. As with the union questions, these characteristics were measured on a 7-point Likert scale. All résumés were equally likely to be rated as male (1.75/7) and white (3.20/7). Additionally, they were all rated with a strong likelihood of belonging to a candidate who is working class (2.55/7). The résumés were also strong signals of the other characteristics, including educational attainment and work experience.
6
N of control and treatment conditions vary slightly because some job postings were taken down during the 24 hours between submission of the first and second application. All results are robust to including only the jobs for which we successfully submitted both applications.
7
We fit a linear probability model, regressing the binary outcome CALLBACKij for application i and job posting j on treatment UNIONij: 
, as most prior audit studies used this specification.