Do Job Applicants Consider Founder Race and Gender? Evidence from a Field Experiment

Do jobseekers consider the race or gender of a firm’s leadership when applying for a job? Prior research on inequality has found that race and gender influence nearly every economic exchange, with particular attention paid to how they affect employer behavior (Castilla 2008, Tilcsik 2011, Kang et al. 2016, Rivera and Tilcsik 2016, Leung 2018, Leung and Koppman 2018, Kuppuswamy and Younkin 2020). However, the conditions that typically encourage employers to discriminate—the need to make consequential choices under uncertainty and with imperfect information—are also present for job applicants. As a result, it stands to reason that a prospective employee might also factor the race or gender of people they would work for into their decisions.

The question of whether the leadership’s race or gender matters to jobseekers is particularly salient for small or entrepreneurial firms. First, the information asymmetries that plague a typical job search can be exacerbated for these firms, making applicants rely on noisy cues when making an assessment about the job. In particular, the identity of new firms is often inextricably linked to that of their founders (Navis and Glynn 2011, Zuzul and Tripsas 2020), raising the founders’ profile and reducing the likelihood that their race and gender goes unnoticed by applicants. Prior research has shown that this increased attention contributes to the challenges that female and minority founders face in raising capital (Chatterji and Seamans 2012, Younkin and Kuppuswamy 2019). Given that entrepreneurial success depends upon both the acquisition of capital and talent (Wasserman 2012, Roach and Sauermann 2015), understanding the behavior of applicants is critical to identifying the degree of disadvantage that underrepresented founders confront.

Although we have extensive research on employer-side discrimination (Pager and Shepherd 2008) and the cost of these biases for both firm and individual, we know less about employee-side biases and their consequences. At first glance, it may seem unlikely that a jobseeker would even notice the gender or race of a founder, let alone factor that into their decision. After all, applicants have limited attention, and it is likely that a founder’s attributes would be buried under more pertinent variables—pay, location, opportunities to advance, or firm prospects (Barbulescu and Bidwell 2013, Cappelli and Hamori 2014). The complexity of the calculus by which an applicant selects a position could therefore obviate any consideration of these demographic characteristics. But the same should be true for employers and investors seeking for employees or funding opportunities, and we know that they are influenced by social categories. Moreover, research suggests that candidates consider a diverse range of factors in their search, including the social (Turban and Greening 1997, Burbano 2016, Carnahan et al. 2017) and political orientation of the firm (Bermiss and McDonald 2018, Carnahan and Greenwood 2018). The question is therefore whether a founder’s race or gender plays a role in the job search process.

Building upon recent work on discrimination in labor markets, we propose three reasons for why job candidates may attend to a founder’s race or gender. First, studies of supply-side processes have established that applicants are highly attuned to their potential fit with an organization (O’Reilly et al. 1991, Fernandez-Mateo and Fernandez 2016). Gender and race are important factors shaping assessments of potential fit; individuals anticipate better fit with people they perceive as sharing their traits (McPherson et al. 2001). Second, research on demand-side processes has documented that similarity in race and gender influences both job offers and opportunities for promotion (Gorman 2005, Kurtulus and Tomaskovic-Devey 2012). Thus, jobseekers may be wary of entering an environment in which they are the local minority, either because they anticipate a challenge “fitting in” or because they anticipate that organizational opportunities will flow through homophilous ties. Therefore, to the degree that homophily influences presumptions of fit and opportunity, we would expect the applicant pool to reflect the attributes of the founder. In a context where most jobseekers are White men, we would expect women and minority founders to get fewer applications compared with White male founders.

Third, jobseekers may use a founder’s race or gender as a proxy for future performance of the firm or as an indicator of job stability. Predicting the future is particularly challenging for entrepreneurial firms that lack clear evidence of their past performance. When quality is hard to measure, people often rely on race or gender when assessing the competence of the person or the viability of their business idea (Thébaud 2015a, Pedulla 2016, Rattan et al. 2019). Prior research does not simply find that evaluators consider a founder’s race and gender in their assessment of a venture’s potential; they repeatedly find that women and minority-backed ventures are devalued (Chatterji and Seamans 2012, Thébaud 2015b, Lee and Huang 2018, Younkin and Kuppuswamy 2018). Therefore, to the degree that women and minority founders are considered less competent and their business ideas less promising, we would expect the candidate pool for White men to be larger and of higher quality.

The dearth of causal evidence for supply-side biases partially reflects the challenge in designing an appropriate empirical test using observational data. For instance, if women or minority founders operate in different types of industries, offer different types of benefits or compensation structures, or even use subtly different language in job postings, then variation in the applicant pool may reflect these organizational or positional differences and not any discriminatory actions. To avoid this, we conducted a field experiment employing a two-step “reverse audit” design (Flory et al. 2015, Hedblom et al. 2019), where we partnered with a startup advertising real, full-time positions on an online job-hosting platform but were able to randomize interested jobseekers into three conditions that varied the apparent race and gender of the founders. This design allowed us to test how the race or gender of the founder affected the likelihood that a given applicant completed their application, holding all aspects of the position and firm constant.

Our results indicate that although applicants were no less likely to apply to jobs in companies with White female founders than companies with White male founders (nor did they request different salaries), applicants were significantly less likely to apply when shown a posting with Black founders and requested a 10% increase in salary (equivalent to $10,000/year) when they did apply. This general pattern was amplified for more qualified candidates, such that the more qualified a candidate was, the less likely they were to apply to a firm with Black founders. We supplement the results from the field experiment with two survey-based experiments to help distinguish between several potential explanations for our findings. These supplemental analyses provide evidence that the penalty is unique to White applicants evaluating Black founders and that it reflects at least two concerns: (1) that a White applicant will be less likely to fit within a firm with Black founders and (2) that a Black-founded firm is less likely to succeed. We also find that White respondents anticipate that working for a White founder will pay more and look better on their resume than working for a Black founder. In contrast, and contrary to our expectations, we do not find any evidence that applicants are less likely to apply to firms with female founders, nor do the survey experiments indicate that people think female-founded firms are more likely to fail. Instead, the evidence is specific to racial minorities. Together, our findings help to identify a previously unrecognized challenge that puts minority founders at a significant disadvantage relative to White founders.

Theory

Research on the effect of bias on hiring typically focuses upon how employer preferences affect the opportunities afforded to different groups of applicants (Tilcsik 2011, Kang et al. 2016, Leung and Koppman 2018). Although the underlying motivation and salient trait (e.g., race, gender, status) may vary with the context, there is consistent evidence that female and minority candidates are disadvantaged at hiring. Historically, frustration with similar structural impediments has served as a spur for entry into entrepreneurship (Sørensen and Sharkey 2014, Thébaud 2015a). Rather than accept the status quo, individuals whose talents are discounted in the labor market may leave it, in the hopes of creating more equitable opportunities for themselves. Here, we investigate how those entrepreneurial aspirations may be derailed by an unrecognized obstacle—embedded in the behavior of job applicants, rather than employers.

Hiring is a two-sided market, in which both employers and jobseekers must agree upon any potential match. We propose that applicants are no less subject to the biases that emerge when making consequential decisions using imperfect information than prospective employers. In particular, we predict that applicants sort into jobs based not merely on their fit with the specified job criteria, but also based on the known attributes of the founders. Below, we draw upon research on labor market sorting and on racial and gender biases in entrepreneurship to explain the rationale behind our predictions.

How Prospective Employees Assess Job Openings

Prior research on labor markets suggests that job candidates focus on two broad sets of factors when considering employment opportunities: financial concerns (e.g., salary, benefits, opportunity costs) and interpersonal ones (e.g., culture, organizational fit). Although the influence of financial incentives is understandable, the influence of nonpecuniary factors is also considerable; applicants have been found to accept lower salaries in order to work for organizations with a social mission (Abraham and Burbano 2022) and to stay longer at firms that share their political ideology (Bode et al. 2015, Bermiss and McDonald 2018). Research into hiring therefore emphasizes the role of both financial enticements and person-organization fit in determining applicant behavior.

And, although these considerations could arise at any stage in the job search process, there is increasing evidence that they exert a robust influence over the very decision to apply in the first place. Research on supply-side processes has documented labor market sorting across occupations, organizational levels, and industries (Barbulescu and Bidwell 2013, Fernandez and Campero 2017, Campero 2021). Notably, job applicants have been shown to refrain from even applying to jobs or occupations when they perceive a lack of fit, irrespective of whether they are otherwise qualified. For instance, startups have been found to employ higher percentages of young employees than older firms (Ouimet and Zarutskie 2014), and one explanation for this is that older candidates are averse to the volatility of startup life. Similarly, firms with more female founders attract more female applicants (Campero and Kacperczyk 2020), suggesting that people prefer to work with people who look like them. One consequence of this self-selection is that two different founders may face two very different applicant pools for otherwise comparable positions (Leung and Koppman 2018, Campero and Fernandez 2019).

Rather than pursuing a volume-based strategy, whereby candidates apply to as many positions as possible to find a match, many candidates weigh the potential costs (e.g., writing a cover letter, tailoring their resume, setting aside time for an interview) against the likelihood of an appealing offer. This calculation involves making predictions about the likelihood of securing an offer and the desirability of the job itself. However, just as employers struggle to determine which applicants are genuinely talented, applicants possess few means of testing firms’ claims regarding culture and opportunities.

This problem is less acute when applying to established firms, because candidates have access to multiple sources of information: public reputations, personal contacts, and an army of current employees posting reviews to Glassdoor or Reddit. Although no one source is perfect, the variety of information helps mitigate uncertainty regarding either the cultural or financial aspects of the position. A candidate focused on how the firm treats women could find multiple accounts from female employees to weigh against the firm’s public statements. By contrast, small and new organizations offer far less information to prospective employees—they are unlikely to have an established reputation, and the limited number of employees means that networked information is likely scarce—and, as a result, candidates are prone to disproportionately value the information they do have. In particular, because a startup’s identity is so often entwined with that of the founders themselves (Fauchart and Gruber 2011, Zuzul and Tripsas 2020), it stands to reason that an applicant would be especially attentive to observable differences between founders. Moreover, just as employers become more likely to rely on noisy heuristics when they lack more detailed evidence of desired skills (Heilman et al. 1988), we propose that candidates considering small or new firms will treat a founder’s race or gender as a proxy for valued but unobserved qualities like the firm’s culture and future prospects.

Founder Attributes and Applicant Re-Assessment

The question then becomes: how might a founder’s race or gender affect an applicant’s inferences about a given organization? We begin with financial incentives, which seem, on the surface, unlikely to change with founder demographics. In many established organizations, the salary, benefits, and range of bonuses may even be fixed at the time of hire or fall within a known range. However, startups are exceptional because many employees accept lower guaranteed salaries in exchange for a potential payoff if the firm succeeds (Sorenson et al. 2021). Moreover, startups have a significantly higher likelihood of failure than established firms (Freeman et al. 1983), and, therefore, predicting future salary also requires an assessment of the firm’s future. As a result, an applicant’s evaluation of the financial merits of working for a startup will be closely tied to their prediction regarding the firm’s future (Ouimet and Zarutskie 2014).

And, although there is no evidence that women or minorities are less able to create profitable or high-growth firms, research consistently finds that evaluators perceive firms with female (Thébaud 2015b, Lee and Huang 2018) or Black founders (Chatterji and Seamans 2012, Younkin and Kuppuswamy 2018) as lower in quality and less likely to succeed. There are two principal explanations for this. First, devaluations often reflect the cost of role incongruity, where the firm is seen as less promising because the founder does not match the common stereotype of a successful founder. For example, Thébaud (2015a) shows that investors perceive business plans as less promising or novel when they are believed to come from female founders, in part because they see innovation as male-typed. Similarly, Lee and Huang (2018) find that female founders of socially oriented ventures are not penalized (whereas founders of purely commercial ventures are) because the alignment between stereotypes of women as “warm” and a prosocial objective mitigates the incongruity of a woman as a founder. In both cases, there is a prevailing assumption that successful founders resemble Elon Musk or Steve Jobs and deviant founders engender greater scrutiny and skepticism simply because they defy this tacit expectation of what a founder should look like.

Second, there is a potentially compounding penalty that arises when evaluators recognize that underrepresented founders face more challenges—like those arising from role incongruity—than White male founders and therefore assess their success as less certain. For instance, Younkin and Kuppuswamy (2019) find that prospective consumers anticipate that the market for a product will be smaller if the founder is Black than when they are White. More generally, awareness of the challenges that underrepresented founders face in raising capital or winning customers may lead potential employees to lower their assessment of the firm’s financial potential. Consistent across both explanations, the race and gender of the founder become influential precisely because of the uncertainty inherent in the entrepreneurial process—lacking empirical evidence of past performance, evaluators are prone to overreliance upon these coarse proxies, leading to negatively biased assessments of the performance of female- and Black-founded firms.

The second question is whether an applicant’s assessment of the startup’s culture, or their fit within the organization, is also liable to be affected by the race or gender of the founder. Although there is little evidence suggesting that race or gender of founders predict consistent differences in startup culture, homophily provides a powerful reason for why an applicant might presume that a founder’s race or gender would affect the applicant’s comfort or fit within the firm. From the selection of friends in preschool to the decision of which house to buy, people consistently exhibit a preference to be around people who look like, or behave like, they do (McPherson et al. 2001). In studies of employment decisions, homophily has been shown to explain both the selection of cofounders (Ruef et al. 2003) and the behavior of job candidates (Campero and Kacperczyk 2020). In these settings, presumably, the individuals weighed a variety of significant concerns, and yet, in each case, the desire to work with similar people proved significant in explaining their actions. Given that startups are small work environments with founders exerting an outsized influence, it seems likely that an applicant would anticipate greater person-organization fit in a firm with a founder of same gender and/or race. Moreover, this effect is compounded if applicants presume that a founder’s traits also predict the composition of the rest of the workforce.

Prior research does offer some support for this idea that applicants infer employee composition from available cues. For instance, in a laboratory experiment, Gaucher et al. (2011) found that when job ads include more masculine words, both men and women thought that there were more men within the occupation. And consistent with the predicted relationship between homophily and fit, when women believed there were more men in an occupation, they found the job less appealing. In contrast, a job ad using more feminine words made it more appealing to women. In our setting, jobseekers might similarly assume that a Black founder would employ more Black people, or that a firm with a female founder would employ more female employees. As a result, candidates might infer from a founder’s race or gender that they would not be a good fit with the firm and withdraw their candidacy.

Although we describe the financial and interpersonal concerns as largely separate issues, there are times when these two domains overlap. Most notably, research has emphasized that applicants assess not just how a position will compensate them in the short term, but also how a position might alter their career trajectory (Bidwell and Briscoe 2010). When applicants assess how the role will shape their future, they must consider a variety of factors—the likelihood for advancement and the training or mentoring they will receive—that are rooted in social ties. These are particularly salient at new firms, because applicants may accept the lower pay or greater risk of a startup in exchange for the opportunity to assume more leadership or responsibility than they would enjoy at a larger firm. But this requires them to predict the likelihood that a founder chooses to entrust them with that leadership, and it is widely understood that mentorship and within-firm opportunities reflect personal relationships (Ibarra 1995). Given the previously discussed strength of homophily in personal interactions, it stands to reason that an applicant might equate outgroup founders as less likely to either become mentors or to favor them with opportunities to advance.

Finally, even if an applicant is not themselves swayed by founder’s race or gender, they may draw inferences about the status loss involved in working for a founder from an underrepresented group. In other words, candidates might be more worried about that working for a minority founder is discounted by others and that accepting a job in a minority-founded firm might therefore hurt their future career prospects. Prior work has shown that applicants are attuned to these external appraisals and alter their behavior in response to perceived status changes. For instance, candidates are more likely to accept a job offer when they receive evidence that most other candidates have also accepted (Coffman et al. 2017). Therefore, even if the applicants are wholly indifferent to the founder’s race and gender, if they anticipate that other candidates will withdraw or that prospective employers will discount the experience, then they should be more likely to withdraw themselves.

Collectively, these considerations suggest that firms with female or minority founders will attract fewer applicants—and that this loss will prove costly. Of course, these same dynamics may also work to attract some applicants, if, for instance, the average female applicant believes a female founder would offer her a better or more supportive environment. The specific effect for any founder might therefore depend upon the composition of the potential applicant pool. However, the net effect—across all occupations and a variety of sectors—is likely to work to the disadvantage of female and minority founders, particularly when they recruit for the technological positions at the center of many successful startups.

In sum, we anticipate that job applicants will be attentive to the race and gender of startup founders and that they will infer from these attributes the prospects and culture of the firm itself. Moreover, for the reasons detailed above, we predict that this attention to founder attributes will lead applicants to lower their estimation of firms with Black or female founders and to be, correspondingly, less likely to apply for a job when a firm has female or minority founders. As a result, it will be more costly for female and minority founders to attract equally qualified employees.

The Field Experiment

Experimental Design

To investigate the effect of founder race and gender on job-entry decisions, we conducted a reverse-audit field experiment. This method involves advertising an open position, introducing a treatment, and then looking for variation in the rate of completed applications. In our study, we partnered with a recently founded technology startup that sought to hire four employees (Jr. Programmer, Marketing, UX Designer, Administrative) as they opened a new office in one of three potential sites (Atlanta, Detroit, Austin). They advertised each of their open positions, in each city, using a national job posting platform and a common description of the firm. Each candidate¹ that expressed an interest was then randomly assigned to one of our three conditions where the founders were two White males, White females, or two Black males.² All interested candidates³ received a reply email inviting them to complete a full application by clicking on a personalized link. The emails were identical except for the names of the two founders (which corresponded to the conditions). When the candidate clicked on their link, it opened the homepage for the startup—which described the business, provided a history of the cofounders, and listed the firm’s open positions. All aspects of the homepage were identical across the three conditions, except the names and photographs of the founders, which varied according to their randomly assigned condition (Online Appendix A, Figure A.1).⁴

The principal outcome of interest in this study was whether a candidate, upon seeing the race and gender of the founders, decided to continue with their application. Therefore, we needed to create a design that ensured that every candidate was aware of the founder before they completed their application. To do this, we built the firm’s homepage in such a way that it was not possible to complete the application without having seen the photographs of the founders. The application itself included both questions of interest to the startup and questions that allowed us to assess differences in the expectations of the candidates (detailed below). All actual applicants were forwarded on to the startup to be considered along with the pool of candidates they received by other means.⁵

A few unique elements of our design should be highlighted. First, because we partnered with an existing firm, all candidates were applying for actual full-time positions. Second, the startup was newly founded and lacked a web presence. Therefore, we were able to design the homepage as described. This also made it very unlikely that the candidates could discover any contradictory information regarding the founders. Third, because most startups are founded by teams of cofounders, rather than solo founders, and because the partner firm themselves had a pair of cofounders, we depicted two cofounders on the site. Fourth, the job-hosting site that the firm used to advertise for their positions collects basic information on each registered user (e.g., education, previous work experience) and encourages them to upload a copy of their curriculum vitae (CV). This allowed us to gather individual-level data about anyone who expressed initial interest in the positions, including those who did not complete their application. In total, 705 candidates expressed interest in one of the positions, and 352 (49.9%) completed an application.

Race and Gender Treatments

To test the effect of founder race and gender, we needed to vary the apparent race/gender of the founder without introducing unintentional variation in the site or startup itself. We did this, as described, by creating a single homepage for the firm but varying the name and photographs of the founders. We selected the names and photographs via a pair of surveys conducted before the field experiment. Specifically, we conducted a pretest in which we hired 250 participants to evaluate the likely race, gender, and education of people with 5 different names (out of a set of 25). In addition, we hired a separate set of 250 participants to evaluate the age, attractiveness, and employment status (employed/student/unemployed) for 4 black-and-white photographs of a person’s face from a set of 100 similarly staged photographs of a person’s face. From these tests, we were able to identify a set of six names that were perceived as being equal in socioeconomic status but that uniformly signaled that the individual was male or female, and Black or White. The pretests also allowed us to identify a corresponding set of six photographs that were perceived as equally attractive, of equivalent age, and equally likely to be presently employed (Online Appendix A, Table A.1). These tests allowed us to create three matching pairs of founders that differed in race or gender, but not in any other potentially influential factor.⁶

Application Patterns Across Conditions

The principal question under investigation is whether an individual’s likelihood of applying for an open position varies with the race or gender of the founder. A total of 705 jobseekers responded to one of our job advertisements. Of these, 352 individuals ultimately applied to one of the jobs. We first note that, conditional on expressing initial interest in the job, applicants for technical jobs (UX Designer and Jr. Programmer) were more likely to complete an application than applicants for nontechnology jobs (Administrative Assistant, Marketing). The overall application probability for tech jobs was 0.66, which is significantly higher than the application probability for nontech jobs, 0.36 (Fisher’s exact test, p < 0.0001). Pooling our results across positions and cities, we find that jobseekers are more likely to complete an application when they believe the founders are White women (55%), compared with when they believe the founders are White men (50%) or Black men (45%) (Fisher’s exact test, p = 0.115).

Because we randomized the conditions across all candidates, we allowed the conditions to be imbalanced at the city-occupation level.⁷ Therefore, we collected additional information on the candidates from their resumes, including their gender, whether they held a college degree, their years of work experience, whether they were a local candidate, and the overall quality of their resume for analyses that could control for city and occupation differences. Table 1 presents a balance table for the field experiment and subsequent two survey experiments, describing the differences in participants assigned to each condition.

Table 1. Balance Table for Three Experiments

One of the challenges with our design was determining the gender of each applicant. The job-hosting site does not collect information on gender, and asking the jobseekers to report it could have influenced their decision to apply. To avoid this bias, we used an algorithmic approach to infer gender. The NamSor classifier calculates weighted gender probabilities using the combination of first and last name, taking into account the cultural origin. For example, the algorithm recognizes that Karen Gallagher is a female, Irish name, whereas Karen Grigoryan is a male, Armenian name. For each name, the classifier returns the value male, female, or unknown, as well as a parameter scale ranging from −1 to +1 to reflect the certainty that a name is male or female. For all names that were not given a confident assignment, we searched for additional information on the jobseekers’ profile and consulted other social networking sites if links were provided. Of all interested jobseekers, 338 were men (48%), 350 were female (50%), and 18 (2%) could not be assigned as either. We find that women are overall less likely to apply than men (42.4% versus 57.1%; Fisher’s exact test, p < 0.001), a similar trend that is reported in Flory et al. (2015) and Leibbrandt and List (2015).

To control for potential differences in human capital, we rely on several measures. First, as part of their initial request, the job-hosting site asked each candidate to specify their level of education and whether or not they had experience working in a similar capacity. These were not screening questions, but jobseekers could choose to apply for a job that they were not qualified for based on their answers. In our sample, 500 applicants (71%) have at least a bachelor’s degree, and 352 (50%) had relevant work experience. To complement these coarse indicators with a more refined measure of human capital, we hired six professional recruiters specializing in high-tech recruiting to evaluate the resumes of all jobseekers (including the ones who ultimately did not apply). The recruiters were given the job descriptions for each position and a short description of the startup, but no information on the founding team, and asked to consider the candidate for a specific position. They then gave each resume a quality score between one and seven and a short one- or two-sentence assessment. Moreover, all jobseekers were sorted into two groups according to their overall recommendation: “callbacks” and “noncallbacks.” The mean quality score for the jobseekers was 3.9 (standard deviation (s.d.) 1.8). In total, 348 individuals received a callback recommendation.

Table 2 presents the results of a logit model estimating the average treatment effect of race and gender of the founder, controlling for the observed position and respondent differences. To gauge the stability of the coefficients for our main treatments, we introduce four logit models and then an ordinary least squares (OLS) and probit model. Model 1 presents a baseline for the two treatment conditions relative to the White male condition and reports the initial indication that candidates were less likely to complete applications for Black male founders (−0.18, 0.07) than for White male founders. Model 2 includes job-specific criteria relative to the referent city (Atlanta) and occupation (Administrative Assistant), and Model 3 includes four observed applicant-level differences (gender, location, education, and amount of work experience); both of these models also report statistically significant negative results for Black male founders.

Table 2. Effect of Founder Characteristics on Job Applications (Study 1)

Model 4 tests our main hypothesis and finds, controlling for city and occupation, that applicants were significantly (−0.33, 0.06) less likely to apply for a position when the founders were two Black men than when they saw two White men. However, we do not find any difference in the likelihood that a candidate will apply to a firm with female founders relative to male founders. To ensure that the results were not sensitive to a specific estimation technique, we also estimated a linear probability model (LPM) (Model 5) and a probit model (Model 6), both of which also find a significant reduction in applications for Black male founders, but no statistically significant difference between White male and female founders.

Beyond differences in the availability of high-quality employees lies the question of their costliness. Specifically, we proposed that minority founders may have to pay a premium in order to acquire equivalent levels of human capital. To test this, we asked the job applicants for their desired annual salary. The question was open-ended, allowing applicants to provide any number. The mean requested salary was $53,129 (s.d. 39,616). Using pooled data across all cities and occupations, we found a significant difference across the three conditions (F(2, 351) = 3.04, p < 0.05). Conditional on applying, candidates who believed the founders were Black men requested annual salaries ∼$10,000 higher than those who believed the founders were White men or White women (mean_{black_male} = $61,013, s.d. = 66,111; mean_{white_male} = $49,639, s.d. = 17,017; mean_{white_female} = $49,791; s.d. = 19,309).⁸

Table 3 presents the results of an OLS model estimating the effect of race and gender of founder on the requested salary, controlling for respondent characteristics. We estimate the models separately for tech jobs and nontech jobs. Model 1 presents the treatment conditions pooled across all job types, and, as described above, the average annual salary requested of Black male founders (11,374, 5,281) was more than $11,000 more than for White male founders. Model 2 shows that for tech jobs, candidates asked for a higher starting salary when shown Black male founders than when shown White founders. In Model 3, we estimate the effects of founder race and gender on requested salary for applicants for nontech jobs and find that the salary request does not vary with the race or gender of the founders. It is important to note that we only have data on salary for applicants that completed their application and, therefore, not for the full sample. As a result, if there are differences in the quality of candidate who apply to Black and White founders, then this may explain any differences in the average salary requested. For example, if White founders receive higher-quality candidates, on average, then it stands to reason these higher-quality candidates would also request higher average salaries.

Table 3. Effect of Founder Characteristics on Requested Annual Salary (Study 1)

To explore this possibility, we examined if the effects of founder race and gender depend on the quality of the applicant. Models 4 and 5 include the interaction effects between the founding team characteristics and the callback rating provided by the professional recruiters. We find that the high-quality (i.e., callback) candidates for tech jobs asked for a higher starting salary when shown a Black founder than when shown a White founder (Model 4). This suggests that the initial penalty is not entirely attributable to differences in the pools; instead, we have some evidence that—controlling for quality—candidates adjust their request according to the founder’s characteristics. At the same time, we do not find a significant interaction effect between the callback rating and founder characteristics among nontech job applicants (Model 5), suggesting that less-skilled workers are less influenced by founder demographics, irrespective of the applicant’s quality.

Pairwise comparison of predictive margins indicates that for tech jobs, high-quality applicants asked an additional $34,524 in annual salary when they were shown Black founders compared with White male founders (F = 7.89, p < 0.01) and an additional $28,048 compared with White female founders (F = 4.62, p < 0.05). For nontech jobs, we find that high-quality candidates asked higher starting salaries than low-quality candidates, but there were no differences by founder race and gender.

Heterogenous Treatment Effects

A natural question raised by our results is whether an applicant’s response to founder race and gender varies with the applicant’s own gender, ability, or other characteristics. Our research design allows us to isolate the effect of founder race and gender on the proportion of all interested jobseekers who ultimately apply, reducing concern about self-selection into treatment, but still allowing for potential heterogeneity in the treatment effect. That is, different subgroups of applicants could respond differently to the race and gender of the founder. Insofar as subgroups differ in terms of human capital or cost of labor, understanding variation in treatment effects is therefore critical for understanding how gender and race shape inequality among entrepreneurs. To further examine how founder race and gender affect applicant interest, we conducted four additional sets of analyses focusing on applicant quality and gender.

Job Applicant Quality

In the prior set of analyses, we estimate the likelihood of an individual application; however, one might argue that the size of the applicant pool matters less than the quality of the applicant pool. We predict that the average candidate will be sensitive to potential negative signals about the firm; if true, it stands to reason that the candidate’s sensitivity to this information will correspond to their alternatives—a bad job may be preferable to no job. This would imply that the likelihood of bias is moderated by the number of options available to the candidate, such that higher-quality candidates will be more likely to withdraw their application than lower-quality ones. In other words, more capable applicants might be more likely to leave the applicant pool when they believe a firm has female or Black founders. If true, then female or minority founders suffer from unequal access to human capital, which could lead to substantial productivity losses and put female- and minority-founded companies at a disadvantage. Therefore, as one means of probing our hypotheses, we explored the possibility that high-quality applicants differ in their behavior from low-quality applicants.

In Table 4, we examine evidence on ability-based sorting by estimating the effect of each treatment on the likelihood to submit a job application for two different subgroups: high- and low-quality candidates, defined in three different ways. First, Model 1 provides logit estimates for the likelihood of an application by the candidates whose CVs were rated as “callback worthy” by a professional recruiter. As predicted, we find that the high-quality callback candidates were less likely to continue (−0.56, 0.09) with their application when shown a Black founder than when shown a White founder. We do not find a significant difference in the response to female founders relative to male founders. In Model 2, we estimate the likelihood of an application by the lower-quality “no callback” candidates and find that their willingness to apply does not vary with the race or gender of the founders. As a second quality test, we separated the sample according to whether the candidates had a college degree or not. In Model 3, we find that college graduates are more likely to discontinue their application when shown a Black founder (−0.39, 0.08) than a White founder. In Model 4, we find that candidates without a degree were no more likely to apply when the founder is White or Black, but that they were more likely to continue their application (0.83, 0.36) when shown a female founder than a male founder. Finally, we split the sample according to which candidates had relevant work experience (Model 5) and which did not (Model 6). Again, we found that the higher-quality (e.g., experienced) candidates were less likely (−0.39, 0.11) to continue their application when shown a Black founder. These results do not change in direction or significance if estimated using probit models or LPMs.

Table 4. Probability of Job Application by Applicant Qualification (Logit)

Candidate Gender

We theorized that homophilous preferences may help to generate the observed pattern of bias. Although we could not identify the candidate’s race with any confidence, we were able to use their first and last name to predict their gender. From this, we can investigate whether the comparable responses to male and female founders reflected indifference to gender or if they reveal bifurcated responses where men prefer male founders and women prefer female founders. We find that women do exhibit a preference for working for female founders—they were more likely to continue their applications for female founders than when confronted with male founders—and they were less likely than men to apply to the firm when the founders were White men. Consistent with Flory et al. (2015), we find that the gender of the founder has no effect on applications by men.

Discussion

The results described above suggest that job candidates are aware of and responsive to the race of a startup’s founder. The net effect of these individual preferences leaves Black founders at a significant disadvantage relative to White male or female founders, because they have both fewer candidates to choose between and a lower quality of candidates to select from. At the same time, our experiment did not yield any evidence of aversion to female founders. Contrary to our expectation, applicants presented with female founders were neither less likely to apply nor did they request higher salaries as compensation.

It is important to acknowledge that although the field experiment identifies a baseline difference, it does not enable us to explain why job candidates behave in this manner. And, as highlighted in the theory, there are a wide range of potential explanations for the phenomena, ranging from simple bias to concerns around fitness, performance, or the preferences of the founders themselves. Moreover, the results are specific to a startup and were only tested for three positions in three locations at a specific point in time. To delve deeper into the mechanism behind the observed main effect, we turn next to two survey-based experiments.

The Survey Experiments

To help illuminate the mechanism driving the behavior observed in the field experiment—and thereby better identify potential remedies—we conducted a pair of survey-based experiments, in which we asked current (or recent) jobseekers to analyze and compare the sites for two fictional startups.

Experimental Design

In the first survey, we recruited 330 U.S.-based jobseekers via a professional firm to evaluate startup features that matter to jobseekers.⁹ They were offered $4 and told that the survey would require up to 15 minutes of their time. No respondents were included, or paid, unless they completed all the survey questions. The results are limited to the 318 respondents that completed all questions, passed the attention check, and spent between 2 and 20 minutes on the survey. Relaxing or enhancing these criteria to include or exclude more respondents does not change the results in direction or significance.

After respondents agreed to participate,¹⁰ they were asked to identify their occupation, whether they were actively looking for a position, and their number of years of work experience. All respondents were then directed to the website for the first startup, held constant across all conditions, and asked to assess the firm’s attractiveness to prospective candidates. Specifically, they were asked to assess the quality of the website (six-point scale), the clarity of the firm’s business model (seven-point scale), and the quality of the founders (seven-point scale). Respondents were then randomly assigned to one of the same three conditions from the field experiment—White males, White females, Black males—and directed to the website from our initial experiment that corresponded to their condition. They were then asked the same set of questions about the firm’s attractiveness to prospective candidates. As a result, each respondent saw two websites, one constant and one variable, answered an identical set of questions after each site, and then answered a set of comparison questions at the end of the survey. The survey closed with a brief demographic set of questions and an attention check. The full list (and order) of questions and an image of the constant site are available in Online Appendix B (Table B.1).

To formulate the comparison questions, we drew upon prior research on hiring practices and bias to identify six testable potential explanations for the observed discrepancy in applications between White and Black founders. For one, applicants may use the founder’s race to make inferences about firm performance and the potential pecuniary benefits (e.g., salary, stability, opportunity to advance) that work to the disadvantage of Black founders. Second, applicants may express equally influential but less tangible sentiments that nonpecuniary benefits (e.g., person-organization fit, workplace culture) correlate with the race of the founder. For instance, given evidence that founders imprint a particular culture on their startups, applicants may be wary that they will be less likely to match the culture if they do not match the founder’s traits. Similarly, given repeated evidence for the ubiquity of homophilous preferences, an applicant may fear that outgroup founders will be less likely to proffer mentorship or tacit forms of support than a founder from their ingroup. In essence, White applicants may anticipate that a Black founder would establish a firm with a workplace culture distinct from, and less supportive to them, than what a White founder would establish. As a result, irrespective of whether the applicant exhibits any overt animus toward Black founders, there is reason to believe that non-Black candidates are liable to withdraw from the application process.

Further, we do not presume that these mechanisms are mutually exclusive but recognize that they may work in combination or isolation. We therefore designed a separate set of questions to test each of the six mechanisms described in our theory. Specifically, after reviewing the second firm, we asked the candidates to compare the two firms and to select choose one of them. To assess the firm’s likelihood of failure, we asked which firm: (1) was more likely to fail and (4) to take more risks. To assess the firm’s potential for growth, we asked which firm: (2) is more likely to be more successful in 5 years and (3) has more potential to grow. To assess concerns related to person-organization fit, we asked which firm would: (5) be a better fit for you and (6) offer better work/life balance. To assess concerns around within-firm opportunities, we asked which firm would be more likely to (7) offer you a mentor. Next, to examine presumptions regarding prospective employers, we asked them to identify which job will (8) look better on a resume. Last, to assess the role of external perceptions, in particular any presumption regarding other candidates, we asked which job will (9) pay better and (10) attract more job applications. Finally, we asked the respondents, “If both firms were hiring in your area, for your position, where would you rather work?”

Experimental Results

When pooling across all respondents, we find few results that might help to explain the pattern detected in the field experiment (Online Appendix A, Table A.2). However, average effects mask substantial variation in how jobseekers respond to treatments. When we limit our analyses solely to White respondents, we find that they were more likely to identify the control firm as superior—financially and in terms of fitness—when they believed the founders of the treatment firm were Black. Table 5 reports the estimated effects of race and gender of the founder on inferences about the firm for White respondents. Specifically, in Model 1, White respondents who were shown White male founders were less likely to identify the firm as at risk for failure (−0.66, 0.34) than White respondents shown Black founders. Similarly, in Model 2, White respondents shown White male founders were also more likely to identify the firm as a potential success (0.58, 0.34) than White respondents shown Black founders. However, we do not find any evidence, in Model 3, that the race of the founder affected perceptions of the firm’s potential. And, in Model 4, we find directionally consistent but not significant results that the idea was less risky when led by White male founders (−0.51, 0.34).

Table 5. Effect of Founder Characteristics on White Applicant Assessments (Study 2)

White respondents also displayed apprehension regarding their fit within an organization with Black founders. In Model 5, White respondents shown White founders were significantly more likely to identify the firm as a good fit (0.90, 0.35) than White respondents shown Black founders. In Model 6, White respondents shown White founders were more likely to identify the firm as offering a good work/life balance (0.58, 0.34) than White respondents shown Black founders. Although not at conventional levels of significance, in Model 7, we find directionally consistent evidence that White respondents shown White founders were more likely to identify a firm as offering mentorship opportunities (0.56, 0.34) than White respondents shown Black founders.

Finally, we found less evidence that respondents would anticipate differences in the short-term returns to working at a firm with Black founders. Model 8 does show that White respondents shown a White founder were more likely to identify the firm as looking good on their resume (0.78, 0.38) than White respondents shown Black founders. However, we do not find any evidence that the respondents predicted a salary differential (Model 9) or difference in the competition for a position (Model 10). Estimating these results using a naïve OLS regression (Online Appendix A, Table A.3) rather than logit does not significantly change the findings.

This initial survey experiment helps illuminate some of the reasons why Black founders faced smaller, and costlier, pools of applicants in the field experiment. Most obviously, the findings suggest that White applicants perceive Black founders as at risk for failure, as poor fits, and unlikely to offer them mentorship. These devaluations may help to explain why White candidates would either withdraw their application or request a higher salary from positions within a Black-founded firm. By contrast, non-White respondents did not respond more favorably to the Black founders, suggesting that they do not receive any compensatory ingroup preference that might offset the assessments of White applicants.

Although these results are useful in identifying a potential explanation for the differences observed in the field experiment, the dearth of Black applicants in the survey limit our ability to draw firm conclusions regarding potential heterogeneity in group responses. For instance, it remains unclear whether all applicants perceive Black founders as poor mentors and unlikely to create supportive organizational cultures or whether these sentiments were specific to non-Black candidates. Similarly, it remains unclear if Black respondents view firms founded by Black founders more favorably. To address this and to add greater clarity to the mechanism driving applicants to withdraw or raise their salary request, we repeated the survey experiment with a balanced sample of White and Black jobseekers.

Tests of Applicant-Founder Homophily

In the second survey, we used the same professional service firm to recruit 480 White and Black jobseekers,¹¹ aiming for a total sample of 400 after the exclusion restrictions. The recruitment, pay, and script were the same as used in the first experiment. As before, the results described below are limited to the 445 participants who completed the entire survey, passed an attention check, and took between 2 and 20 minutes to complete the survey. Relaxing or enhancing these criteria to include or exclude more respondents does not change the results in direction or significance. This experiment was preregistered with the Open Science Foundation¹² prior to the recruitment of subjects.

Experimental Design

The experimental design was identical to the first experiment with two principal exceptions: first, we only compared the White male and Black male founder conditions, and second, the respondents were balanced such that an equal number of Black and White respondents were sent to each of the two conditions. This meant that, after respondents agreed to participate, they were asked to identify their occupation, whether they were currently looking for a position, and their number of years of work experience. All respondents were directed to the same startup, received the same questions regarding site/firm quality, and then were randomly sent to one of the variable sites and asked the same set of questions. The respondents were then asked a modified set of comparison questions, listed below; a brief demographic set of questions; and an attention check.

The first round revealed that some information is hard to discern from a site and some outcomes do not appear affected by the race of the founders. In recognition of this, we reduced the number of comparison questions to focus our study on understanding the areas with significant differences. Specifically, we asked the respondents to identify which firm: (1) is more likely to fail, (2) is more likely to be more successful in 5 years, (3) will be a better fit for you, (4) will offer you a mentor, (5) will offer better work/life balance, (6) looks better on a resume, and (7) will pay better.¹³ Finally, we asked the respondents, “If both firms were hiring in your area, for your position, where would you rather work?”

Experimental Results

The question at the heart of this study was whether the race of an applicant influenced the value attached to the race of the founder. Additionally, we sought to determine whether there was a general antipathy toward cross-racial pairings, in which case Black respondents might also penalize White founders, or whether the penalty was unique to the case of Black founders and White applicants.

Mean comparison tests provide some evidence that the penalty is unique to White applicants evaluating Black founders, raising a hiring challenge for these founders that we do not find among White founders. For example, White respondents shown a Black founder believe that the firm is more likely to fail (mean = 0.786, s.d. = 0.412) than White respondents shown a White founder (mean = 0.666, s.d. = 0.473, t (227) = 2.03, p = 0.044), whereas Black respondents do not provide different estimates when shown a White or Black founder. Similarly, White respondents shown a Black founder perceive the firm as a worse fit for themselves (mean = 0.330, s.d. = 0.472) than White respondents shown a White founder (mean = 0.444, s.d. = 0.499, t (227) = 1.78, p = 0.077); Black respondents, again, do not differ. We also find that White respondents anticipate that working for a White founder will pay more and look better on their resume than working for a Black founder; Black respondents did not vary on either measure. These results are identical to those found via one-way ANOVA.

For robustness, we present a logit estimation of the likelihood that an applicant preferred the treatment condition for White (Table 6) and Black respondents separately, controlling for all observable differences between respondents. These results are comparable to those presented above. In Model 1, we find that White applicants shown a White founder were less likely (−0.64, 0.32) to identify the firm as liable to fail than those shown the same firm with a Black founder. In Model 3, we find that White respondents shown a White founder were more likely to identify the firm as offering a good personal fit (0.56, 0.30) than White respondents shown a Black founder. In Model 6, we find that White respondents were more likely (0.72, 041) to identify a firm as good for their resume when the founders were White than they were if the founders were Black. Finally, in Model 7, we find that White respondents are more likely to identify the firm as offering higher salary (1.09, 0.45) when the founders were White than were White respondents shown a Black founder. We also find that the coefficients for success, mentorship, and work/life balance are positive for White founders, but the results were not significant. When we limit the sample to Black respondents, we do not find that the race of the founder is significant for any of seven outcomes (Online Appendix B, Table B.1). In addition, estimating the results via naïve OLS (Online Appendix B, Table B.2) or excluding respondents outside the United States did not significantly alter the results. In each case, we find no evidence that Black respondents differ in their assessments of the startup, but White respondents consistently rate the startup more favorably when shown a White founder.

Table 6. Effect of Founder Characteristics on White Applicant Assessments (Study 3)

Discussion

In summary, we find that there is a significant difference between the perceptions of Black and White founders, but the effect depends on the race of the jobseeker. White respondents exhibit a preference for White founders, but Black respondents do not favor one set of founders over the other. These results are consistent with recent work by Pager and Pedulla (2015), which finds that Black jobseekers apply more broadly than White ones. Therefore, although it remains possible, we find no evidence of a widespread applicant homophily where all applicants favor founders of their own ethnic group, nor do we find evidence of widespread statistical discrimination whereby all applicants penalize Black founders for being atypical members of the entrepreneurial class. Instead, we solely find evidence—consistent with what was identified in Study 2—that White applicants prefer firms with White founders to firms with Black founders, even when the startups are themselves identical. The net result of this imbalance leaves Black founders at a sizeable disadvantage relative to White ones when it comes to acquiring talented employees.

Conclusion

Do job candidates discriminate against female or minority founders? The question at the heart of this paper is not a complicated one, and there are good reasons to assume that the answer is “no.” Job candidates make complex decisions that involve dozens of different factors from location to salary to personal fit. Some of these they know precisely, others they know fall in a range, and others can only be inferred. Amidst this sea of information, it is reasonable to assume that the gender or race of the founder amounts to no more than a single small ripple.

Our results suggest that this optimistic assumption is not an entirely accurate one. Evidence from our field experiment suggests that job applicants’ behavior and their assessments of the firm itself are shaped by founder attributes. Specifically, we find that job candidates are less likely to complete an application they had already signaled an interest in when they learn that the founders are Black than if they believe that the founders are White. Moreover, we find that the likelihood of withdrawing an application increases with the quality of the candidate, leaving Black founders with a pool of candidates both smaller and lower in quality than their White peers. Moreover, we find evidence that this is at least partially motivated by concerns that a White candidate will be less likely to fit within a firm with Black founders and partially by concerns about the viability of the firm. Consistent with work showing that Black job candidates do not try to avoid discrimination (Pager and Pedulla 2015), we do not find evidence that Black applicants prefer Black founders—only that White applicants prefer White ones, suggesting that the patterns we observe are not a result of general ingroup bias or race-based homophily (cf. Campero and Kacperczyk 2020).

These results auger a worrisome self-fulfilling prophecy whereby an applicant’s fear that Black-founded firms will underperform could leave those founders reliant upon less capable employees, generating the very underperformance that the applicant feared—only for very different reasons. Nor is it immediately obvious how a Black founder could counter these concerns as the most common devices—evidence that contradicts the stereotype—are not readily available to most startup founders. It is our hope that future researchers will seek to extend this line of inquiry to uncover potential mitigation strategies and to identify proactive steps that the founders can take to attract a stronger pool of candidates. Moreover, future work should test the generalizability of these findings beyond the context of startups. For instance, would we expect that Black managers at established firms would face similar challenges? And does the penalty persist as the firm ages and grows?

Fortunately, there is reason for optimism that the results described here would not extend to those cases and are limited to the experience of new ventures. We theorize that attention to the founder’s race is greater in the absence of available information, which would suggest that an applicant would be less likely to rely on such cues when assessing established organizations or even startups with measurable evidence of performance. However, the concerns regarding cultural fit may prove more persistent because they reflect concerns about interpersonal interactions, not firm performance. A recent study by Brenner et al. (2023) finds that female CEOs of large public companies are more likely to be viewed more harshly by their male employees, especially when the firm has made investments in diversity, equity, and inclusion initiatives, suggesting that demographic characteristics continue to play a role as the firm grows.

It is also important to mention although while we predicted a similar penalty for female founders, we do not find any evidence that a founder’s gender affects the decision to apply. This finding differs from recent work by Campero and Kacperczyk (2020) showing that female software engineers are more likely to apply to firms with female founders, which they interpret as evidence of homophily among jobseekers. However, the experimental approach we employed allows us to isolate the influence of gender in a way that was not possible in their setting. Their results may reflect the fact that female founders are more likely to run certain type of businesses, use a certain type of language in their job postings, or offer certain types of benefits attracts more female candidates. Our research design allows us to keep these factors constant, which may help to explain the differences between our findings.

What we do find is evidence of an unrecognized challenge facing Black founders. Across all three experiments, the results are consistent: jobseekers are less willing to work for Black founders. As a result, Black founders must accept lower-quality applicants or pay higher salaries to compensate for the fact that applicants perceive their firms as higher risk and lower fit than identical firms with White founders. Coupled with past evidence of the difficulties minority founders face raising capital, our findings reinforce the fact that to assess, and improve, the performance of Black founders, we must first identify the varied and subtle ways in which race makes the process more arduous.