As an HR professional, it is important to understand the overall makeup of an organization, especially company staff. Diversity is an ongoing concern for some organizations. There are employment laws

We use a natural field experiment to test several hypotheses on effective means to attract minority candidates for top professional careers. By randomly varying the content in recruiting materials of a major financial services corporation with more than 10,000 employees, we find that signaling explicit interest in employee diversity more than doubles the interest in openings among racial minority candidates, as well as the likelihood that they apply and are selected. Impacts on gender diversity are less sharp and generally not significant. Jeffrey A. Flory is Associate Professor of Economics at Claremont McKenna College. Andreas Leibbrandt is Professor of Economics at Monash University. Christina Rott is Assistant Professor of Economics at Vrije Universiteit Amsterdam. Olga Stoddard is Assistant Professor of Economics at Brigham Young University ([email protected]). The authors thank Rebecca Jack for excellent research assistance and gratefully acknowledge financial support from the Bias Interrupters Working Group. The authors also thank participants of SOCCAM, ESA AFE, IMEBESS, and Science of Diversity and Inclusion Initiative (SODI) conferences, as well as seminars at University of California (San Diego), Brigham Young University, Claremont McKenna College, University of Melbourne, Maastricht University, Amsterdam Business School, Vrije Universiteit of Amsterdam, and Tilburg University for their helpful comments on earlier drafts of the paper. The authors are listed in alphabetic order. The authors declare that they have no relevant or material financial interests that relate to the research described in this paper. Institutional Review Board (IRB) approval was obtained to conduct this research involving the collection of data on human subjects. The data are available at olgastoddard.byu.edu/research.

[Submitted May 2018; accepted April 2019]; doi:10.3368/jhr.56.1.0518-9489R1 JEL Classification: J01 and D01 ISSN 0022-166X E-ISSN 1548-8004ª2021 by the Board of Regents of the University of Wisconsin System Supplementary materials are freely available online at: http://uwpress.wisc.edu/journals/journals/ jhr-supplementary.html THE JOURNAL OF HUMAN RESOURCES 56 1 I. Introduction Understanding and eliminating barriers to career progress based on bias against people on the basis of their race, gender, and other social identity, rather than on skills and ability, is an important issue in the U.S. labor market. Removing social identity–based frictions in the flow of human capital to its most productive uses in- creases economic efficiency, making demographic diversity an increasing priority for firms. Estimates suggest companies spend nearly $10 billion a year on diversity and inclusion training initiatives (Hansen2003), with some dedicating as much as $200 million a year or more to diversity programs (Catalyst 2005), and the Chief Diversity Officer has now become an established leadership position in the country’slargest firms. 1However, despite significant education gains among underrepresented groups, and substantial resources devoted to enhancing employee diversity in high-profile occupations, many firms still struggle to increase representation by individuals from minority groups. Through a field experiment in a major U.S. corporation, we show how simple changes in presentation of recruitment materials can have powerful impacts on selection into high-profile opportunities by people in underrepresented groups, and ethnic minorities in particular.

The managerial landscape in many U.S. companies continues to resemble mono- cultures. Almost 90 percent of Fortune 500 CEOs are white males, while less than 4 percent are African American or Hispanic, and less than 6 percent are women. 2Among all U.S. companies with 100 or more employees, the proportion of Black men in man- agement barely increased from 3 percent in 1985 to 3.3 percent in 2014, and the pro- portion of white women has stayed mostly flat since 2000 at under 30 percent (Dobbin and Kalev 2016). These percentages have remained remarkably low, despite signifi- cant advances in educational achievements of racial minorities and women over the last few decades. African Americans and women, for example, account for ever larger proportions of MBA-holders in the United States, rising from 4 percent in 1990 to 14 percent in 2015 for African Americans and from 22 percent in 1980 to 47 percent in 2014 for women (National Center for Education Statistics 2018). This disparity sug- gests important barriers in the career trajectories for racial minorities and women, but it also suggests significant opportunities for organizations to increase demographic di- versity among employees. However, the key question remains how.

Nearly all Fortune 500 companies and almost half of all mid-size companies in the United States have programs to enhance employee diversity (Dobbin and Kalev 2016). These programs contain a range of company-specific initiatives to reduce discrimination and unintentional biases in employee hiring and promotion and to provide resources for underrepresented groups, including scholarships, mentoring, 1. Korn Ferry, the top executive recruiting firm, states that about 60 percent of the Fortune 500 companies have Chief Diversity Officers, or the equivalent: https://www.kornferry.com/functions/divhttps://www.kornferry .com/functions/diversity-officersersity-officers (accessed May 29, 2020).

2. See, for example, http://fortune.com/2015/12/23/2015-women-fortune-500-ceos, http://www2.ucsc.edu /whorulesamerica/power/rise_and_fall_of_diversity.html, http://fortune.com/2016/06/06/lessons-fortune-500, http://fortune.com/2013/05/09/women-ceos-in-the-fortune-500 (accessed April 21, 2020). 74 The Journal of Human Resources and support groups. However, much remains unknown about the actual impacts of programs aiming to foster workplace diversity. This makes it very difficult to eval- uate their cost-effectiveness. To make matters worse, there is suggestive evidence that some diversity programs can actually backfire (Gilbert and Ivancevich 2000; McKay and Avery 2005; Kalev, Dobbin, and Kelly 2006; Apfelbaum, Stephens, and Reagans 2016), increasing the urgency for rigorous causal analysis of these different approaches.

In order to better understand how minority candidates can be attracted for future management positions, we conduct a natural field experiment in one of the largest firms in the financial services industry—a setting widely perceived to lack both racial and gender diversity. By randomizing the content of recruiting information across individ- uals, we exogenously vary whether and how the employer signals that it values diversity among its employees. We also systematically vary whether the diversity statements are backed by facts in order to assess the importance of supporting information when using this type of approach to raise employee diversity. This is important since some orga- nizations may be unable to furnish evidence on stated diversity values, and it is unclear whether individuals respond to unsupported diversity statements or instead view it as “cheap talk.”Finally, we compare the impact of statements that directly target candi- dates from underrepresented groups to statements that use a less direct appeal—since there may be settings in which an overt approach to diversity recruiting is impractical or undesirable, and since there is suggestive evidence that some types of direct signals can actually push away skilled employees from racial minority groups (Leibbrandt and List 2018).

This work is related to a growing literature on understanding and eliminating barri- ers to labor market entry and career progress based on race, gender, and other social identities (Hinton et al. 2010; Giuliano, Levine, and Leonard 2011; Bertrand and Duflo 2017; Blau and Kahn 2017). A rich vein of this literature includes field experiments and studies on gender differences in willingness to enter competitive environments (Buser, Neiderle, and Oosterbeek 2014; Flory, Leibbrandt, and List 2014; Preece and Stoddard 2015, 2016; Flory et al. 2018) and how affirmative action policies and quotas can reduce the gender gap (Balafoutas and Sutter 2012; Niederle, Segal, and Vesterlund 2013; Leibbrandt et al. 2018; Ibanez and Riener 2018). While there are several related studies showing that the way a job is described can affect the applicant pool (Ashraf, Bandiera, and Lee 2014; Dal Bó, Finan, and Rossi 2013; Marinescu and Wolthoff 2016), no prior research has investigated how signaling a firm’s valuation of diversity affects interest and applications by people from underrepresented ethnic/racial mi- nority groups.

The paper arguably most closely related toours is by Del Carpio and Guadalupe (2018).

In a series of two field experiments in Lima and Mexico City, the authors investigate whether varying information provided to potential applicants to a coding program is effective in attracting women. Randomized messages contained information about role models, career prospects, and peer network. The authors find that exposure to the message significantly increased application rates by women, with the statement about role models being the most effective. While this paper sheds light on effective ways to narrow the gender gap in historically male-dominated fields like tech, it does not explore racial or ethnic diversity, which is at the core of our study. Flory, Leibbrandt, Rott, and Stoddard 75 II. Hypotheses and Experimental Design A. Research Hypotheses Underrepresented groups may be dissuaded by workplaces they perceive as lacking in demographic diversity, such as race and gender, for several reasons. From a pecuniary perspective, avoiding certain work environments may be a rational payoff-maximizing response to income risks from working in settings with little or no representation of one’s own demographic group. This may be due to expectations of discrimination by supervisors or unconscious biases in performance evaluations, causing an individual to believe her effort will be less rewarded in these settings. It may also be a strategic avoidance of risk that actual performance might become lower in certain work settings.

Working in an environment where there are few or no other individuals of the same identity group is likely to make that identity salient, and several studies have shown that emphasizing social identities such as race and gender can undermine performance, lower effort and expected success, and increase anxiety—often referred to as“stereo- type threat”(Steele and Aronson 1995; see Bertrand and Duflo 2017 for a review). This may cause some from less represented groups to seek work settings where their social identity is less salient. From a nonpecuniary perspective, individuals may have a pref- erence for work settings that include others from their own group, or workers from underrepresented groups may prefer environments with greater representation from a variety of different demographics.

This suggests that being perceived as lacking in diversity of social identities may cause an employer to have difficulty attracting individuals from underrepresented groups. However, it also suggests a firm may be able to increase its ability to attract such candidates by presenting itself as diverse or as placing high value on the diversity of its workforce. This leads to our first main hypothesis.

Hypothesis 1: Projecting diversity among employees as an organizational value and priority will increase interest among underrepresented groups and achieve greater demographic diversity in recruiting outcomes.

However, in some circles there is concern about an approach to workplace diversity that is too overt. There have also been reports of instances of backlash against diversity efforts, and discontent among individuals excluded from these efforts—typically, ethnic- majority males. 3Some firms and organizations may therefore be reluctant to appear as if they are intentionally trying to court underrepresented groups, while still wanting to increase their rates of recruitment.

One approach is to use an appeal to cognitive diversity, by encouraging interest among individuals from a variety ofeducationalbackgrounds. The conceptual basis for this type of approach is threefold. First, appealing to fields of study other than those typically associated with the industry, occupation, or firm can have a mechanical effect if those fields have greater proportions of underrepresented groups. Second, expressing an interest in diversity of educational background may be interpreted by job-seekers as a signal of an openness or desire for more general diversity among employees (including dimensions of social identity), which may attract candidates from under- represented groups for reasons similar to the rationale for the more direct approach 3. For examples, see Green (2017), Emerson (2017), Toten (2017), andCEB Talent Daily(2017). 76 The Journal of Human Resources discussed above. Third, if it breaks stereotypes about skillsets needed in an industry or occupation, this may make some groups more willing to opt in (similar to studies showing increased selection into competitive settings by women when altering gender- task stereotypes—see, for example, Shurchkov 2012; Flory, Leibbrandt, and List 2014).

Breaking stereotypes may also push individuals across a System 1/System 2 divide (Stanovich and West 2000; Kahneman 2011), replacing an intuitive aversion to non- diverse environments with more reflective responses that cause some to opt in as a payoff-maximizing choice. This leads to our second main hypothesis.

Hypothesis 2: Projecting a desire for employees from a variety of academic fields and training, including those not commonly associated with the industry, will increase interest by underrepresented groups and achieve greater demo- graphic diversity among selected candidates.

Finally, it is also important to know whether words are enough, or if a proven track record or commitment to employee diversity is important. The practical relevance of this question is particularly strong for organizations with little existing diversity. There are good reasons to expect facts of this nature may matter. On the one hand, individuals might interpret diversity-friendly statements as cheap talk—a cynical effort by the firm to give the appearance of trying to address a public demand or help promote a social good but lacking any real commitment behind it. On the other hand, even if the appeal for diversity is interpreted as being made in good faith, it might have far less effect (and potentially even backfire) if not backed up with evidence regarding the firm’s priorities, values, or current conditions. For example, an attempt to get underrepresented groups to apply may be interpreted by some as an indicator of a“diversity problem”(little current diversity, lack of support once hired, etc.), which could dampen the impacts of a diversity appeal (or even push away some candidates that might have otherwise applied). This leads to our third main hypothesis.

Hypothesis 3: Including factual information to support claims about the value of diversity among employees to the organization will increase the ability of pro- diversity statements to raise interest among underrepresented groups.

B. Experiment Design Overview To test these hypotheses and uncover how to increase workplace diversity, we run a field experiment in a high-profile sector perceived as lacking in diversity. The financial services industry is widely thought of as dominated by white males—this is discussed, for ex- ample, in the management and sociology literatures (Rivera 2015) and regularly sur- faces in the popular press. 4This may be for good reason. In 2010, African Americans accounted for 2.7 percent of senior staff in financial services, while Hispanics accounted for 2.9 percent (Government Accountability Office 2013). In 2014, a survey by the Financial Timesfound 23.7 percent of senior roles in finance were occupied by 4. A sample of headlines from the last three years includes, for example,“‘White Male’Culture at Banks Is Difficult Even for One of the Industry’s Top Leaders,”“Wall Street’s Young Bankers Are Still Mostly White and Male, Report Says,”“What It’s Like to Be Something Other Than White and Male in the Hedge Fund Business,”“The Financial Industry Doesn’t Want You to Know about Its Lack of Diversity,”“These Charts Show Just How White and Male Wall Street Really Is.”Also see Brymer (2016), Crowe and Kiersz (2015), Alden (2014), andFinancial Times(2017). Flory, Leibbrandt, Rott, and Stoddard 77 women, while a recent study of U.S. financial services firms put the percentage for women even lower—at 20–22 percent. 5Perceptions of lack of diversity therefore appear to have at least some support in the data. We take advantage of this setting to test whether different types of information can affect interest by underrepresented groups and recruiting outcomes.

The experiment is embedded in a recruitment drive to fill positions in a professional development program. This highly selective program introduces undergraduate freshmen and sophomores to careers in the financial industry and helps them build their profes- sional networks and strengthen important skills like interviewing, elevator-pitching, résumé design, etc. Acceptance provides a critical“foot in the door”opportunity for participants, substantially boosting their chances of post-graduate employment at the firm and their readiness to enter and advance in the financial sector more broadly. 6 Figure 1 illustrates the experimental protocol. The firm recruits participants by send- ing an email advertisement to its nationwide network of campus contacts, including career services centers, student organizations, individual business schools, departments, and job boards. The email (shown in Online Appendix Figure A1 ) consists of a brief announcement, along with a hyperlink to a webpage that has information about the firm, the program, and how to apply. Individuals who click on the hyperlink in the announcement first enter their name and university, are then randomized into one of the treatments, and then proceed to the landing page where the treatment is delivered.

More precisely, we use the landing page to test the effects of different types of infor- mation on increasing interest from underrepresented groups. A randomly assigned statement at the top of the page either (i) provides information about the value the Control Figure 1 Overview of Experimental Protocol Notes: Treatments are randomly assigned at individual level.

5. SeeFinancial Times(2017, https://ig.ft.com/managements-missing-women-data/) and Jaekel and St-Onge (2016).

6. In 2016, participants in this program were four times more likely than other applicants to be hired as an intern, the majority of whom become full-time employees after graduation. Since the program is also designed to build the participants’professional image, acumen, and network in the financial sector more broadly, it is also likely to increase entrance to other major financial firms. 78 The Journal of Human Resources firm places on diversity among its employees, (ii) encourages individuals from a broad range of fields of study to apply, or (iii) includes a neutral statement to serve as our control condition. Statements were randomly assigned at the individual level. 7 Besides the treatment script, the webpage is identical for all individuals who visit it.

(See Online Appendix Figure A2 .) C. Treatments The treatments are designed to identify best practice and to test the three hypotheses outlined above, plus a minor hypothesis on whether it matters if the reason for wanting a diverse group of employees is its impact on firm performance or instead its inherent value to the firm’s culture. Table 1 summarizes the treatment messages. 8(For the complete text Table 1 Experimental Design Message Type Treatment Support Type Description Control T0 N/A A diversity-neutral statement Diversity I (direct diversity signal)T1 Nonsupported Statement emphasizing firm’s valuation of diversity as a competitive advantage that raises firm productivity T2 Fact-supported T1 plus listing of exact percentages of women and ethnically diverse new hires in the prior year Diversity II (direct diversity signal)T3 Nonsupported Statement emphasizing firm’s valuation of diversity as a key part of the company’s culture T4 Fact-supported T3, presented as quote spoken by the company’s CEO, proving endorsement at the highest levels of company leadership Major (indirect diversity signal)T5 Nonsupported Statement encouraging individuals from a variety of fields of study T6 Fact-supported T5 plus listing the exact number of majors represented in last year’s program Notes: Individuals randomly received one of seven different treatment statements. The experiment follows a three by two design. There are three main message types (Diversity I, Diversity II, Major), and for each there are two support types (fact-supported and nonsupported). For the exact wording of each treatment, see Online Appendix Table A1 . 7. We removed a few students from the sample who entered the landing page multiple times and were exposed to different messages.

8. Besides the treatment language, the webpage for each treatment is identical. Flory, Leibbrandt, Rott, and Stoddard 79 in each treatment, see Online Appendix Table A1 .) Our experiment follows a three-by- two design, with three main message types, each split into a fact-supported and a non- supported version. The first two message types signal the value of diversity to the firm:

Diversity Iemphasizes its value to firm productivity, andDiversity IIemphasizes its value to firm culture. These signals are direct, communicating a desire by the firm to employ people from a variety of backgrounds and a high value placed on diversity per se. These treatments enable us to test Hypothesis 1.

The third message type tests an indirect approach to see if a more diverse pool can be attracted without explicitly appealing to applicant origins or social background, or even mentioning diversity per se. This message type (Major) attempts to encourage interest from ethnic minorities and women by appealing to a broader variety in major field of study than is typically associated with finance careers, citing subject areas with greater representation by ethnic minorities and women—such as ethnic studies, nursing, and psychology. 9This treatment is aimed at examining Hypothesis 2.

Finally, we split each of the three message types into one treatment with no evidence to support its broader claim (nonsupported) and one that presents evidence to back up the statement (supported). The facts used are the high percentage of recent hires accounted for by ethnic minorities and women (Diversity I), a direct quote from the company’s CEO showing endorsement of the statement at the highest levels of company leader- ship (Diversity II), and information on breadth of student majors among past successful applicants (Major). This treatment allows us to examine Hypothesis 3.

At the bottom of the webpage, after seeing one of the seven messages, individuals can click on a hyperlink to learn more about the program and submit their application. By linking with the firm’s applicant tracking database, we are able to capture the behavior of each individual who lands on the webpage—which treatment message she sees, whether she expresses interest in the program by starting an application or asking to be notified of similar events, whether she submits the application, and whether she is selected to participate in the program. Employees at the firm selecting participants from the pool of candidates do not know which treatment applicants received.

D. Outcome Variables We examine three critical recruiting outcomes to analyze the impact of the different statements on race and gender diversity: (i) the percentage of different types of individuals who express interest by beginning an application or asking to be notified of similar future events, (ii) the percentage of different types of individuals who complete and submit an application to the program, and (iii) the percentage of different types of individuals selected for the program. We also look at two measures for each variable that are key to assessing impacts on diversity outcomes: the gap between underrepresented groups and overrepresented groups, and the behavior of each group (underrepresented and over- represented) considered on its own. 9. In 2015, for example, the proportion of majors accounted for by African and Hispanic Americans rises 62 percent in moving from business to ethnic studies and rises 25 percent in moving from business to psychology (13 and 7 percentage points, respectively), while the proportion of majors accounted for by women rises 64 percent in moving from business to psychology and 51 percent in moving from business to ethnic studies (30 and 24 percentage points)(National Center for Education Statistics 2018). 80 The Journal of Human Resources The signal of interest provides the most complete measure of treatment effects of the statements on the appeal of the opportunity to individuals, since the decision to submit an application conditional on interest is likely driven by specifics of the pro- gram such as its date or location. 10However, in addition to applicant interest, submitted applications are perhaps just as important to organizations and to diversity recruiting efforts, so we report the effects on both. We also examine evidence on whether the impacts on interest among underrepresented/minority groups translate to impacts on their representation among selected candidates. This last measure is an indicator for the qualifications of the marginal candidates attracted by the treatments, since the selection committee was blind to the treatments.

E. Definition of Underrepresented Groups and Identification of Diversity Characteristics In this study, we place particular emphasis on certain underrepresented races. In the work setting and related entry-level positions in the financial industry that we examine, the two groups that are by far the most underrepresented and highest priority for the firm’s diversity recruiting objectives are African Americans and Hispanics. 11 While women and Asians are also underrepresented in financial industry leadership roles at the upper levels, it is not clear they are underrepresented in entry-level up to mid-level leadership positions in the financial industry. 12 To identify ethnicity/race (and gender), we use two independent sets of data. First, we use data from self-reports. Individuals had the possibility to self-identify their ethnicity/ race (and gender) during the application process, and 285 individuals did so. Second, we employed an independent research assistant to code the ethnicity. To do so, the research assistant found each individual on social networking sites using their name and uni- versity. Ethnicity was determined based on information such as native language, school clubs and societies of which they were a member, hometown, profile pictures, etc. When ethnicity was not immediately clear from this information, other publicly available personal websites, blogs, or news articles were used. At times, research into the ety- mology of names was also considered to help determine ethnicity. We use the same 10. For example, some percentage of those who would otherwise apply will learn they have conflicts with the date of the program or decide it is located too far away and thus ultimately will not complete and submit their application.

11. Other significantly underrepresented (but much smaller) groups are, for example, Native American and Hawaiian. However, due to their very small numbers, we do not separately analyze their behavior and instead pool them in the group“other”ethnicities, together with Asians, white, two or more races, and nonidentified.

The vast majority of this group (92 percent) is composed of whites and Asians, which are not underrepresented in the finance sector. The results for whites and Asians only are very similar. See Online Appendix Tables A2, A5, and A6 for further details on ethnic composition of candidates.

12. In 2015 Asian Americans accounted for about 5 percent of the U.S. labor force (Bureau of Labor Statistics:

https://www.bls.gov/opub/reports/race-and-ethnicity/2015/home.htm), 11 percent of financial industry pro- fessionals, and 8 percent of first- and mid-level management (EEO: https://www1.eeoc.gov/eeoc/statistics /employment/jobpat-eeo1/2015/index.cfm#centercol; https://www1.eeoc.gov/eeoc/statistics/employment/jobpat -eeo1/2015/index.cfm#select_label); while women accounted for about 47 percent of the US labor force, (U.S.

Department of Labor: https://www.dol.gov/wb/stats/NEWSTATS/facts/women_lf.htm#one), 52 percent of financial industry professionals, and 48 percent of first- and mid-level management (same EEO references as above). Women make up approximately 45 percent of the S&P 500 companies’workforce (Catalyst 2015), and Asians hold about 47 percent of professional jobs in Silicon Valley tech companies and roughly a quarter of finance sector professional employment (U.S. Equal Employment Opportunity Commission 2015). Flory, Leibbrandt, Rott, and Stoddard 81 categories for the variables ethnicity/race as used by the company in its application form.

We coded ethnicity as follows: Asian; Black or African American; Hispanic or Latino; White; Two or More Races; Native, Hawaiian, or Other; and Cannot Tell.

As a check on the accuracy of the second approach in identifying race (and gender), we compare the race data identified through this process to the 285 individuals who applied and for whom we already had self-identified race. We find that all individuals who we identified as African American or Hispanic through this approach also self- identified as such or as having two or more races, that more than 99 percent of those identified through this process as female also self-identified as female, and that more than 98 percent of those identified as male also self-identified as male—this suggests that we were able to quite accurately identify ethnicity and gender. 13 III. Experimental Findings A. Overview and Global Effects In total, the experiment generated a relatively large sample of 1,121 individuals, with a substantial number belonging to the underrepresented group (N=166, 14.8 percent of the sample: 6.3 percent African Americans, 8.5 percent Hispanic). 14 As expected, Asians (47.7 percent) and women (48.4 percent) are not underrepresented in our sam- ple. 15Of the 1,121 individuals, 385 signal interest after viewing the treatment message and program information (62 African Americans and Hispanics and 211 women), 285 submit applications (50 African Americans and Hispanics and 156 women), and 47 are selected to participate in the program (15 African Americans and Hispanics and 33 women). Randomization of individuals into the treatment cells resulted in a fairly well-balanced distribution—each message was seen by approximately 160 individu- als, ranging from 150 individuals who saw theDiversity I(supported) message to 165 individuals who saw theDiversity II(nonsupported) message. About 15 percent of each treatment group is African American or Hispanic (ranging from 13 percent to 18 percent across the seven groups), and about 48 percent of each treatment group is women (from 13. We also find that the Spearman’s correlation coefficient across the values for race is 0.80 (p=0.000) when looking across all ethnicities–including white, Asian, Native/Hawaiian/Other–and is 0.96 (p=0.000) for gender (see Online Appendix Tables A5 and A6 ). Note that any measurement error in the identification of race and gender (that is, noise) makes itmoredifficult to identify the impact of treatments targeted to affect racial/ gender minorities. Throughout the analysis, we use the coded information. Given the high correlation between coded and self-identified ethnicity and gender, we are confident that we capture the treatment effect on ethnic minorities and women.

14. While our sample size of subjects from underrepresented groups may appear somewhat limited at first glance, we note that 166 individuals is relatively large compared to total sample sizes in many typical laboratory experimental studies. One of the advantages of lab experiments is that they draw from a population with less variation than is common in field experiments (university undergraduates) and are thus able to obtain relatively precise estimates even with small samples. In this field experiment, we also fortunately benefit from this advantage, as our sample is drawn from undergraduates.

15. Overall, 1,264 individuals clicked on the hyperlink in the email announcement, entered their name and university, and proceeded to the landing page where the treatment was delivered (33 individuals clicked on the hyperlink but did not pass through all the way to the landing page). However, we had to exclude 135 individuals from the sample who logged in multiple times from different IP addresses and saw different treatment messages and eight individuals who saw the message after they had already applied to the program. This leaves us with a total sample of 1,121 individuals who saw the control or one of the six treatment messages. 82 The Journal of Human Resources 41 percent to 53 percent), with no significant differences in percentages across treatment groups. (See also Online Appendix Table A2 .) We first examine impacts of the statements on interest of all individuals, ignoring demographic characteristics. We find that interest in the company and/or a career in the financial sector is higher in all six treatments compared to the control. Pooling all six treatment messages together, we see they raise the overall proportion of individuals interested in the program by 25 percent (seven percentage points), from 28 percent of those who visited the landing page with the status quoControlmessage, to 35 percent of those who visited the landing page with one of the six messages communicating employer prioritization of diversity among its employees (w 2-test,p=0.057). Splitting by the two different types of approaches to attracting employee diversity (Diversityvs.

Major), we find they each lead to similar positive impacts on individuals overall—a seven percentage point rise forDiversity(w 2-test,p=0.076) and an eight percentage point rise for theMajormessages (w 2-test,p=0.069).

B. Effects on Underrepresented Groups We now turn to impacts of the statements on the underrepresented groups in recruiting outcomes. There are two distinct and important dimensions with respect to treatment effects on demographic composition of candidates and recruits: impacts on thegap between underrepresented and nonunderrepresented and impacts onindividualsfrom underrepresented and nonunderrepresented groups. We examine both. We also first pool theDiversitymessages separately from theMajormessages since they represent two fundamentally different approaches to raising interest among underrepresented groups and because we find little difference in impacts within theDiversitycategory overall (see further below).

Figure 2 illustrates the gap between underrepresented and nonunderrepresented ethnicities in the proportion of individuals who express interest (Panel A), the pro- portion who submit an application (Panel B), and the proportion who are selected (Panel C) in each of the three main message types (Control,Diversity, andMajor). We see that in theControlcondition, underrepresented ethnicities were 13 percentage points less likely to express interest than nonunderrepresented, five percentage points less likely to apply, and did not differ in the rate that they were selected. Turning to Figure 2 Impact of Messages on Ethnicity Gap Notes: Difference between percentage of underrepresented and nonunderrepresented individuals expressing interest (Panel A), submitting an application (Panel B), and being selected (Panel C). 90 percent confidence intervals are calculated with robust standard errors. Flory, Leibbrandt, Rott, and Stoddard 83 theDiversitycondition, the picture sharply changes: individuals from underrepresented groups are now 15 percentage pointsmorelikely to be interested, 14 percentage points morelikely to apply, and ten percentage points more likely to be selected. That is, the Diversitycondition changes the gap by 28 percentage points (p=0.006; OLS with robust SE) in favor of underrepresented groups for expressing interest, by 19 percentage points (p=0.052) for applying (in both cases, not only closing, but reversing the gap), and by ten percentage points (p=0.073) for being selected.

Table 2 reports results from three linear probability models that examine the ro- bustness of these findings, accounting for gender of the candidate. Panel A shows coefficient estimates for regressions of the outcomes on gender, ethnicity, treatment type, as well as ethnicity-by-treatment and gender-by-treatment interactions. The coefficient estimates for the interaction termAfrican/Hispanic American·Diversity represent the estimated impacts of theDiversitytreatment on the ethnicity gap for each outcome variable (that is, the difference-in-difference estimates–the amount that Diversityincreases each outcome variable for underrepresented ethnicities, relative to nonunderrepresented). As the estimates show,Diversityraises the percentage of un- derrepresented groups expressing interest by 30 percentage points (p<0.01), the per- centage applying by 21 percentage points (p<0.05), and the percentage being selected by 11 percentage points (p=0.053), above that of nonunderrepresented.

Turning to how the treatments impact behavior and outcomes for individuals, Figure 3 illustrates the effects among the two different groups (African American/Hispanic and other ethnicities). It also sheds light on what is driving the above effects on the ethnicity gap for each outcome. The top row reports the percentage of African Americans and Hispanics who express interest (Panel A), submit applications (Panel B), and are se- lected (Panel C) depending on the three main message types in the job advertisement (Control,Diversity,Major).

As we see in the Panels A and B, African Americans and Hispanics strongly respond to theDiversitymessages. Panel A shows that their interest in the open position almost triples in moving from theControlto theDiversitymessage. While only 17 percent of ethnic minorities signal interest inControl, 48 percent do so inDiversity(w 2,p=0.005).

Further, in Panel B we observe that the proportion of ethnic minorities submitting an application more than doubles in moving fromControltoDiversity, going from 17 percent to 38 percent (w 2,p=0.05).

The encouragement effect ofDiversityon African and Hispanic Americans also translates into a threefold increase in the likelihood of being selected, compared to their likelihood of being selected inControl, and induces a sixfold increase in the likelihood of being selected relative to nonunderrepresented individuals. This provides indirect evidence that theDiversitytreatments not only encourage applications from African Americans and Hispanic Americans per se, but from highly qualified African Ameri- cans and Hispanic Americans, as the selection committee is blind to treatment and does not know which message type each applicant sees. More precisely, while only 4 percent of underrepresented ethnicities inControlare selected for openings, 12 percent of ethnic minorities inDiversityare selected (w 2,p=0.25). 16 Furthermore, while the same 16. The share of selected applicants is another way of comparing qualifications of applicants across condi- tions. If theDiversitymessages attract highly qualified African Americans/Hispanics, the share of selected African American/Hispanic applicants inDiversityshould be at least as large as inControl(andMajor). This is indeed the case: while 33 percent of African American/Hispanic applicants are selected inDiversity, the shares 84 The Journal of Human Resources Table 2 Probability of Expressing Interest, Submitting Application, and Being Chosen Interested Applied Selected (1) (2) (3) Panel A: Treatment Effects on Ethnicity Gap, Gender Gap, and Overall Diversity 0.0336 0.0222-0.0109 (0.0539) (0.0480) (0.0198) Major 0.0892 0.0674-0.0130 (0.0613) (0.0550) (0.0199) African American/Hispanic-0.145-0.0614-0.00758 (0.0884) (0.0863) (0.0436) African American/Hispanic·Diversity 0.301*** 0.207** 0.107* (0.104) (0.101) (0.0552) African American/Hispanic·Major 0.0271-0.00641 0.0155 (0.114) (0.109) (0.0550) Female 0.106 0.0798 0.0430 (0.0713) (0.0655) (0.0327) Female·Diversity-0.0128-0.0141-0.0207 (0.0806) (0.0740) (0.0361) Female·Major-0.0312-0.0191 0.0252 (0.0890) (0.0819) (0.0398) Constant 0.249*** 0.182*** 0.0241 (0.0472) (0.0421) (0.0179) Panel B: Treatment Effects on Ethnic-Minority Individuals Diversity Effect on African American/Hispanic Men 0.335*** 0.229** 0.096* [F-testp-value] [0.0014] [0.0252] [0.0607] Diversity Effect on African American/Hispanic Women 0.322*** 0.215** 0.0756 [F-testp-value] [0.0012] [0.0262] [0.1941] Major Effect on African American/Hispanic Men 0.1163 0.0610 0.00252 [F-testp-value] [0.3092] [0.5799] [0.9601] Major Effect on African American/Hispanic Women 0.0851 0.0419 0.0277 [F-testp-value] [0.4381] [0.6898] [0.6359] Observations 1,121 1,121 1,121 Notes: Estimates from a linear probability model. Dependent variables are dummy variables taking the value one if the individual expresses interest in the program (Model 1), submits an application (Model 2), or is selected (Model 3). The explanatory variables Diversity, Major, African American/Hispanic, and Female are dummy variables taking the value one if the individual is in the respective treatment or demographic group. Diversity Effect on African American/Hispanic Men is the sum of the coefficients for Diversity and African American/Hispanic·Diversity, and Diversity Effect on African American/Hispanic Women is the sum of the same coefficients plus the coefficient for Female·Diversity. Major Effect on African American/Hispanic Men is the sum of the coefficients for Major and African American/Hispanic· Major, and Major Effect on African American/Hispanic Women is the sum of the same coefficients plus the coefficient for Female·Major. Robust standard errors in parentheses,F-testp-values in brackets; *p<0.1, **p<0.05, ***p<0.01. Flory, Leibbrandt, Rott, and Stoddard 85 proportion of underrepresented ethnicities as nonunderrepresented are selected for the program underControl(4 percent), under theDiversitymessage type underrepresented ethnicities are ten percentage points (500 percent) more likely to be selected than nonunderrepresented (w 2,p=0.000).

The estimates in Panel B of Table 2 confirm the positive and highly significant impact ofDiversitymessages on underrepresented ethnicities’interest (Column 1,p=0.001 for men,p=0.001 for women) and application rates (Column 2,p=0.025 for men, p=0.026 for women), compared to interest and application rates of underrepresented ethnicities in theControlcondition. Furthermore, Column 3 shows that for individuals from underrepresented groups who view the page, theDiversitymessages raise the likelihood that they will be selected for the program by an estimated 9.6 percentage points (p=0.061) for males and an estimated 7.5 percentage points (p=0.194) for females, compared to their likelihood of being selected under theControl.

To summarize, nonparametric tests and regression estimates both show substantial effects from theDiversitymessages on all three recruiting outcomes—interest, appli- cations, and selection. This is true both when looking at the impacts on the difference between underrepresented and nonunderrepresented ethnicities, as well as the effects on underrepresented ethnicities alone (ignoring the effects on nonunderrepresented).

Indeed, impacts on the gap are almost entirely driven by the effects on underrepresented group individuals. Finally, we find no significant differences in treatment effects be- tween African Americans and Hispanics (p>0.43). 17 Figure 3 Impact of Messages Notes: Percentage of individuals expressing interest (Panels A and D), submitting an application (Panels B and E), and being selected (Panels C and F). The top (bottom) row shows the respective percentages for African Americans and Hispanics (other individuals). 90 percent confidence intervals are calculated with robust standard errors.

are 17 percent and 40 percent inControlandMajor, respectively (and 36 percent among African American/ Hispanic applicants who do not see any of the treatment messages).

17. In attracting candidate interest, for example, theDiversitystatements raise interest among African Americans by 40 percentage points for men (p<0.01) and 37 percentage points for women (p<0.01), and raise interest among Hispanic Americans by 29 percentage points for men (p<0.05) and 28 percentage points for 86 The Journal of Human Resources Several other notable findings are illustrated in Figures 2 and 3. First, we observe in Figure 2 that theMajortreatments appear to have no positive impact on the gap between underrepresented ethnicities andnonunderrepresented for any of the three outcomes, a finding confirmed in the regression results in Panel A of Table 2, where the estimates for the interaction termAfrican/Hispanic American·Majorare quite small and not significantly different from zero. Turning to Figure 3 helps us see why— the top row shows underrepresented ethnicities do not respond nearly as strongly to the indirectMajortreatments. While their expression of interest and submission of appli- cations are both higher inMajorthan inControl(increases of eight percentage points and three percentage points, respectively), the difference is not statistically significant (w 2-tests,p=0.428, andp=0.704). This is confirmed by the results shown in Panel B of Table 2, where the estimated impacts ofMajorare positive for all three outcomes, but never significantly so (regression estimateF-tests,p>0.309).

Figures 2 and 3 further illustrate the strong effect of the directDiversitymes- sages compared to the indirectMajormessages. As pointed out earlier, the ethnicity gap in rates of interest and applications reverses inDiversity, which is not the case inMajor.

In addition, while 48 percent of African Americans and Hispanics signal interest upon seeing aDiversitymessage, only 25 percent do so inMajor(w 2-test,p=0.010), and while 38 percent of African Americans and Hispanics apply underDiversity, only 20 percent do so underMajor(w 2-test,p=0.042).F-tests for the regressions in Table 2 (not shown) confirm the same pattern (p=0.014 andp=0.046 for African American/ Hispanic men,p=0.010 andp=0.042 for African American/Hispanic women). The difference between the impact ofDiversityand the impact ofMajoron the selection rates is in the same direction (stronger impacts forDiversity), but is less pronounced (w 2-tests,p=0.150; regression estimateF-tests,p=0.043 andp=0.338 for African American/Hispanic men and women, respectively).

Another noteworthy finding relates to the response to diversity appeals among in- dividuals from ethnic groups that are not underrepresented in this sector. In the bottom row of Figure 3, we see that there is no discouragement effect on rates of interest or applications from theDiversitymessages on overrepresented ethnic groups (whites and Asians in this setting). Ethnic nonminority candidates, upon seeing theDiversityand Majormessages, are even slightly more likely to signal interest (33 percent inDiversity and 38 percent inMajor, compared to 30 percent inControl) and to apply (24 percent inDiversityand 28 percent inMajor, compared to 22 percent inControl), although these differences are not significant (w 2-tests,p=0.483 andp=0.651 forDiversity, p=0.105 andp=0.166 forMajor). This is also visible in the regression estimates in Panel B of Table 2, which show positive and nonsignificant coefficients for the impacts ofDiversityandMajoron interest and application rates among both men and women in the nonunderrepresented group. This provides evidence that theDiversitytreatments benefit members of underrepresented groups and helps employer diversity objectives, without discouraging members of overrepresented groups. Combined with the findings discussed earlier on the positive overall impacts on interest across all individuals, this suggests little to no downside in recruiting outcomes from using the diversity appeals.

Figure 4 shows the patterns when considering theDiversity IandDiversity IItreatments separately. Overall, the difference in impacts is negligible for both underrepresented women (p<0.05), the difference in effects between African Americans and Hispanics not significant (p>0.50).

See Online Appendix 1 and Table A7 for full results on each ethnic minority group examined separately. Flory, Leibbrandt, Rott, and Stoddard 87 ethnicities (Panel A) and for individuals outside the underrepresented groups.

Similarly, impacts of each treatment on the ethnicity gap are also quite close to each other—moving the gap in favor of underrepresented ethnicity by 31 versus 25 per- centage points for interest, by 21 versus 17 percentage points for completed applica- tions, and by nine versus ten percentage points for being selected. Regression results reported in Online Appendix Table A3 confirm that there are no significant differ- ences in the effects on candidates from underrepresented ethnicities fromDiversity I compared toDiversity II, and that the same results discussed above forDiversityhold when considering the impacts of the two statements separately. We therefore use the pooledDiversitytreatments as the basis for our main findings, as discussed above.

C. Gender Findings Although women are not underrepresented in our setting, they may still respond to our treatments. Turning our analysis to women, we find they appear much less sensitive in general to the treatments as compared to underrepresented ethnicities. While their interest and application rates are higher inDiversityandMajorthan inControl, these differences are not significant at conventional levels. For example, the proportions of women who signal interest are 33 percent inControl, 40 percent inDiversity, and 40 percent inMajor,while the proportions who complete an application are 25 percent in Control, 29 percent inDiversity, and 30 percent inMajor(p>0.267 for pairwisew 2-tests for all four treatment–control comparisons). For applicant selection, 7 percent of all women who see theControlstatement are selected compared to 5 percent of those who see aDiversitystatement and 8 percent of those who see aMajorstatement. There is also little evidence of effects of the treatment statements on the gender gap. As shown in Table 2, the estimates for the gender-by-treatment interaction terms are small and not significantly different from zero. Figure 4 Proportion of Individuals Expressing Interest, Applying, and Being Selected Notes: Percentages are shown for African Americans and Hispanics (Panel A) and for other individuals (Panel B). 90 percent confidence intervals are calculated with robust standard errors. 88 The Journal of Human Resources This picture changes little when considering the impacts ofDiversity IandDiversity IIseparately. When looking at the comparisons betweenControlandDiversity I, and ControlandDiversity II, interest and application rates among women, as well as se- lection by the firm, are once again generally higher than in theControl, but none of the differences are statistically significant at conventional levels (w 2-tests,p>0.21). In addition, neither theDiversity Inor theDiversity IItreatment has a statistically signif- icant impact on the gender gap across the three outcomes, as shown by the coefficient estimates for the gender-by-treatment interaction terms in Online Appendix Table A3 .

D. Factual Support The third main hypothesis this experiment was designed to test is that adding concrete facts or statistics to diversity or inclusivity statements increases their impact by raising their credibility. Interestingly, we find little evidence that providing factual support for the pro-diversity orientation of the firm has an effect. Pro-diversity statements without supporting evidence were no less effective than those with supporting evidence, whether facts about diversity among recent recruiting outcomes were presented as inDiversity I orMajoror revealed as a quote by the company’s CEO as inDiversity II. Pairwise tests of the difference between the fact-supported and nonsupported statements within Diversity I,Diversity II, andMajorshow no significant difference for signaling inter- est (underrepresented ethnicities:p>0.73; women:p>0.21; all individuals:p>0.29), for applying (underrepresented ethnicities:p>0.29; women:p>0.26; all individu- als:p>0.24), and for being selected (underrepresented ethnicities:p>0.17; all indi- viduals:p>0.39). The only exception is among women for the selection outcome, where a higher proportion of women who saw the fact-supported version ofDiversity I were selected for the program compared to those who saw the nonsupported version (6.35 percent compared to 0 percent,w 2-test,p=0.02). (The difference in selection rates for women between fact-supported and nonsupported underDiversity IIandMajor are not significant,p>0.25.) Linear probability model regression results reported in Online Appendix Table A4 confirm these results.

One potential explanation for the lack of observed impact of fact-supported state- ments is that their impact may depend on the position advertised. For example, the CEO statement, provided in Treatment 4, may be less important for students for a one-day professional development program than for job-seekers for full-time, high-profile po- sitions. Another potential explanation is that the prospective applicants did not perceive the additional information to be more informative than the statement itself. For example, listing the share of diverse new hires in Treatment 2 may not help if the share is about equal to what the candidates expect—this new information may not change their prior beliefs. While our current experimental design does not allow us to disentangle com- peting explanations for the ineffectiveness of the fact-supported statements, this pro- vides an important direction for future research. IV. Discussion and Conclusion Companies have tried different strategies to increase diversity, and there is some evidence that affirmative action, such as quotas (Balafoutas and Sutter 2012; Niederle and Vesterlund 2013; Ibanez and Riener 2018), mentoring and advice Flory, Leibbrandt, Rott, and Stoddard 89 institutions (Blau et al. 2010; Rodriguez-Planas 2012; Brandts, Groenert, and Rott 2015) or changes in remuneration structure (Niederle and Vesterlund 2007; Flory, Leibbrandt, and List 2014), can influence job-seeker sorting in the labor market, with impacts on gender and ethnic diversity in the workplace. The findings we report highlight the value of a far less invasive approach, using simple adjustments in language to signal important dimensions of a firm’s value of employee diversity. Our results suggest that signals valuing workplace diversity have important implications for net impacts on the size and demographic composition of applicant pools, and more gen- erally for job-seekers heading into high-profile careers.

The implications of our results are important in several regards. On the one hand, this method of using statements to signal active valorization of diversity offers em- ployers and organizations an easy and cost-effective tool to enhance diversity in sit- uations where using quotas or altering compensation structures isnot feasible, costly to implement, or undesirable for other reasons. The fact that diversity messages increase the proportion of ethnic-minority applicants who are actually selected suggests the messages not only pull in more ethnic minority applicants, but pull in those who are strong candidates. On the other hand, identifying a method not requiring resource- intensive interventions and not linked with financial incentives has tremendous value.

It points toward a range of potential applications far beyond firms and labor markets:

since our results stem from simple changes in language and signals of values, they may speak to a broad range of environments where lack of diversity is a concern. Enroll- ment in higher education, political participation (voting, running for election), and civic engagement are just a few examples where language and signals might be lev- eraged to increase interest among underrepresented groups and break up monocultures (or prevent them from forming) in other spheres critical for the functioning of a healthy democratic society.

Finally, our finding that diversity messages sharply raise interest and application rates by African Americans and Hispanic Americans, but do not dissuade nonunderrepresented ethnicities is important. It suggests that this approach can lead to an increase in diversity without a dip in applicant quality or adverse effects on the majority group. In closing, it is important to bear in mind that our results highlight effective approaches to attract indi- viduals from underrepresented minorities, but have less to say about how to retain them.

When it comes to sustainably changing the demographic composition of employees in an organization, hiring is a critical piece, but forms part of a larger overall endeavor with other dimensions that must also be addressed in order to create lasting change.

References Alden, W. 2014.“Wall Street’s Young Bankers Are Still Mostly White and Male, Report Says.” New York Times, September 30.

Apfelbaum, E., N. Stephens, and R. Reagans. 2016.“Beyond One-Size-Fits-All: Tailoring Diversity Approaches to the Representation of Social Groups.”Journal of Personality and Social Psychology111(4):547–66. http://dx.doi.org/10.1037/pspi0000071 Ashraf, Nava, Oriana Bandiera, and Scott S. Lee. 2014.“Do-Gooders and Go-Getters: Career Incentives, Selection, and Performance in Public Service Delivery.”STICERD Economic Organization and Public Policy Discussion Papers Series 54. 90 The Journal of Human Resources Balafoutas, Loukas, and M. Sutter. 2012.“Affirmative Action Policies Promote Women and Do Not Harm Efficiency in the Laboratory.”Science335(6068):579–82.

Bertrand, M., and E. Duflo. 2017.“Field Experiments on Discrimination.”InHandbook of Economic Field Experiments, Volume 1, ed. Abhijit Vinayak Banerjee and Esther Duflo, 309– 93. New York: Elsevier.

Blau, F.D., J.M. Currie, R.T. Croson, and D.K. Ginther. 2010.“Can Mentoring Help Female Assistant Professors? Interim Results from a Randomized Trial.”American Economic Review 100(2):348–52.

Blau, F., and L. Kahn. 2017.“The Gender Wage Gap: Extent, Trends, and Explanations.”NBER Working Paper 21913. Cambridge, MA: NBER.

Brandts, J., V. Groenert, and C. Rott. 2015.“The Impact of Advice on Women’s and Men’s Selection into Competition.”Management Science61(5):1018–35.

Brymer, R.A. 2016.“What New Hiring Methods Say about Wall Street’s Diversity Problem.” Fortune, June 20.

Buser, T., M. Niederle, and H. Oosterbeek. 2014.“Gender, Competitiveness, and Career Choices.” Quarterly Journal of Economics129(3):1409–47.

Catalyst. 2005.“2005 Catalyst Member Benchmarking Report.”https://www.catalyst.org/wp -content/uploads/2019/02/2005MemberBenchmarkingReport.pdf (accessed February 11, 2020).

———. 2015.“Pyramid: Women in S&P 500 Companies.”http://tinyurl.com/qcnwko5 (accessed February 11, 2020).

CEB Talent Daily. 2017.“Diversity Backlash: Lean in Report Highlights America’s Blind Spots on Gender Inequality.”October 11.

Crowe, Portia, and Andy Kiersz. 2015.“These Charts Show Just How White and Male Wall Street Really Is.”Business Insider, August 25.

Dal Bó, Ernesto, Frederico Finan, and Martín A. Rossi. 2013.“Strengthening State Capabilities:

The Role of Financial Incentives in the Call to Public Service.”Quarterly Journal of Eco- nomics128(3):1169– 218.

Del Carpio, L., and M. Guadalupe. 2018.“More Women in Tech? Evidence from a Field Experiment Addressing Social Identity.”Unpublished.

Dobbin, F., and A. Kalev. 2016.“Why Diversity Programs Fail.”Harvard Business Review, August 21.

Emerson, Joelle. 2017.“Colorblind Diversity Efforts Don’t Work.”Harvard Business Review, September 11.

Financial Times. 2017.“Management’s Missing Women Series.”https://www.ft.com /management-women (accessed April 16, 2020).

Flory, J.A., U. Gneezy, K.L. Leonard, and J.A. List. 2018.“Gender, Age, and Competition:

A Disappearing Gap?”Journal of Economic Behavior and Organization150:256–76.

Flory, J.A., A. Leibbrandt, and J.A. List. 2014.“Do Competitive Workplaces Deter Female Workers? A Large-Scale Natural Field Experiment on Job Entry Decisions.”Review of Eco- nomic Studies82(1):122–55.

Gilbert, J.A., and J.M. Ivancevich. 2000.“Valuing Diversity: A Tale of Two Organizations.” Academy of Management Executive14(1):93–105.

Giuliano, L., D. Levine, and J. Leonard. 2011.“Racial Bias in the Manager–Employee Rela- tionship: An Analysis of Quits, Dismissals, and Promotions at a Large Retail Firm.”Journal of Human Resources46(1):26–52.

Government Accountability Office. 2013.“Diversity Management: Trends and Practices in the Financial Services Industry and Agencies after the Recent Financial Crisis.”https://www.gao .gov/assets/660/653814.pdf (accessed April 21, 2020).

Green, Jeff. 2017.“Deloitte Thinks Diversity Groups Are Passé.”Bloomberg Businessweek, July 19. Flory, Leibbrandt, Rott, and Stoddard 91 Hansen, F. 2003.“Diversity’s Business Case: Doesn’t Add Up.”Workforce82(4):28–29.

Hinton, I., J. Howell, E. Merwin, S. Stern, S. Turner, I. Williams, and M. Wilson. 2010.“The Educational Pipeline for Health Care Professionals: Understanding the Source of Racial Differences.”Journal of Human Resources45(1):116–56.

Ibanez, M., and G. Riener. 2018.“Sorting through Affirmative Action: Three Field Experiments in Colombia.”Journal of Labor Economics36(2):437–78.

Jaekel, Astrid, and Elizabeth St-Onge. 2016.“Why Women Aren’t Making It to the Top of Financial Services Firms.”Harvard Business Reviewhttps://hbr.org/2016/10/why-women -arent-making-it-to-the-top-of-financial-services-firms (accessed April 21, 2020).

Kahneman D. 2011.Decisions, Fast and Slow. New York: Farrar, Straus, and Giroux.

Kalev, A., F. Dobbin, and E. Kelly. 2006. Best Practices or Best Guesses? Assessing the Efficacy of Corporate Affirmative Action and Diversity Policies.American Sociological Review 71(4):589–617.

Leibbrandt, A., and J. List. 2018.“When Equal Employment Opportunity Statements Backfire:

Evidence from a Natural Field Experiment on Job-Entry Decisions.”NBER Working Paper 25035. Cambridge, MA: NBER.

Marinescu, Ioana, and Ronald Wolthoff. 2016.“Opening the Black Box of the Matching Function: The Power of Words.”NBER Working Paper 22508. Cambridge, MA: NBER.

McKay, P.F., and D.R. Avery. 2005.“Warning! Diversity Recruitment Could Backfire.”Journal of Management Inquiry14(4):330–36.

National Center for Education Statistics. 2018.“Baccalaureate and Beyond.”https://nces.ed .gov/ (accessed February 11, 2020).

Niederle, M., and L. Vesterlund. 2007.“Do Women Shy Away From Competition? Do Men Compete Too Much?”Quarterly Journal of Economics122(3):1067–1101.

Niederle, M., C. Segal, and L. Vesterlund. 2013.“How Costly Is Diversity? Affirmative Action in Light of Gender Differences in Competitiveness.”Management Science59(1):1–16.

Preece, J., and O. Stoddard. 2015.“Why Women Don’t Run: Experimental Evidence on Gender Differences in Competition Aversion.”Journal of Economic Behavior and Organization 117:296–308.

———. 2016.“Run, Jane, Run! Gendered Responses to Political Party Recruitment.”Political Behavior38(3):561–77.

Rivera, L. 2015.“Firms Are Wasting Millions Recruiting on Only a Few Campuses.”Harvard Business Review.October 23, p. 1–8.

Rodriguez-Planas, N. 2012.“Longer-Term Impacts of Mentoring, Educational Services, and Learning Incentives: Evidence from a Randomized Trial in the United States.”American Economic Journal: Applied Economics4(4):121–39.

Shurchkov, Olga. 2012.“Under Pressure: Gender Differences in Output Quality and Quantity under Competition and Time Constraints.”Journal of the European Economic Association 10(5):1189–213.

Stanovich, K.E., and R.F. West. 2000.“Individual Differences in Reasoning: Implications for the Rationality Debate?”Behavioral Brain Sciences23:645–65.

Steele, C.M., and Joshua Aronson. 1995.“Stereotype Threat and the Intellectual Test Performance of African Americans.”Journal of Personality and Social Psychology69(5):797–811.

Toten, M. 2017.“Backlash against Diversity and Inclusion Gains Momentum.”WorkplaceInfo, May 3.

U.S. Equal Employment Opportunity Commission. 2015.“Diversity in the Finance Industry Report.”https://www.eeoc.gov/eeoc/statistics/reports/finance/index.html (accessed February 11, 2020). 92 The Journal of Human Resources Copyright ofJournal ofHuman Resources isthe property ofUniversity ofWisconsin Press and itscontent maynotbecopied oremailed tomultiple sitesorposted toalistserv without the copyright holder'sexpresswrittenpermission. However,usersmayprint, download, or email articles forindividual use.