Diversity Thresholds: How Social Norms, Visibility, And ...

r Academy of Management Journal2019, Vol. 62, No. 1, 144–171.https://doi.org/10.5465/amj.2017.0440

DIVERSITY THRESHOLDS: HOW SOCIAL NORMS,VISIBILITY, AND SCRUTINY RELATE TO

GROUP COMPOSITION

EDWARD H. CHANGKATHERINE L. MILKMANUniversity of Pennsylvania

DOLLY CHUGHNew York University

MODUPE AKINOLAColumbia University

Across a field study and four experiments, we examine how social norms and scrutinyaffect decisions about adding members of underrepresented populations (e.g., women,racial minorities) to groups. When groups are scrutinized, we theorize that decisionmakers strive to match the diversity observed in peer groups due to impression man-agement concerns, thereby conforming to the descriptive social norm. We examine thisfirst in the context of U.S. corporate boards, where firms face pressure to increase genderdiversity. Analyses of S&P 1500 boards reveal that significantly more boards includeexactly two women (the descriptive social norm) than would be expected by chance.This overrepresentation of two-women boards—a phenomenon we call “twokenism”—ismore pronounced amongmore visible companies, consistent with our theorizing aroundimpression management and scrutiny. Experimental data corroborate these findingsand provide support for our theoretical mechanism: decision makers are discontinu-ously less likely to add a woman to a board once it includes two women (the socialnorm), and decision makers’ likelihood of adding a woman or minority to a group isinfluenced by the descriptive social norms and scrutiny faced. Together, these findingsprovide a new perspective on the persistent underrepresentation of women and mi-norities in organizations.

In recent years, many groups have faced negativescrutiny for their lack of diversity. For instance, theAcademy of Motion Picture Arts and Sciences facedbacklash in 2015 and in 2016 when all 20 actorsnominated for Academy Awards in the lead andsupporting acting categories were white. Thissparked an #OscarsSoWhite meme and a plan to

double female and minority membership in theAcademy by 2020 (Ryan, 2016). When Twitter madean initial public offeringwith nowomen on its boardof directors in 2013, the company faced an out-pouring of negative media attention, with numerousoutlets claiming that the lack of gender diversitywould cause problems for the company (Merchant,2013; Miller, 2013). And when Donald Trump an-nounced the members of his presidential cabinet in2017, the New York Times ran a front-page storytallying the women and racial minorities Trump’scabinet included and comparing its (lack of) di-versity to the demographic compositions of all othermodern U.S. administrations (Lee, 2017). These ex-amples illustrate that when groups lack diversity,negative scrutiny—or critical attention paid to partic-ular behaviors (Sutton & Galunic, 1996)—can ensue.

Little is known, however, about when a group’sdiversity will be judged negatively or how groups

We thank Mabel Abraham, John Beshears, Iris Bohnet,Gerard Cachon, Adam Galinsky, Peter Henry, Mary Kern,BlytheMcGarvie, Michael Norton, Devin Pope, Alex Rees-Jones, Deborah Small, Guhan Subramanian, and seminarparticipants at Cornell, Harvard, Johns Hopkins, UCLA,the University of Pennsylvania, AOM, BDRM, BSPA,ECDC, SJDM, SPSP, TADC, and Yale Whitebox for feed-back on this research. We also thank Cullen Blake for theidea that led to this paper and Rachel Tosney for excellentresearch assistance. Finally, we are grateful for supportfrom the Wharton Behavioral Lab and the Wharton RiskManagement and Decision Processes Center.

144

Copyright of the Academy of Management, all rights reserved. Contents may not be copied, emailed, posted to a listserv, or otherwise transmitted without the copyright holder’s expresswritten permission. Users may print, download, or email articles for individual use only.

https://doi.org/10.5465/amj.2017.0440

will respond to the possibility of negative scrutinyregarding their diversity. While scholarship hasestablished that diversity is not perceived objec-tively, or equivalently, by all observers and in allcontexts (Unzueta & Binning, 2010, 2012; Unzueta,Knowles, & Ho, 2012), it remains ambiguous as towhen group members and those perceiving groupsjudge a group’s diversity to be so insufficient as towarrant action or attention. Further, although pastwork has established that organizations respond toreputational threats such as social movement boy-cotts (King, 2008; McDonnell & King, 2013), it isunclear how those responsible for group composi-tion may behave when facing the threat of reper-cussions for displaying insufficient diversity. Inthis paper, we address these questions by analyzinga decade of data on the composition of U.S. corpo-rate boards in theStandard andPoor’s (S&P) 1500 andby conducting a series of supplemental experiments.

We propose that to avoid facing negative scrutiny,those responsible for forming groupsmay seek safetyin numbers by looking to the average behavior ofothers when setting implicit or explicit goals aboutthe diversity of groups. Descriptive social norms—defined as the average observed behavior of in-dividuals or groups in a population (Prentice &Miller, 1993)—have been shown to serve as refer-ence points for behavior in a variety of contexts,setting expectations about what is appropriate andeffective (Coffman, Featherstone, & Kessler, 2017;Goldstein, Cialdini, & Griskevicius, 2008; Nolan,Schultz, Cialdini, Goldstein, & Griskevicius, 2008),particularly in situations where appropriate behav-ior is ambiguous or uncertain (Festinger, 1954;Sherif, 1936). Decision makers and firms may thuslook to relevant others to understand what the de-scriptive social norms for diversity are, and theymaythen imitate these levels of diversity, both because ofthe reputational threat associated with negativescrutiny and because of uncertainty about what ad-equate diversity entails (DiMaggio & Powell, 1983).This behavior should be evenmore prevalent amonghighly visible groups or organizations because thenegative consequences of failing to conform can begreater for high-profile groups (Gardberg&Fombrun,2006). The actions of highly visible groups are morelikely to be scrutinized in the first place (Chiu &Sharfman, 2011), and organizations generally respondmore strongly to more visible threats (King, 2008).

We combine our theorizing about descriptive so-cial norms, scrutiny, andvisibilitywith past researchon goal setting to make a novel prediction. Specifi-cally, we predict that individuals responsible for

group compositions will respond to pressures todiversify in a similar fashion, leading to an over-abundance of groups with identical levels of di-versity. Past research has shown that goals—like thegoal to match the diversity of peer groups—are oftenhighly motivating (Locke & Latham, 2002), but in-dividuals relax efforts to achieve desirable outcomesafter reaching salient goal thresholds in many set-tings (Heath, Larrick, & Wu, 1999). This relaxing ofeffort has been shown to lead goal seekers’ perfor-mance to cluster around salient goal thresholds(Pope & Simonsohn, 2011). We predict that thistendency will lead scrutinized groups to clusteraround the social norm for diversity set by theirpeers. In other words, rather than continuing to in-crease diversity in response to external pressures(e.g., the threat of negative scrutiny), those with thepower to shape group diversity should be less likelyto increase the diversity of a group once the grouphas reached the descriptive social norm for diversityset by peers. This behavior will lead to improbablyhomogeneous diversity levels across groups.

We test our theorizing first in the context of U.S.corporate boards, a setting where firms face negativescrutiny for failing to include adequate gender di-versity (Merchant, 2013; Miller, 2013). Analyses ofS&P 1500 boards reveal that significantly moreboards include exactly two women (the descriptivesocial norm) than would be expected by chance,supporting our prediction that groups will respondto pressures to diversify in a similar fashion, lead-ing to an overabundance of groups with identicallevels of diversity at the descriptive social norm.This overrepresentation of two-women boards ismore pronounced among more visible companies,consistent with our theorizing on impressionmanagement and scrutiny. In additional studies,we experimentally manipulate descriptive socialnorms, scrutiny, and visibility to show that each ofthese influences group diversity decisions as ourtheory predicts. We find that these effects hold insettings beyond corporate boards and for socialcategories besides gender.

Our work provides a more complete under-standing of diversity-related hiring decisions, tellingus when women and racial minorities will be par-ticularly attractive candidates for inclusion ingroups and when groups will reduce their efforts toincrease diversity. Further, rather than focusing onlyon individual-level or firm-level explanations forwhywomen and racial minorities may ormay not beadded to groups, we highlight how external entitiessuch as peers (who help shape descriptive social

2019 145Chang, Milkman, Chugh, and Akinola

norms) and outsider scrutiny can shape group di-versity decisions. By illuminating these critical fac-tors that influence group diversity decisions, weprovide theoretical guidance about potential newways to improve diversity in organizations, and weprovide practical guidance to help predict whatlevels of diversity wemight expect to see in differentcontexts. Our research suggests that itmay behelpfulto increase scrutiny around diversity decisions andattempt to make other social norms, besides de-scriptive social norms, salient to decision makers inorder to increase the number of women and racialminorities selected into groups.

THEORY AND HYPOTHESES

Descriptive Social Norms

Descriptive social norms—defined as the averageobserved behavior of individuals or groups in apopulation (Prentice & Miller, 1993)—exert a potentinfluence on decisions. According to past research,descriptive social norms influence the behavior ofindividuals and groups for two primary reasons.First, they establish what is socially acceptable. Be-cause following the norm means avoiding outlierstatus, individuals and groups can feel reassured thatif existing norms are followed, social ostracism willnot ensue (Schultz, Nolan, Cialdini, Goldstein, &Griskevicius, 2007). By following a descriptive socialnorm, individuals and groups essentially insulatethemselves from the risk of being singled out becausethey are—by definition—doing what many of theirpeers are doing. Individuals, groups, and organiza-tions that negatively deviate from any descriptivenorm are muchmore likely to be singled out and facenegative consequences (Ahmadjian & Robinson, 2001;Zavyalova, Pfarrer, Reger, & Shapiro, 2012).

Second, descriptive social norms contain infor-mation aboutwhat behaviors are likely to be effectiveor adaptive (Cialdini, 2007). If the majority of othershave elected to partake in a specific action or be-havior (making it the descriptive social norm), thenthat signals that the norm may be a wise course ofaction (e.g., if everyone else is using this brand ofsoap, it must be a good brand of soap to use). Thissocial information is even more important when theappropriate behavior is unclear or when situationsare ambiguous or uncertain, as extant research hasshown that social norms affect behavior to a greaterdegree in such settings (Festinger, 1954; Sherif,1936). In effect, descriptive social norms can func-tion as heuristics for decision making, providing

a guide for appropriate or wise behavior in a widerange of situations.

By conveying bothwhat is appropriate andwhat islikely to be effective, descriptive social norms pro-duce powerful effects on judgments and decisions(Cialdini, 2003; Cialdini, Reno, & Kallgren, 1990). Alarge body of empirical evidence has shown thatdescriptive social norms serve as salient referencepoints for behavior in many contexts, ranging fromenergy consumption to job acceptance decisions(Coffman et al., 2017; Goldstein et al., 2008; Nolanet al., 2008).Wepropose that descriptive social normsshould influence decisions made about group di-versity just as they influence decisions in other con-texts. Next, we describe research on scrutiny andimpression management that illuminates why thoseresponsible for decisions influencing group diversitymay feel pressure to follow descriptive social norms.

How Scrutiny of Group Diversity May DriveConformity to Descriptive Social Norms

Scrutiny refers to obtrusive and critical attentionpaid to particular behaviors (Sutton & Galunic, 1996),and it can come from a variety of sources. For ex-ample, the media is one common source of scrutinycapable of influencing an organization’s reputationand value and shaping others’ perceptions of itslegitimacy. Naturally, organizations compete to re-ceive positive and avoid negative media exposure(Fombrun, 1995; Fombrun & Shanley, 1990; Pollock& Rindova, 2003). Scrutiny can also come fromother sources, such as shareholders (e.g., institu-tional investors placing pressure on firms to engagein socially responsible behaviors) and policymakers(e.g., through regulations and the imposition of re-wards or penalties for certain behaviors [Aguilera,Rupp, Williams, & Ganapathi, 2007; Campbell,2007]). The public also often directly scrutinizesorganizations, mobilizing in ways that may drawwanted or unwanted attention to particular behaviors(e.g., through social movement boycotts [McDonnell &King, 2013]).

In general, groups and organizations have strongincentives to avoid negative scrutiny. Negativescrutiny can be detrimental to reputation and legiti-macy (Desai, 2011), so in order to avoid negativescrutiny, groups frequently attempt to manage im-pressions around scrutinized behaviors (Bolino,Kacmar, Turnley, & Gilstrap, 2008; Elsbach, Sutton,& Principe, 1998). Impression management describesattempts bygroupsororganizations topositively shapehow they are perceived (Elsbach & Sutton, 1992), and

146 FebruaryAcademy of Management Journal

it may occur even in anticipation of the possibility ofnegative events. For example, Elsbach et al. (1998)documented how hospitals use anticipatory impres-sion management tactics in order to prevent potentialnegative scrutiny.

In recent years, scrutiny has increased surround-ing the diversity of groups. For example, the mediahas scrutinized companies for insufficient genderdiversity on their boards of directors (Merchant,2013; Miller, 2013); presidents for insufficient raceand gender diversity in their cabinets (Lee, 2017) andtheir U.S. Supreme Court nominees (Totenberg,2016); and the Academy of Motion Picture Arts andSciences for insufficient racial diversity among theirOscar nominees (Buckley, 2016; Ryan, 2016). Impor-tantly, scrutiny is often applied selectively: ratherthan simultaneously emphasizing racial, gender, andsocioeconomic diversity, for instance, scrutiny oftenfocuses more narrowly on a single dimension of di-versity. For example, while groups such as corporateboards have facedconsiderablenegative scrutiny for alack of gender diversity, there has been far less atten-tion to their lack of racial diversity.

Scrutiny surrounding diversity naturally moti-vates impression management concerns. An im-portant question, then, is how decision makerswho shape the composition of high-profile groupswithin organizations may seek to manage diver-sity in order to avoid negative scrutiny.We proposethat past research on descriptive social norms pro-vides key insights. If groups or organizations aremotivated to avoid negative scrutiny, then follow-ing the descriptive social norm for diversity essen-tially ensures that they will not be singled out forinadequate diversity. Further, because it is often un-clear what an “objective” benchmark for strong per-formanceshouldbe in the contextof decisions arounddiversity (Bell & Hartmann, 2007; Shemla, Meyer,Greer, & Jehn, 2016; Unzueta et al., 2012), descriptivesocial norms should be particularly informative inguiding behavior around diversity. Thus, groups andorganizations (and the decision makers responsiblefor their composition)may treat the descriptive socialnorm for diversity as a goal for impression manage-ment reasons.

The Implications of Descriptive Social Norms asDiversity Goals

Past research on goal setting offers insight intowhat will happen when those who shape groupcomposition share the same explicit or implicit goal.Goals serve as reference points, causing individuals

to expend considerable effort in the hope of achiev-ing an unmet goal and then to relax their efforts afterachieving it (Heath et al., 1999; Locke & Latham,2002). This has been shown to lead to performanceclusteringaroundsalient goal thresholds innumerouscontexts. For instance, professional baseball playersfinish seasons disproportionately oftenwith a battingaverage just above .300 (a salient threshold widelybelieved to separate good hitters from great ones[Moskowitz & Wertheim, 2011; Pope & Simonsohn,2011]), and marathon runners finish races dispro-portionately often in the minute right before salient,round-number thresholds (e.g., the minute just underthree hours [Allen, Dechow, Pope, & Wu, 2016]). Wetherefore expect toobserveanexcessmassor clusteringof groups at (or just above) the descriptive social normfor diversity.1

Hypothesis 1a. Groups’ diversity levels will clusterat (or just above) the descriptive social norm set bypeers for diversity.

While Hypothesis 1a pertains to group composi-tion, group composition is the result of decisionsregarding which members to add to a group. Ifreaching the descriptive social norm for diversity is agoal of those who shape group compositions, thenefforts to increase group diversity (in the form ofadding underrepresented group members) shoulddecline precipitously once the descriptive socialnorm for diversity is achieved. Empirically, thisrelaxing of effort after reaching a goal threshold hasbeen observed in several contexts. In the contextof baseball, as just mentioned, batters reduce theirat-bat appearances near the end of the season oncethey have exceeded the salient .300 batting aver-age threshold that separates good hitters from greatones (Pope & Simonsohn, 2011). In the context ofScholastic Assessment Test (SAT) scores, studentsare disproportionately less likely to retake theSAT once they surpass a salient threshold such as ascore of 1,000 (roughly the average score set by theCollege Board and a salient round number) (Pope &Simonsohn, 2011). In our context of diversity andgroup composition decisions, we predict that groupsare less likely to increase their diversity once they

1 Becausedescriptive social norms are averages, they arerarely whole numbers (e.g., the average number of womenper boardwas 1.36women in the S&P 1500 in 2013). Sincegroups cannot have fractional numbers of women or racialminorities, we expect clustering at “or just above” the de-scriptive social norm (i.e., at the smallest whole numberabove the descriptive social norm).


have already reached the descriptive social norm fordiversity established by peers.

Hypothesis 1b. Groups (and the individuals whoshape their composition) will add newmembers fromunderrepresented populations at a lower rate oncethey have surpassed the pertinent descriptive socialnorm for diversity.

Importantly, we only expect descriptive socialnorms to serve as goals when it comes to scrutinizeddimensions of diversity. Without any scrutiny on agiven dimension of diversity, there should be noimpression management motives and thus no desireto follow the descriptive social norm. For example,we would expect to find support for Hypotheses 1aand 1b when it comes to gender diversity in settingswhere inadequate gender diversity has been scruti-nized (e.g., on corporate boards), but not in settingswhere gender diversity has not been scrutinized.Thus, we propose that scrutiny (or the threat ofnegative scrutiny) is required in order to produce ourhypothesized clustering and threshold effects.

Hypothesis 2. Scrutiny moderates the effects of de-scriptive social norms on group diversity decisions.Specifically, descriptive social norms will only influ-ence group diversity decisions and outcomes whenscrutiny is present on a given diversity dimension.

The Moderating Role of Visibility

If groups and organizations manage impressionsaround diversity to avoid negative scrutiny, thistendency should be more pronounced among morevisible groups and organizations. We follow pastresearch and use the term “visibility” to describehow much attention individuals, groups, or organi-zations typically receive (Chiu & Sharfman, 2011),regardless ofwhy they are receiving this attention (asopposed to our use of the term “scrutiny,” whichrefers to attention paid to a particular behavior suchas a group’s gender diversity). When firms are morevisible (e.g., because they operate in more visibleindustries or because they have higher overall mediaexposure), they face greater external pressures toengage in legitimacy-seeking behaviors (Gardberg &Fombrun, 2006) and are alsomore likely to engage inlegitimacy-enhancing behaviors such as corporatesocial performance initiatives (Chiu & Sharfman,2011). For example, firms respond more to boycottswhen they receive more media attention (King,2008), and firms engage in more prosocial activitieswhen boycotts are more threatening because of in-creased media attention (McDonnell & King, 2013).

Past research has shown that conforming to de-scriptive social norms (i.e., mimicking the behaviorof peer firms) is one way to enhance legitimacy(DiMaggio & Powell, 1983), suggesting that descrip-tive social norms should influence the diversityof groups on scrutinized diversity dimensions to agreater degree when those groups are more visible.Further, the actions of more visible firms receivemore attention, which can magnify the negative con-sequences of failing to conform to social norms.

Past researchon individual judgment anddecisionmaking has made similar predictions regarding theeffects of visibility on conformity to descriptive so-cial norms. Social norms influence behavior to agreater degree when individuals and their behaviorsare more visible (Cialdini & Trost, 1998). In particu-lar, individuals tend to look to social norms to guidetheir behavior most frequently when the behavior inquestion is public or observable (Cialdini, Kallgren,& Reno, 1991; Cialdini et al., 1990; Kallgren, Reno, &Cialdini, 2000; Shaffer, 1983). For example, studieshave found that monitoring employees can improveconformity to ethical norms in the context of employeetheft (Pierce, Snow, & McAfee, 2015), monitoring canimprove conformity to hand hygiene norms in hospi-tals (Staats, Dai, Hofmann, & Milkman, 2016), and be-ing in a public setting (as opposed to a private setting)can make women more likely to conform to gendernorms regarding assertiveness (Swim & Hyers, 1999).On an individual level, we would thus expect moreconformity to descriptive social norms when out-comes aremore visible. Thus, research and theorizingon both individuals and firms have suggested thatmore visible groups should be more likely to conformto social norms around diversity on scrutinized di-versity dimensions.

Hypothesis 3. Visibility moderates the effects of de-scriptive social norms on groupdiversity decisions onscrutinized diversity dimensions. Specifically, morevisible groups will be more likely to follow the de-scriptive social norm for diversity on scrutinized di-versity dimensions than less visible groups.

OVERVIEW OF STUDIES

The remainder of this paper proceeds as follows.We begin by examining our hypotheses in the field,exploring whether they make accurate predictionsabout the compositionandevolutionofU.S. corporateboards. In Study 1A,we present analyses of S&P 1500board composition data from 2013 that test for excessclustering of corporate boards at the descriptive socialnorm for gender diversity (Hypothesis 1a). We also


examine whether this pattern is more extreme amongmore visible companies (Hypothesis 3). In Study 1B,we present analyses of board member additions todetermine whether boards are discontinuously lesslikely to add female directors once they have reachedthe descriptive social norm for gender diversity (Hy-pothesis 1b). In Study 1C, we run an online experi-ment to test for evidence of the same pattern ofdiscontinuities in board member selection found inthe field in Study 1B in a stylized hypothetical de-cision environment where we can randomize thenumber of women on a board and control for theavailabilityof qualified candidates (Hypothesis1b). InStudies 2A and 2B, we seek evidence that scrutiny,descriptive social norms about diversity, and goalthresholds influence the gender of group membersselected for open positions, and we experimentallymanipulate social norms and scrutiny to test Hy-potheses 1b and 2. Finally, in Study 3, we examinehowsocialnormsandgroupvisibility affect the raceofgroupmembers selected foropenpositions, andwedothisbyexperimentallymanipulating socialnormsandvisibility to test Hypotheses 1b and 3. Together, thesestudies help establish the external validity, internalvalidity, and generalizability of our theories.

STUDY 1: CORPORATE BOARDS

We first test our theories in the context ofU.S. corporate boards. This context is an importantorganizational setting that is economically signifi-cant, as boards control trillions of dollars. It is alsohighly relevant to policy, as in recent years numer-ous countries have passed laws about the gendercomposition of the corporate boards of public com-panies (Bainbridge & Henderson, 2014; Forbes &Milliken, 1999; Smale & Miller, 2015).

STUDY 1A: CLUSTERING OF U.S. CORPORATEBOARD COMPOSITIONS AROUND THE

SOCIAL NORM

In Study 1A, we analyzed the most recent availableS&P 1500 corporate board composition data (from2013) to test whether descriptive social norms influ-ence board composition. Given the importance ofscrutiny toour theoreticalmodel (seeHypothesis2),wefirst sought to establishwhich dimensions of corporateboard diversity faced scrutiny at the time of data col-lection. An analysis of news articles from 2013 in thenews database LexisNexis revealed that of 98 newspa-per articles that mentioned “board diversity,” 97%mentioned gender diversity, while 18% mentioned

racial or ethnic diversity (the second most frequentlymentioned social category). In addition, several coun-tries in Europe have recently passed laws mandatingminimum levels of gender diversity on the boards ofpublic companies under their jurisdiction (Smale &Miller, 2015), but no such laws have beenpassed aboutother types of diversity. Given that the majority of at-tention regardingdiversity on corporate boards focuseson gender diversity, in this study we tested for (andonlyexpected toobserve) socialnormeffectspertainingto the gender diversity of U.S. corporate boards.

On S&P 1500 corporate boards, the average num-ber of women was 1.36 in 2013, and this descriptivesocial norm received significant media coverage,with all newspaper articles in the LexisNexis data-base about board gender diversity in 2013 focusingon the average number or percentage of women onboards. We therefore expected to observe an excessof boards with exactly two women, as boards withtwo women just exceed the peer norm for gender di-versity (Hypothesis 1a). We also predicted that thisexcessofexactly twowomenperboardwouldbemoreprevalent amongmore visible companies—those thatreceive more overall media attention (Hypothesis 3).

Methods

Data. Our dataset was compiled by InstitutionalShareholder Services (ISS). The ISS Director Datawe analyzed contains detailed information about theboards of directors for 1,514 companies that representthe S&P Composite 1500, which is composed of threeindices: the S&P 500, the S&P MidCap 400, and theS&P SmallCap 600. The S&P 1500 represents roughly90% of the total U.S. stock market capitalization, andwe also focus on the far more visible subset of com-panies in theS&P500,2which represents roughly90%of the total market capitalization of the S&P 1500and 80% of the total market capitalization of the U.S.stock market (S&P Dow Jones Indices, 2015).

The ISS dataset we analyzed includes informationon the individual members of the boards of directorsfor each of the 1,514 companies in the S&PComposite1500, including each director’s name, gender, andethnicity.3 The dataset is updated annually, and for

2 AGoogle search for the term “S&P500” returns 400 timesas many results as a Google search for the term “S&P 1500,”andaGoogleScholarsearch for the term“S&P500” returns20times as many articles as a search for the term “S&P 1500.”

3 ISS data on director gender were complete, but in 31instances director ethnicity was missing or blank. Wemanually searched Google and company websites to fill inthese missing data.


our primary analysis we relied on the 2013 data asthemost recent data available to us as of June 5, 2015,when we first accessed the ISS database.

Additional datawere collected on each company’smedia mentions (from LexisNexis), industry (fromNASDAQ), year of initial public offering (fromBloomberg and company websites), market capital-ization (from the Center for Research in SecurityPrices and Google Finance), and percentage in-stitutional ownership (from Bloomberg), and thesedata were used to perform robustness checks andinvestigate the moderating effect of visibility.

Analysis strategy. To test Hypothesis 1a, we re-lied on a comparison of the actual distribution ofmale and female directors on corporate boards withthe distribution we would expect if those directorswere assigned to boards in a gender-neutral manner.We determined the expected distribution using aMonte Carlo simulation method (Rubinstein &Kroese, 2011). Specifically, we took existing 2013S&P 1500 and S&P 500 data on directors and boardseats from the ISS dataset and then randomly reas-signed directors to different boards, generating10,000 simulateddistributions of directors to boards.Because we randomly reassigned actual directors toboards in each of our simulations, these simulationsproduced the board composition distribution wewould expect to see if gender played no role in boardmember selection. In other words, given the avail-able pool of board seats and directors, our simula-tions told us howmany women we should expect tosee on each board if boards ignored gender whenselecting board members.

We reassigned existing directors in our simula-tions to provide a conservative test of whether thereexist anomalous sorting patterns of female directorsto boards.4 In each simulation, we took as given thenumber of boards, the size of each board, and the

number of board seats each director held based onthe statisticswe observed in the 2013 ISSdataset. Forexample, if company a had nine board members inthe ISS dataset, then in each simulation company awas assigned nine distinct board members. Simi-larly, if director Zed held two different board seats inthe ISS dataset, then director Zed ended each simu-lation holding seats on two different corporateboards.

Running this simulation 10,000 times producedrandomassignments of all directors to all boards thatreflected the same number of directors, number ofboards, and board sizes we observed in the ISSdataset. For each simulation result, we consideredhow many company boards were assigned zero fe-male directors, one female director, two female di-rectors, etc. We then calculated the mean of thesevalues across all 10,000 simulations. These meanstold us how many companies we would expect, onaverage, to observe with exactly zero, one, two, andso on female directors if available board seats in theISS dataset were randomly assigned to available di-rectors. Our simulations also told us how rare a givenassortment was, giving us bounds in the form ofconfidence intervals around each mean to indicatethe likelihood under random assignment that wewould observe a certain fractionof boards containinga specific number of women (e.g., in what fraction of10,000 simulations we obtained such a result).

Although this simulation strategy has been usedand validated in a number of empirical papers(e.g., Dezs}o, Ross, & Uribe, 2016; Gino & Pierce,2010),we also conducted placebo simulationswith acharacteristic other than gender to ensure that anyobserved deviations from our simulations on genderwere not an artifact of our simulation method (see“Robustness Checks”).

Results

Summary statistics.For companies in our dataset,the modal number of directors on a board was nine,themediannumberwas nine, and95%of companieshad between six and 14 directors. Because we wereinterested in understanding the distribution of theabsolute number of women on each board, boardswith outlier numbers of seats could have exertedundue influence on our analyses. For our primaryanalyses, we therefore trimmed our dataset to in-clude only companies with a total number of di-rectors in the middle 95% of the distribution,excluding companies with outlier numbers of di-rectors (i.e., fewer than six or more than 14), leaving

4 One common explanation for the limited number ofwomen on corporate boards is that there are not enoughqualified women to serve on boards. We thus assume theuniverse of people qualified to serve on boards consistsonly of thosewho actually sit on boards, so our simulationsgauge whether we find anomalous sorting even if we as-sume no more qualified women exist to serve on boards.This extremely conservative assumption is certainly in-correct, but given that the universe of qualified womenmust be larger than the set who already serve on boards,finding evidence of clustering at the social norm underour assumptions would be even more remarkable (sincerelaxing this assumption would make it easier for the ob-served gender distribution to deviate from our simulatedexpected distribution).


us with 1,441 companies to analyze. However, theresults of our analyses remain meaningfully un-changed in terms of magnitude and statistical sig-nificance if we repeat them without trimming theseoutliers (see Online Supplement5).

The 1,441 companies in our trimmed dataset in-cluded 13,440 distinct board seats and 11,185 distinctdirectors, as some directors held board seats on mul-tiple company boards. In our trimmeddataset, 84%ofdirectors held exactly one board seat, 13% held twoboard seats, 3%held three board seats, and fewer than1%held four or five board seats. Of the 11,185 uniquedirectors represented in our trimmed dataset, 14%(n 5 1,558) were female, and women held 15% (n 51,963) of the available board seats (see Table 1).Ninety-one percent (n 5 10,150) of directors wereCaucasian, 3.7% (n5 417)wereBlack, 3.0% (n5 335)were Asian, 1.7% (n5 192) were Hispanic, and 0.8%(n 5 91) were classified as belonging to a differentethnic group (see Table 1). The average age of the di-rectors in our trimmed dataset was 62.9 years, with astandard deviation of 8.9 years. Fifty-eight (4.0%) ofthe companies had female CEOs. See Table 2 for acorrelation matrix describing our data.

Do boards cluster around the descriptive socialnorm for gender diversity? Hypothesis 1a suggeststhat we should find an excess of boards with exactlytwo women (since the relevant descriptive socialnorm was that an average board in the S&P 1500 in-cluded 1.36 women in 2013 and an average board intheS&P500 included1.89women in 2013). Based onsimulations of the S&P 1500, there were 8% fewercompanies with no women than would be expected(p5 0.019), and consistent with Hypothesis 1a therewere 12%more boardswith exactly twowomen thanwouldbe expected (p5 0.008). Boards including otherfrequencies of women were in line with expectations(see Figure 1, Panel A). Similarly, for the S&P 500 andconsistent with Hypothesis 1a, there were 45% morecompanies with exactly two female board membersthan would be expected (p , 0.001). There were also45% fewer companies with no female boardmembersthan we would expect (p , 0.001), and boards in-cluding other frequencies of women again arose at therates expected (see Figure 1, Panel B). Thus, Hypoth-esis 1a is supported, and in light of the far higher visi-bility of S&P 500 companies than other companies inthe S&P 1500, these patterns provide suggestive evi-dence in support of Hypothesis 3.

To provide further support for Hypothesis 1a, weanalyzedadditionalhistoricaldataoncorporateboardcompositions to assess whether historical descriptivesocial norms also determined where clustering oc-curred. In years when the average number of womenper board (i.e., the descriptive social norm)was belowone, our theorizing predicts an overrepresentation ofboards with exactly onewoman (i.e., “tokenism,” or agroup including exactly one woman [Kanter, 1977]);in years when the average number of women perboard was between one and two (e.g., 1.36 womenper board in 2013), our theorizing predicts an over-representation of boards with exactly two women.We name the phenomenon whereby a group in-cludes exactly two women “twokenism,” which isa portmanteau of the number “two” and the term“tokenism” originally used by Kanter (1977). Werepeated our simulations using 12 years of historicaldata to seewhether thedescriptive social normdid infact predict where an excess of boards arose in eachdistribution.

We gathered additional data on the compositionof S&P 1500 boards from 2002 to 2012 from the Risk-Metrics Directors Legacy dataset (for the years 2002 to2006)6 and ISS (RiskMetrics) Director Data (for the

TABLE 1Summary Statistics Describing S&P 1500 Dataset

Proportion of all Directors (%)

Male 86Female 14Caucasian 91Asian 3.0Black 3.7Hispanic 1.7Other Ethnicity 0.811 Board Seat 842 Board Seats 133 Board Seats 2.84 Board Seats 0.375 Board Seats 0.07

5 Online supplement can be accessed at: https://osf.io/562yg/?view_only51d8c31b6e5a94b0aa40ee90ac95f3f5a.

6 Data captured prior to 2002 in the RiskMetrics Di-rectors Legacy dataset appear to have substantial variationin quality and reliability. For example, although the data-set is meant to include information about S&P 1500 com-panies, and there are roughly 1500 companies in the S&P1500, the 2001 dataset included information about 1,797companies supposedly in the S&P 1500, suggesting that itwas unreliable. This is why we began our analyses withdata from 2002. ISS Director Data are only available goingback to 2007.


https://osf.io/562yg/?view_only&tnqh_x003D;1d8c31b6e5a94b0aa40ee90ac95f3f5a



years 2007 to 2012) on August 22, 2016. For each yearfrom2002 to2012,we repeatedour simulation strategyto calculate how many boards would be expected toincludeexactlyoneorexactly two femaledirectors.Wethen compared these simulation-based expectations tothe number of boards we actually observed with ex-actly one or exactly two female directors.

As illustrated in Figure 2, we found a statisticallysignificant overrepresentation of boards with ex-actly one woman when the descriptive social normwas below one woman per board, and we found astatistically significant overrepresentation of boardswith exactly twowomenwhen the descriptive socialnorm rose above one woman per board. In 2002and 2003, the descriptive social norm for genderdiversity—or the average number of women perboard—was less than one woman, and we see sta-tistically significant tokenism in these two years,but we do not find statistically significant two-kenism in these years. From 2005 to 2013, the de-scriptive social norm for gender diversity exceededone woman, and we see statistically significanttwokenism in these years, but we do not find sta-tistically significant tokenism in these years. In2004, the first year that the descriptive social normfor gender diversity exceeded one woman in theS&P 1500, we still observe statistically significanttokenismanddonot yet find statistically significanttwokenism.

When we ran an ordinary least squares (OLS) re-gression with robust standard errors clustered atthe firm level to predict the extent of tokenism(or the overrepresentation of boards including onewoman) or twokenism (or the overrepresentationof boards including two women) in each year as afunction of whether the descriptive social normfor gender diversity exceeded one woman in thatyear, we found that the descriptive social normexceeding one woman was a significant negativepredictor of tokenism (b 5 20.11; p , 0.001) and a

significant positive predictor of twokenism (b 50.12; p 5 0.002). This provides further support forHypothesis 1a and our theorizing that descriptivesocial norms help determine salient thresholds fordiversity.

Are more visible companies more likely to ex-hibit twokenism? To test Hypothesis 3 in this con-text, we examined whether companies that receivemore media attention were more likely to includeexactly two women on their boards. We used mediaattention as a proxy for visibility to align with pastresearch on organizational visibility (Brammer &Millington, 2006; Chiu & Sharfman, 2011; King,2008; McDonnell & King, 2013). We searchedLexisNexis for all media mentions (including news-papers, Web-based publications, magazines, etc.) ofeach of the companies in the S&P1500 in 2012 (meanmedia mentions of a company5 307; SD5 441). Wegathered 2012 data on media attention so we couldexamine whether past media attention predictedfuture (2013) twokenism.We then analyzedwhethermedia attention in 2012 predicted whether compa-nies would include exactly two women on theirboards in 2013.

We ordered the companies in our dataset by thenumber of media mentions each company receivedin 2012 and created deciles (i.e., 10 bins of 144companies each) based on this ordering. Thus,the first decile contained the companies most fre-quently mentioned in the media in 2012, whilethe last decile contained the companies least fre-quently mentioned in the media in 2012. Aftersegmenting the companies in our dataset by theamount of media attention they were subjected toin 2012, we repeated our basic simulation strategybut limited each simulation to include only thecompanies in a given decile. This strategy allowedus to determine how many companies we wouldexpect to see with exactly two women on theirboards in 2013 in each of the deciles. We ran 1,000

TABLE 2Correlation Matrix for S&P 1500 Board Data in 2013 (n 5 1,441)

1 2 3 4 5 6

1. Size of Board 1.002. Number of Female Directors 0.51*** 1.003. Number of Racial Minority Directors 0.36*** 0.30*** 1.004. Logarithm of Market Capitalization 0.44*** 0.36*** 0.31*** 1.005. Logarithm of Media Mentions 0.43*** 0.38*** 0.32*** 0.59*** 1.006. Member of S&P 500 0.43*** 0.36*** 0.29*** 0.71*** 0.59*** 1.00

***p , 0.001


FIGURE 1Comparison of Actual Distribution of Women on (A) S&P 1500 Boards and (B) S&P 500 Boards with Simulated

Expected Distribution of Women

50

40

30

20

10

0

50

40

30

20

10

0

0 1 2 3 4 5 6 7

0 1 2 3 4 5 6 7

Number of Women on Board

Actual Expected

Panel A—S&P 1500

Actual Expected

Panel B—S&P 500

Number of Women on Board

Descriptive Social Norm of1.36 Women per Board in

the S&P 1500 in 2013

Descriptive Social Norm of1.89 Women per Board in

the S&P 500 in 2013

Per

cen

tage

of

Boa

rds

in S

&P

150

0P

erce

nta

ge o

f B

oard

s in

S&

P 5

00


simulations for each decile, generating a new ex-pected number of companies with exactly two fe-male directors each time. Thus, for each decile, wegenerated an expected number of companies withexactly two women on their boards based on oursimulations, and we were able to compare this ex-pectation with the actual number of companiesincluding exactly twowomen on their boards in our2013 board data.

The results of our simulations for the differentmedia attention deciles are depicted in Figure 3.To test the hypothesis that the likelihood of havingexactly twowomen on a company’s board increasesfor more visible companies, we ran an OLS re-gression with robust standard errors. We used thelogarithm of the average number of mediamentionsin a given decile to predict the absolute differencebetween the observed and expected number ofcompanies with exactly twowomen on their boardsin each decile. The logarithm of media mentions ofthe decilewas a significant predictor of the absolutedifference between observed and expected boardswith exactly two female directors (b 5 6.12, p 50.014). The positive coefficient of log media men-tions indicates that deciles containing more visible

companies were more likely to display twokenism,supporting Hypothesis 3.7

Robustness checks. To further validate our simu-lation strategy and ensure our results were not an ar-tifact of the way we constructed an expecteddistribution of the number of boards including varyingnumbers of female directors, we conducted a series ofplacebo simulations (Gino & Pierce, 2010). Specifi-cally, in these placebo simulations we produced ex-pected distributions of the number of boards thatwould include varying numbers of directors with an-other characteristic (i.e., not gender) that should notshowgoal-related clustering effects because of a lack ofscrutiny on that characteristic (e.g., board members

FIGURE 2How Tokenism and Twokenism Shifted as Social Norms Changed from 2002 to 2013

Per

cen

tage

Ove

rrep

rese

nta

tion

of

S&

P 1

500

Boa

rds

wit

hE

xact

ly O

ne

or T

wo

Wom

en*

25

20

15

–15

10

5

0

–5

–10

The Average Number of Women

Tokenism (Exactly One Woman) Twokenism (Exactly Two Women)

2002 2003 2004 2005 2006 2007

Year

2008 2009 2010 2011 2012 2013

on S&P 1500 Boards FirstExceeded One Woman in 2004

7 See the Online Supplement for additional specifica-tions of this regression to test the robustness of this findingand for a table reporting detailed regression results. Weused as predictors either the logarithm of the averagenumber of mediamentions or the decile rank, andwe usedas outcomes either the absolute overrepresentation ofboards with exactly two women or the percent over-representation of boards with exactly two women. Allyielded findings that were statistically significant andmeaningfully unchanged.


whose ages ended with an arbitrary number). Wefound no significant differences between the expectednumbers of boards and the actual numbers of boards inany of our placebo simulations, suggesting that thelarge deviations we see in our simulations studyinggender were not an artifact of the way we constructedour baseline expectations or null distributions (see theOnline Supplement for complete details about ourplacebo simulations).

In addition to conducting placebo simulations toensure the robustness of our simulation methodol-ogy, we conducted numerous additional robustnesschecks to ensure our results were not driven byoutliers or by a small subset of boards by repeatingour baseline simulations with different cuts of ourdata. First, we checked that our findings were robustto board size. To do this, we used our standard sim-ulation strategy but limited the data to boards of size6 or fewer, 7, 8, 9, 10, 11, 12, and 13 or more. Theunderrepresentation of companies with no womenon their boards and the overrepresentation of com-panies with exactly two women on their boards isrobust across all board sizes tested (see Table 3),although the clustering at the social norm of two islargely driven by companies with larger boards, andfuture research exploring the reasons for this couldyield interesting insights.

Our results are also robust across industries, andtheyhold regardless of the genderof a company’sCEO

(see Online Supplement). We also examinedwhetherthe length of time the companyhasbeenpublic affectsthe likelihood that the company has exactly twowomen on its board. In general, our results appear tobe robust regardless of when a company went public(see Online Supplement). Finally, when we examinehow our results relate to themarket capitalization of acompany, we find that twokenism is more prevalentamong companies with higher market capitalization(see Online Supplement), which are also the mostfrequently mentioned by the media (the correlationbetween the logarithm of a company’s market capi-talization and the logarithm of its number of mediamentions in 20135 0.59; p, 0.001).

STUDY 1B: THRESHOLD EFFECTS IN BOARDMEMBER SELECTION AT THE SOCIAL NORM

In Study 1B, we analyzed the gender of new boardmembers added to company boards over time forevidence consistent with our theories. We predictedthat boards would be discontinuously less likely toadd additional women once they had met the rele-vant descriptive social norm for gender diversity(Hypothesis 1b). Given that the descriptive socialnorm for gender diversity in the S&P 1500 first sur-passed one woman in 2004, we examined all boardmember additions from 2004 to 2013 to test whetherboards during this time periodwere discontinuously

FIGURE 3Firm Visibility Moderates the Extent of Twokenism

50

40

30

20

10

01 2

Decile of Lexis Nexis Mentions in 2012 (1 = Highest 10%; 10 = Lowest 10%)

3 4 5 6 7 8 9 10

Actual ExpectedP

erce

nta

ge o

f B

oard

s w

ith

Exa

ctly

Tw

o W

omen

in

201

3


less likely to add additional female directors oncethey already included two women on their boards.

Method

Data. For these analyses, we used a subset of thedata described in Study 1a. Specifically, we used ISSDirector Data describing board compositions from2007 to 2013 and the RiskMetrics Directors Legacydataset describing board compositions from 2004 to2006 to examine the 9,989 board member additionsin the S&P 1500 from 2004 to 2013.

Analysis strategy. Using data on all board mem-ber additions from 2004 to 2013, we estimated anOLS regressionwith robust standard errors to predictwhether each newly added board member was fe-male.8 We included as predictors both the numberof women currently on a board (to control for thepossibility that boards have either increasing or de-creasingmarginal value for female directors), as well

as an indicator for whether the board included atleast two women (our primary predictor of a dis-continuity in a groups’ desire to add more womenafter exceeding the social norm for gender diversity),and we clustered standard errors by firm. We reportthese regressions with and without fixed effects forboard size, fixed effects for industry, fixed effects forstock market index, and a continuous control for acompany’s market capitalization.

Results

Summary statistics. Of the 9,989 board additionsfrom 2004 to 2013, 16.5% (1,649) were additions offemale directors. The 9,989 board member additionsfrom 2004 to 2013 represent 8,328 distinct directors(i.e., some directors were added to multiple boardsduring this time period), and 16.2% (1,347) of the dis-tinct directors were female. On average, boards in thisdataset added 5.25 directors during this nine-year span.

Do boards add fewer women once they havereached the descriptive social norm? As shown inTable 4,Model 1, for the S&P 1500, the coefficient onour primary predictor of whether a board added afemale director—an indicator for whether the boardalready included at least two women—was negativeand significant in our primary regression specifica-tion (b 5 20.039; p 5 0.017). As shown in Table 4,

TABLE 3Comparison of Actual and Expected Number of Female Directors Across S&P 1500 Boards of Different Sizes

Size of Board n

Excess Percentage ofBoards Observed with0 Female Directors

Excess Percentage ofBoards Observed with1 Female Director



6 or fewer 124 2.74 25.79 216.96 37.36(2.89) (11.23) (28.15) (109.43)

7 199 22.60 4.38 10.06 266.06(3.34) (8.98) (16.79) (54.18)

8 241 215.98** 23.15** 28.42 224.81(5.57) (8.04) (11.18) (24.50)

9 283 226.32** 14.16* 20.60* 234.31*(8.05) (7.22) (9.09) (14.04)

10 235 238.00*** 18.88* 16.51 216.12(10.85) (8.58) (10.14) (13.57)

11 198 250.56*** 6.67 28.69** 7.47(14.71) (9.95) (10.95) (13.42)

12 100 276.36** 222.97 64.40*** 26.40(28.08) (15.09) (15.38) (17.34)

13 or more 134 258.37* 220.70 49.34*** 22.85(24.01) (13.63) (13.10) (15.55)

Notes:This table reports the difference between the actual percent of boardswith a given number of female directors and the simulated expectedpercent of boards with that number of female directors conditional on the size of the board. Standard deviations are reported in parentheses.

*p , 0.05**p , 0.01

***p , 0.001

8 We rely on a linear model because it yields more in-terpretable coefficients than a logit specification, and thismethod also allows us to correct for heteroskedasticity inthe standard errors (Angrist & Pischke, 2008; see Brands &Fernandez-Mateo, 2017 for a similar procedure). However,logistic regressions yield similar results and are reported inthe Online Supplement.


Model 3, for the S&P 500 (roughly the 500 most vis-ible and valuable companies in the S&P 1500), thecoefficient on the indicator variable was negativeand even more highly significant (b 5 20.092; p ,0.001). This suggests that companies are less likely toadd additional women to their boards once theirboards havemet the social norm for gender diversityby including two women, providing support forHypothesis 1b. The larger effect size in the (highlyvisible) S&P 500 also provides some suggestivesupport for Hypothesis 3. Adding in fixed effectsfor board size, fixed effects for industry, fixed effectsfor stock market index, and a continuous control formarket capitalization (see Table 4, Models 2 and 4),we still find that our predictor of a discontinuity issignificant in the S&P 1500 (b 5 20.034; p 5 0.037)and in the S&P 500 (b 5 20.090; p , 0.001).

Do more visible companies show largerdiscontinuities at the descriptive social norm?To test Hypothesis 3 in Study 1B, we examinedwhether there was an interaction between media at-tention and our primary predictor of whether a boardadded a femaledirector—an indicator forwhether theboard already included at least twowomen.We againsearchedLexisNexis for allmediamentions of each ofthe companies in the S&P 1500, and we gathered ad-ditional data to look at media mentions for each yearstarting in 2004 to see if media attention in year t – 1predicted whether a newly added board member inyear t was female. For our analyses, we used the cen-tered logarithm ofmediamentions rather than the rawnumber of mediamentions because the distribution of

media mentions is highly skewed (skewness 5 2.57;skewness test for normality p , 0.001; kurtosis 510.37; kurtosis test for normality p, 0.001).

Our results are depicted in Table 5. As predicted,we find a significant negative interaction betweenthe centered logarithm of the number of mediamentions in year t–1andhaving twoormorewomenon a board in predicting whether a newly addedboard member in year twas female (b520.021; p50.042; Model 1). Adding in fixed effects for boardsize, fixed effects for industry, fixed effects for stockmarket index, and a continuous control for marketcapitalization, we still find a significant negativeinteraction between the centered logarithm ofmediamentions and having two or more women (b 520.021; p 5 0.041; Model 2). These results suggestthat more visible companies show larger disconti-nuities in board member additions at the descriptivesocial norm of two women per board.

STUDY 1C: ONLINE EXPERIMENT REPLICATINGTHRESHOLD EFFECTS

In Study 1C, we sought to replicate our findingsregarding threshold effects from Study 1B in anonline experiment that allowed us to randomly as-sign the number of women in a group and control forthe availability of qualified candidates. Specifically, weinvestigatedwhether individuals in a controlled settingare less likely to addwomen to a corporate boardwhenthe board has met or surpassed the social norm forgender diversity by including two or more women.

TABLE 4Boards Less Likely to Add Additional Women Once They Include at Least Two Women

Board Added Woman5 1

Model 1 Model 2 Model 3 Model 4

S&P 1500 S&P 1500 S&P 500 S&P 500

Number of Women on Board 20.0033 20.039*** 20.0056 20.035*(0.0079) (0.0090) (0.012) (0.015)

Indicator for Two or More Women on Board 20.039* 20.034* 20.092*** 20.090***(0.016) (0.016) (0.023) (0.024)

Controls Present? No Yes No YesObservations 9,989 9,936 4,131 4,117R2 0.0032 0.030 0.017 0.045

Notes: This table shows a series of OLS regressions predicting whether boards addwomen conditional on the number of women already onthe board andwhether the board hadmet the descriptive social norm for gender diversity (i.e., already had at least twowomen) in the S&P 1500(Models 1 and 2) and the S&P 500 (a subset of the S&P 1500; Models 3 and 4). Robust standard errors are in parentheses. When controls arepresent, regressions include fixed effects for board size, fixed effects for industry, fixed effects for stockmarket index, and a continuous controlfor market capitalization.

*p , 0.05***p , 0.001


Method

Participants.A total of 479 U.S. participants wererecruited through Amazon’s Mechanical Turk toparticipate in a short online research study (55%male; 77%Caucasian). These participants were paid$0.25 for completing a survey they were told wouldtake approximately five minutes of their time. Sam-ple size was determined a priori, data analysis wasconducted only once all data were collected, and wedid not exclude any data.

Procedures. In a pilot study (see the Online Sup-plement for details), we first established that ourstudy population was indeed aware that two is theaverage number of women on U.S. corporate boards(i.e., two women is the descriptive social norm forgender diversity).

After establishing an awareness of descriptivesocial norms in the pilot study, we ran our primarystudy. In this study, participants were asked toimagine they had been tasked with helping a com-pany select a newmember for its board of directors.

They were then exposed to a list of 10 names andtold the current board consisted of the individualson that list. Participants were randomly assigned toone of four experimental conditions where zero,one, two, or three of the names of board memberswere female.

Study participants were next presentedwith threehypothetical candidates for an opening on the boardin question and asked to choose one to add to theboard. The candidates were all described as quali-fied, but one was a CEO, one was a current boardmember at another company, and one was a con-sultant with expertise in the industry. We randomlyvaried which candidate had a female name (JillDavis) and which candidates had male names (Mat-thew Anderson and Todd Miller), and we randomlyvaried which name was associated with each quali-fication.9 Following Castilla and Benard (2010), wepresented three candidates for the available boardseat rather than one male and one female to reducesuspicion that our study was about gender. Our de-pendent variable of interest was what fraction ofparticipants in each condition would choose a fe-male candidate.

Finally, participants completed demographic ques-tions and amanipulation check question, which askedthem to recall how many men and howmany womenwere present on the corporate board they had seen atthe beginning of the survey. Study materials and acorrelation matrix of all variables collected in thisstudy are available in the Online Supplement.

Results

First, our manipulation check confirmed our ma-nipulation was successful: participants recalled sig-nificantly more women on boards that includedthree women than two (p, 0.001), two women thanone (p , 0.001), and one woman than zero (p 50.015).

Second, a x2 test of independence showed a mar-ginally significant relationship between the numberof women on the board and whether the participantchose the female candidate (x2(3, n 5 479) 5 7.51,

TABLE 5More Visible Companies Show Larger Discontinuities at

the Descriptive Social Norm

Board AddedWoman5 1

Model 1 Model 2

Number of Women on Board 20.017* 20.043***(0.0083) (0.0090)

Indicator for TwoorMoreWomenonBoard

20.27 20.023(0.017) (0.017)

Centered Logarithm of MediaMentions

0.026*** 0.017**(0.0042) (0.0049)

Number of Women on Board3Centered Logarithm of MediaMentions

20.0018 20.00018(0.0049) (0.0053)

Indicator for TwoorMoreWomenonBoard3 Centered Logarithm ofMedia Mentions

20.021* 20.021*(0.010) (0.010)

Controls Present? No YesObservations 9,781 9,743R2 0.012 0.033

Notes: This table shows two OLS regressions predictingwhether boards add women conditional on the number of womenalready on the board and whether the board had met the de-scriptive social norm for gender diversity (i.e., already had at leasttwowomen), interactedwith the centered logarithmof thenumberofmediamentions a company receives. Robust standard errors arein parentheses. When controls are present, regressions includefixed effects for board size, fixed effects for industry, fixed effectsfor stock market index, and a continuous control for marketcapitalization.

*p , 0.05***p , 0.001

9 Participants were most likely to choose the candidatewhowas a CEO (p, 0.001), regardless of gender. However,because we randomly assigned qualifications to the can-didates, we do not need to control for candidate qualifi-cation in order for our tests to provide unbiased estimatesof the causal effects of our manipulations. In addition, wedid not find any significant interactions between genderand candidate qualifications.


p 5 0.057). Consistent with Hypothesis 1b and rep-licating our results from Study 1B, participants weresignificantly less likely to choose the female candi-date and increase the gender diversity of the boardonce the board included at least two women. Par-ticipants shown a corporate board with exactly twofemale members were significantly less likely tochoose the female candidate for the open seat (M 536.0%, SD5 0.48) than were participants who wereshown a corporate board with one female member(M 5 50.4%, SD 5 0.50; z 5 2.26, p 5 0.024; seeFigure 4).10We then ran anOLSwith robust standarderrors to predict the likelihood a participant chosethe female candidate, replicating our empirical an-alyses of boardmember additions fromS&P1500 andS&P 500 data from Study 1B. We again included thenumber ofwomen currently on the board as a controlvariable, in addition to an indicator variable forwhether the board included at least two women as apredictor of a discontinuity. The coefficient on theindicator variable was negative and marginally sig-nificant (b 5 20.19, p 5 0.062; see Table 6), sug-gesting that participants in our experimentwere alsodiscontinuously less likely to increase the genderdiversity of the board once the board had at leasttwo women, and providing additional support forHypothesis 1b.

DISCUSSION OF STUDY 1

Study 1A shows that U.S. corporate boards aredisproportionately likely to include exactly thenumber of women needed to minimally exceed thedescriptive social norm for female representationin peer groups. This evidence is consistent withHypothesis 1a, which proposes that the composi-tion of groups facing scrutiny on a diversity di-mension will cluster around the descriptive socialnorm for that type of diversity. Further, historicalanalyses show that descriptive social norms pre-dicted the shift from tokenism (an overabundanceof boards with exactly one female director) to two-kenism (an overabundance of boards with exactlytwo female directors), providing additional supportforHypothesis 1a—i.e., that the clusteringwedetectis driven by the descriptive social norm for genderdiversity.

Study 1B provides support for Hypothesis 1b,which states that groups facing scrutiny on a di-versity dimension will be less likely to addmembersof the relevant underrepresented group once theyhave reached the descriptive social norm for di-versity. We find that U.S. corporate boards are dis-continuously less likely to add additional womenonce they have reached the descriptive social normfor diversity by including two female directors. InStudy 1C, we replicate this finding in a stylized ex-periment where we randomly assign the number ofwomen to ahypothetical corporate board andcontrolfor the availability of qualified candidates. WhileStudy 1C lacks the realism of Studies 1A and 1B, itconfirms our hypothesis in an environment wherewe can randomly assign board composition, pro-viding convergent evidence that there exists a causalrelationship between board composition and thegender of new board members.

Consistent with Hypothesis 2, which predicts thatscrutiny is a necessary condition for social normsto influence diversity, we do not see evidence ofclustering at the social norm when we look at boardmembers’ race or ethnicity in supplemental ana-lyses.11 There is far less scrutiny of corporate boards’racial diversity compared with the scrutiny boardsface regarding gender diversity (e.g., only 18% ofnews articles about board diversity in 2013 dis-cussed racial diversity while 97% discussed genderdiversity, and no laws have been passed establishingracial quotas on corporate boards in any country), socorporate boards may have fewer impression man-agement motives regarding the recruitment of racialor ethnic minorities compared to women.

Finally, consistent with Hypothesis 3, we findevidence that companies that are more visible (asmeasured by media coverage in the previous year)are more likely to include exactly two women ontheir boards, consistent with our theory that theclustering we detect at the social norm is driven inpart by impression management concerns. In Study1B, we also find that companies that aremore visibleshow larger discontinuities at the descriptive socialnorm of two women per board when adding addi-tional female board members.

Past research has suggested that these findingsare worrisome from a policy perspective. Researchon the benefits of gender diversity on corporateboards has suggested that at least three female10 We found amain effect of participant gender such that

female participants were significantly more likely toselect the female candidate (p 5 0.019), but we found nosignificant interaction between participant gender anddecisions.

11 A more detailed discussion of simulation analysesregarding director race and ethnicity can be found in theOnline Supplement.


directors are needed before boards experience tangi-ble benefits from gender diversity (Konrad, Kramer, &Erkut, 2008; Torchia, Calabro, & Huse, 2011). Bystopping at twowomen, boardsmaybemissingout onkey benefits that can ensue from greater gender di-versity. Further, our results suggest that the push forgender parity on boardsmay not generate results for along time. In Study 1A, we depict the evolution ofdescriptive social norms regarding gender diversityon corporate boards over a 12-year span, and theseresults suggest that descriptive social norms changequite slowly over time.

In spite of the evidence provided by our empiricalanalyses of archival board composition data sup-porting our theorizing and hypotheses, Studies 1A

and 1B are ultimately only correlational studies, andthushave limitations.Wecannot completely rule outconcerns about reverse causality or other confounds,such as firm performance. In addition, because we donotobserveboardmemberselectiondecisionsdirectly,we can only explore the mechanisms responsible forthe effects we have documented indirectly. Thereare many factors at play that affect who is added tocorporate boards (e.g., legal constraints can preventpeople from serving on multiple boards; bias andstereotyping may affect board member selection), andwe focus only on the roles played by descriptive socialnorms, scrutiny, and visibility. We also unfortunatelycannot disentangle the specific motives of individualcompanies.

In order to providemore confidence in our results,in Study 1C we replicated threshold effects at thedescriptive social norm in an experimental settingwhere we could randomize the number of womenin a group. This gives us greater confidence that theresults found in Study 1B are not driven by endo-geneity or the fact that there are not enough quali-fied women for director positions. However, weacknowledge that Study 1C is a stylized experimentthat does not accurately represent corporate boarddecision-making processes. First, our experiment isconducted at the level of the individual, whileboards are groups. Second, board members havemuch more experience and many more constraintsthey must attend to, while we use a relatively un-informed sample and intentionally strip awaymany

FIGURE 4Participants in Study 1C Less Likely to Increase Gender Diversity of Boards Once Boards Include Two

Women (and Thus Exceed the Social Norm)

42

50

36 35

Zero Women

Experimentally Manipulated Number of Women on Board

Per

cen

tage

Ch

oosi

ng

Fem

ale

Can

did

ate

One Woman Two Women Three Women

60

50

40

30

20

10

0

TABLE 6Regression Predicting the Selection of the Female

Candidate to Serve on a Corporate Board in Study 1C

B

Number of Women on Original Board 0.040 (0.045)Original Board Has Two or More Women 20.19† (0.10)Observations 479R2 0.013

Notes: This table shows the results of an OLS regression pre-dicting whether participants added a woman to a board condi-tional on the number of women already on the board andwhetherthe board hadmet the descriptive social norm for gender diversity(i.e., already included at least twowomen). Robust standard errorsare in parentheses.

† p, 0.1


of the complications of the board member selectionprocess for simplicity.

In spite of these limitations, these studies collec-tively provide empirical evidence that group com-position and group diversity decisions can beaffected by threshold effects at the descriptive socialnorm. In our subsequent studies, we provide addi-tional experimental evidence that directly tests ourtheoretical model to examine the influences of de-scriptive social norms, scrutiny, and visibility ongroup diversity decisions.

STUDY 2: EXPERIMENTALLY MANIPULATINGSOCIAL NORMS AND SCRUTINY

In Study 2, we sought to test our theoreticalmodel more directly by manipulating—rather thanmeasuring—the descriptive social norms and scrutinyassociated with the inclusion of females in a group. Inaddition, we sought to explore these phenomena in anew setting to establish their generalizability to groupsbesides corporate boards.

STUDY 2A: GROUP DIVERSITY, SOCIALNORMS, AND SCRUTINY

In Study 2A, we tested whether manipulatingdescriptive social norms and scrutiny affects deci-sions about whether to add a female candidate to amajority-male group with a sample of participantswith work experience. Specifically, we investigatedwhether, as predicted in Hypothesis 2, individualsstrive to meet descriptive social norms for diversitywhen under threat of possible scrutiny, but not incases where scrutiny is absent.

Method

Participants. A total of 556 master of businessadministration (MBA) students completed thisstudy. This represented the entire incoming class at aU.S. business school. Fifty-seven percent of the par-ticipants were male, 25% had previous managerialexperiencebefore starting theirMBA,andparticipants’average age was 27.7 years. Sample size was de-termined a priori, data analysis was conducted onlyonce all data were collected, we did not exclude anydata, and we report all measures and manipulations.

Procedures. Participants were asked to imaginetheir companyhad given them the task of assemblinga seven-person panel for submission to an industryconference. Theywere told six of the seven panelistshad already been determined, and they were

responsible for selecting the final panelist. All par-ticipants saw an image of two women and four men,representing the six predetermined panelists. Par-ticipants were randomly assigned to one of four ex-perimental conditions (surpassed social norm orunmet social norm 3 unscrutinized or scrutinized),described below.

Participants saw images of five seven-personpanels,representing other panel submissions to the confer-ence. Participants randomly assigned to the surpassedsocial norm condition saw that four of these otherpanel submissions had one woman each, while onepanel submissionhadnowomenon it (i.e., the averagenumber of women on other panels was 0.8); partici-pants randomly assigned to the unmet social normcondition saw that four of thesepanel submissionshadthree women each, while one panel submission hadtwowomenonit (i.e., theaveragenumberofwomenonother panels was 2.8). Therefore, in the surpassed so-cial norm condition, the participant’s current panel(which included two women) already exceeded thedescriptive social norm for gender diversity (0.8women); in the unmet social norm condition, the par-ticipant’s current panel (which included two women)was below the descriptive social norm for gender di-versity (2.8 women).

Participants were told panels were generally ac-cepted based on speaker quality and years of in-dustry experience of the panelists. Participantsrandomly assigned to the unscrutinized conditionwere told the review process was “blind:” the namesand photos of the panelists would not be submittedfor evaluation (i.e., it would be impossible for thepanels to be scrutinized with regards to gendercomposition). Participants randomly assigned to thescrutinized condition saw no such statement. Pastresearch has suggested that impressionmanagementconcerns often arise when people simply knowthey are being evaluated (Leary & Kowalski, 1990),suggesting that when the evaluation process is notblind, scrutiny can be expected to affect decisions.12

12 In a separate pilot study, we asked participants to ratethe extent to which they agreed or disagreed, on a seven-point scale, with the statements, “My decision is underscrutiny with regards to the gender diversity of the panel”and “The reviewer will pay attention to the gender di-versity of the panel when deciding which panels to ac-cept.” Participants in the scrutinized condition reportedsignificantly higher scrutiny on gender diversity than par-ticipants in the unscrutinized condition (Mscrutinized 5 3.67,SDscrutinized 5 1.84; Munscrutinized 5 2.66, SDunscrutinized 51.92, t(150)5 3.34, p5 0.001).


Participants were then shown two potentialcandidates—Candidate A and Candidate B—for thefinal panelist. One image depicted a female candi-date who had 10 years of industry experience and aspeaker rating of 4.6; the other image depicted amalecandidate who had 12 years of industry experienceand a speaker rating of 4.8. Which candidate waspresented first as Candidate A (versus second asCandidate B) was randomized. Participants thenrated their preference for the two candidates on ascale from 1 to 7, where 1 was labeled as “Stronglyprefer Candidate A” and 7 was labeled as “Stronglyprefer Candidate B.” Study materials and a correla-tion matrix of all variables collected in this studyare available in the Online Supplement.

Results

Consistent with Hypothesis 1b, and as illustratedin Figure 5A, participants in the scrutinized con-dition had a significantly stronger preference for thefemale candidate in the unmet social norm condi-tion than in the surpassed social norm condition(t(277) 5 2.24; p 5 0.026). In other words, partici-pants whose diversity decisions could be scruti-nized found it much more desirable to add a femalecandidate to a group when the group had not yetmet the social norm for gender diversity, com-pared to when the group had surpassed the socialnorm.However, consistentwithHypothesis 2, therewere no differences in the preferences expressedfor the female candidate between the unmet socialnorm and the surpassed social norm conditionswhen diversity decisions were not under scrutiny(t(275) 5 0.22; p 5 0.83).

Next, we checked whether there was a significantinteraction between surpassed social norms and thepresence of scrutiny. We estimated an OLS re-gression to predict the preference for the femalecandidate with indicators for our scrutinized con-dition, our unmet social norm condition, and theinteraction between these two conditions (seeTable 7, Model 1). The interaction term was positivebut did not reach standard levels of statistical sig-nificance (p 5 0.14).13 To strengthen our statisticalpower to detect an interaction, we conducted afollow-up study with incentivized decisions (notethat we could not increase the sample size in this

study because it already included every incomingMBA student at our selected university, and wewere not able to incentivize the decisions of MBAstudents).

STUDY 2B: REPLICATING STUDY 2AWITH INCENTIVES

InStudy2B,we sought to replicate our results fromStudy 2A but with real monetary stakes that wouldincrease our statistical power to detect an interactionbetween the presence of scrutiny and a surpassedsocial norm for diversity. Again, we experimentallymanipulated scrutiny and descriptive social normsto test for a causal relationship between these vari-ables and the demographic characteristics of a newlyselected group member.

Method

Participants. Two hundred U.S. participants(51.5% male) were recruited through Amazon’sMechanical Turk and paid $0.15 to participate ina short online research study. Sample size wasdetermined a priori, data analysis was conductedonly once all data were collected, we did not ex-clude any data, and we report all measures andmanipulations.

Procedures. As in Study 2A, participants wereasked to imagine their company had given them thetask of assembling a seven-person panel for sub-mission to an industry conference, that six of theseven panelists had already been determined (twowomen and four men), and that the participantswere responsible for selecting the final panelist.Again, participants were randomly assigned to oneof four experimental conditions.

In this study, we simplified the way the de-scriptive social norm was manipulated. Partici-pants randomly assigned to the surpassed socialnorm condition were told competitive intelligencesuggested the other panel submissions would have1.25 women on average; participants randomlyassigned to the unmet social norm condition weretold the other panel submissions would have 2.75women on average.

Participants were then told a reviewer wouldevaluate all panel submissions and choose to “ac-cept” 75% of them. If their panel submission wasaccepted, participants would receive a bonus pay-ment. All participants were initially allocated a$0.25 bonus, but participants had to “pay” to selectthe last panelist, and this cost was deducted from

13 We found a significant main effect of gender such thatwomen had significantly higher preferences for the femalecandidate compared to men (p 5 0.022), but there was nosignificant interaction.


their promisedbonus. Participants randomly assignedto the unscrutinized condition were told the reviewprocess was “blind:” the names and photos of thepanelists would not be submitted for evaluation (i.e.,it would be impossible for the panels to be scruti-nized with regards to gender composition). Partici-pants randomly assigned to the scrutinized conditionsaw no such statement.

Participants were then offered the choice amongthree candidates for their final panelist. One imagedepicted a female candidate who had 10 years ofindustry experience, a speaker rating of 4.6, andcost $0.15 to select. The other images depictedmale candidateswhohad similar qualifications (11or 12 years of industry experience; speaker ratings

of 4.5 or 4.8) and cost $0.10 and $0.11 to select.Our outcome of interest was what fraction of par-ticipants in each condition selected the femalecandidate. We made the female candidate slightlymore expensive to reflect research suggesting thatwomen are more expensive to recruit or hire incontextswhere diversity is lacking (e.g., on corporateboards and other contexts where less than 50% ofthe workforce is female [see Leslie, Manchester, &Dahm, 2017]). Finally, participants reported theirgender identity and whether they had ever attendedor organized a conference in the past 10 years. Studymaterials and a correlation matrix of all variablescollected in this study are available in the OnlineSupplement.

FIGURE 5AParticipants’ Preferences for Women Are Influenced by Social Norms and Scrutiny in Study 2A

5

4.5

4

3.5

3

2.5

2

1.5

1

Pre

fere

nce

for

Fem

ale

Can

did

ate

Unscrutinized

Surpassed SocialNorm

Unmet Social Norm

Scrutinized

FIGURE 5BInteraction between Social Norms and Scrutiny in Study 2B

60

50

40

30

20

10

0Per

cen

tage

Ch

oosi

ng

Fem

ale

Can

did

ate


Unscrutinized Scrutinized

Unmet Social Norm


Results

Consistent with Hypothesis 1b, participants inthe scrutinized condition were significantly morelikely to select the female candidate in the unmetsocial norm condition than in the surpassed socialnorm condition (z5 2.94; p5 0.0033; see Figure 5B).Consistent with Hypothesis 2, there were no suchdifferences in the likelihood of selecting the femalecandidate in the unmet social norm and the sur-passed social norm conditions when there was noscrutiny (z 5 0.24; p 5 0.81).

To test for an interaction between the presence ofscrutiny and unmet social norms, we estimated anOLS regressionwith robust standard errors to predictthe choice of a female candidate with indicators forour scrutinized condition, our unmet social normcondition, and the interaction between these twoconditions (see Table 7, Model 2). We found thatthe interaction term was positive and statisticallysignificant (b 5 0.27; p 5 0.028). This further sup-ports Hypothesis 2—i.e., that when shaping thecomposition of groups, decision makers will onlyconform to the social norm for diversity when theyare under scrutiny.

DISCUSSION OF STUDY 2

Studies 2A and 2B directly manipulate scrutinyand descriptive social norms to provide direct testsof Hypotheses 1b and 2 and show that decisionmakers responsible for shaping group composition

strive to increase group diversity when the group inquestion has not yetmet the social norm for diversityon a scrutinized dimension (gender in the case ofthese studies). However, motivation to further in-crease diversity is reduced once the social norm hasbeen met, and social norms do not exert this influ-ence when scrutiny is not present.

STUDY 3: THE MODERATING EFFECTOF VISIBILITY

In Study 3, we manipulated descriptive socialnorms and a group’s visibility to investigate whetherthe influence of descriptive social norms on de-cisions about group diversity is moderated by agroup’s visibility, and we also extend our study ofgroup diversity to explore a social category besidesgender.

Method

Participants. A total of 603 U.S. participants(52.9% male; 80.4% Caucasian) were recruitedthroughAmazon’sMechanical Turk to participate ina short online research study in exchange for $0.30.Sample size was determined a priori, data analysiswas conducted only once all data were collected, wedid not exclude any data, andwe report all measuresand manipulations.

Procedures. Participantswere told to imagine theywere the manager of a team of five people and werehiring a sixth teammember. All participants saw animage of one black man and four white men repre-senting their current team. Theywere also told theirhuman resources (HR) department cared about theracial diversity of teams and the HR departmentcould review team compositions and choose topunish teams deemed to have inadequate racialdiversity, creating scrutiny on the dimension of ra-cial diversity in all conditions. Participants werethen randomly assigned to one of four experimentalconditions.

Participants randomly assigned to the surpassedsocial norm condition were told that other teams oftheir size included an average of 0.25 black people.Participants randomly assigned to the unmet socialnorm condition were told that other teams of theirsize included an average of 1.75 black people.

To manipulate visibility, participants were ran-domly assigned to learn either: (1) their team was“not very important” in the company so there wasa low probability that the HR department wouldreview the composition of their team (the low

TABLE 7Regression Predicting Preference for Female Candidates

to Serve on Panels in Studies 2A and 2B

Study 2A Study 2B

Model 1 Model 2

DV: Rating of FemaleCandidate

Chose FemaleCandidate

Scrutinized 0.65** (0.24) 0.031 (0.075)Unmet Social Norm 0.051 (0.24) 0.018 (0.075)Scrutinized 3 Unmet

Social Norm0.51 (0.34) 0.27*(0.12)

Observations 556 200R2 0.056 0.084

Notes: These OLS regressions present the preference for thefemale candidate to serve on a panel in Studies 2A and 2B. Scru-tinized is an indicator for the scrutinized condition. Unmet SocialNorm is an indicator for the unmet social norm condition. Robuststandard errors are in parentheses.

*p , 0.05**p , 0.01


visibility condition), or (2) their team was “veryimportant” in the company so there was a highprobability that the HR department would reviewthe composition of their team (the high visibilitycondition).14

Participants were then offered the choice of twocandidates for their new team member. One imagedepicted a black male candidate, who would comewith a bonus of $0.03 to participants if they chosehim; the other image depicted a white male candi-date, who would come with a bonus of $0.10 toparticipants if they chose him. We incentivizedparticipants to choose the white man in order toovercome social desirability concerns and placesome cost on increasing diversity. Participants weretold they would keep the bonus associated withthe candidate they chose unless the HR departmentreviewed their team and chose to penalize their teamfor a lack of racial diversity.

Finally, participants reported their racial andgender identities. Study materials and a correlationmatrix of all variables collected in this study areavailable in the Online Supplement.

Results and Discussion

Consistent with Hypothesis 1b and all previousstudies, participants were significantly more likelyto select the black candidate in the unmet socialnorm condition than in the surpassed social normcondition (z5 4.28; p, 0.001; see Figure 6). In otherwords, decisionmakers added the black candidate totheir group at a lower rate once their group had sur-passed the descriptive social norm for racial di-versity. In addition, there was a significant maineffect of visibility, such that participants were sig-nificantly more likely to select the black candidatewhen their teamwas highly visible than when it wasnot (z5 9.25; p , 0.001).

To test Hypothesis 3, which states that visibilitymoderates the effect of descriptive social norms, wetested for an interaction between visibility and socialnorms. To do this, we estimated an OLS regression

with robust standard errors to predict the choice ofthe black candidate with indicators for our high vis-ibility condition, our unmet social norm condition,and the interaction between these two conditions (seeTable 8). Consistent with Hypothesis 3, we found theinteraction term between visibility and norms waspositive and statistically significant (b 5 0.15; p 50.043).

Overall, Study 3 conceptually replicates our pre-vious studies, extends our findings to under-represented groups besides women, and shows themoderating effect of visibility on decisions aboutgroup diversity.

GENERAL DISCUSSION

Across four experiments and one field study, weoffer convergent evidence that those who shape thediversity of groups attend to and seek to conform tothe descriptive social norms for diversity set by peergroupswhen under scrutiny. In Study 1,we showedthat U.S. corporate boards are disproportionatelylikely to include exactly two women (the de-scriptive social norm), and they appear to lose mo-tivation to add additional women once they havematched the descriptive social norm by includingtwo female directors. We also found that theseeffects are more pronounced among more visiblecompanies, consistent with our theory that theseeffects are driven in part by scrutiny and impressionmanagement motives. In addition, we did not findany clusteringwhenwe analyzed data on the race orethnicity of board members in our field data, con-sistent with our theory that scrutiny is required toproduce clustering at the descriptive social norm.15

In Studies 2 and 3, we directly manipulated de-scriptive social norms, scrutiny, and visibility toshow that each of these influences group diversitydecisions as our theory predicts in groups besidescorporate boards and when we examine social cat-egories besides gender.

Theoretical and Practical Implications

Our theory and findings help us understandhow decision makers with the power to shapegroup compositions respond to the threat of nega-tive scrutiny surrounding diversity. Individuals

14 In a separate pilot study, we asked participants to ratethe extent to which they agreed or disagreed, on a seven-point scale, with the statements, “My team is visible in thecompany” and “My team receives a lot of attention in thecompany.” Participants in the high visibility condition re-ported significantly higher scores on these items than par-ticipants in the low visibility condition (Mhigh_visibility5 6.39,SDhigh_visibility 5 0.97; Mlow_visibility 5 2.25, SDlow_visibility 51.49; t(147)5 20.04; p, 0.001).

15 As discussed in Study 1A, an analysis of media at-tention to board diversity in 2013 showed that 97%of sucharticles discussed gender diversity, while just 18% evenmentioned racial or ethnic diversity.


responsible for group compositions look to de-scriptive social norms, matching the levels of di-versity found in peer groups at an unusually highrate. This behavior leads to homogeneous levelsof diversity across groups, providing another con-tributing explanation for the persistent under-representation of women and racial minorities inmany organizational contexts. Our work also helpsprovide a fuller understanding of diversity-relatedhiring decisions, illuminating when women andracial minorities will be particularly attractivecandidates for inclusion in groups andwhen groupscan be expected to lessen their efforts to increasediversity.

Our findings suggest newavenues for policymakersseeking to increase diversity. Rather than simplytargeting bias and stereotyping among thosemakinghiring decisions (e.g., through diversity training)or seeking to shape underrepresented candidates’preferences and skill sets (e.g., by training womento negotiate), more interventions may be neededto change the perceived norms around diversity.Groups appear to cluster at the descriptive socialnorm for diversity because it is an adaptive im-pression management strategy: by clustering atthe social norm, they can escape negative scrutinyregarding their diversity levels. However, the factthat groups can escape negative scrutiny once theyreach the descriptive social norm for diversity im-plies that those scrutinizing these groups (e.g.,shareholders, the media) may be too easily satis-fied. Shifting the standards of those who scrutinizediversity, as well as those of the decision makerscapable of shaping group diversity, from focusingon descriptive social norms in peer groups to

instead achieving more ambitious norms (e.g.,matching the levels of diversity in the general pop-ulation) may be a promising new avenue for in-creasing the diversity of highly visible, scrutinizedgroups. If powerful institutions or individuals en-dorse new norms regarding gender and racial repre-sentation, perhaps this could lead to changes in thenorms that influence group composition decisions(Paluck & Shepherd, 2012). For instance, decisions bythe Supreme Court have been shown to change atti-tudes and perceptions of norms in the realm of gayrights (Tankard & Paluck, 2017).

Our work also points to scrutiny as a lever forchange. Scrutiny can come from a variety of sources,but some sources may be more influential thanothers (Oliver, 1991). Applying greater scrutiny togroup diversity should lead groups to increase their

TABLE 8Regression Predicting the Selection of a Black Candidate

for a Team in Study 3

B

High Visibility 0.30*** (0.055)Unmet Social Norm 0.089 (0.053)High Visibility 3 Unmet Social Norm 0.15* (0.043)Observations 603R2 0.18

Notes:ThisOLS regressionpredictswhether participants chosethe black candidate to serve on a team in Study 3.HighVisibility isan indicator for the high visibility condition. Unmet Social Normis an indicator for the unmet social norm condition. Robust stan-dard errors are in parentheses.

*p , 0.05***p , 0.001

FIGURE 6Interaction between Social Norms and Visibility in Study 3


Unmet Social Norm

Per

cen

tage

Ch

oosi

ng

Bla

ck C

and

idat

e

Low Visibility High Visibility

60

70

80

90

50

40

30

20

10

0


diversity. One extreme form of scrutiny when itcomes to diversity is to enforce legal penalties onpublic companies for a failure to diversify. How-ever, even when policymakers have establishedlaws mandating minimum levels of gender di-versity on the corporate boards of public compa-nies, some companies have elected to becomeprivate rather than comply with the laws (Miller,2014). Forced compliance therefore comes with therisk of creating at least some reactance (Dobbin,Schrage, &Kalev, 2015). An alternative tomandateddiversity may be to shower positive attention ongroups that reach high levels of diversity. Treatingdiversity as an ideal may help reshape perceptionsof the relevant norm, leading injunctive norms (ornorms about ideals) to overshadow descriptive so-cial norms.

Limitations and Future Research

One paradox suggested by our theorizing andempirics surrounds changing descriptive norms:U.S. corporate boards shifted from clustering at onewoman to clustering at two women (albeit slowly)over the last 20 years in spite of the fact that ourtheorizing about diversity thresholds would predicta stagnation of board diversity at the one-womanthreshold. A noteworthy fact, however, is that thisshift in clustering followed the passage of Norway’s“Women on Boards” act in 2003. This legislationrequired public and state-owned companies inNorway to include at least 40% women on theirboards, and it may have made the topic of genderdiversity on corporate boards in the United Statesmore salient at that time. This law could have in-creased scrutiny of boards with few women andmade the need for gender diversity more salient,driving the shift to twokenism from tokenism. Fu-ture research exploring how descriptive socialnorms can be shifted in the context of diversitywould be extremely valuable.

Another puzzling question raised by our findingsis whether more diverse groups may actually dis-criminate more than less diverse groups. We cannotevaluate whether any specific group or organizationis actively “managing” diversity for impressionmanagement reasons. However, overall, we do see apattern suggesting that this is the case and that—contrary to the expectation that more diverse groupswill attract more women and racial minority candi-dates (Avery, 2003; Avery & McKay, 2006)—suchgroups are less likely to select women and racialminorities than others after reaching the descriptive

social norm for diversity. It would be valuable forfuture research to examine when and how socialnorms around diversity can hurt rather than helpwomen and minorities.

Although our field and experimental studies pro-vide convergent evidence in support of our theoryand hypotheses, in our experiments we only ex-amine the judgments and decisions of individuals,while group member selection processes are variedand complex and often involve many decisionmakers. Extensive past research has shown thatstudies of individual decisions and insights aboutindividual psychology can further our understandingof group and organizational outcomes (Greve, 2008;Highhouse, Brooks, & Gregarus, 2009; Simon &Houghton, 2003; Staw, 1991). However, there areunquestionably limitations in our approach.

Weonly test our theorizing in a single field setting(albeit in an economically and organizationallyimportant one). Future research examining howthese phenomena play out in other important or-ganizational contexts would undoubtedly be use-ful. Our experiments may also be susceptible todemand effects, which could limit their externalvalidity. In addition, in our field setting and in ourexperiments, the groups we examine are relativelysmall in size (i.e., fewer than 20 members). Addi-tional research exploring how group sizemoderatesthe effects of descriptive social norms and scrutinycould be informative. For example, in larger groups,the behavior of peer groups could feel less relevantas the size of the group might create a greater senseof its uniqueness, thereby reducing pressure toconform to descriptive social norms. Alternatively,larger groups may feel more scrutinized because oftheir size, leading them to react more dramaticallyto descriptive social norms.

Finally, more research into the psychologicalmechanisms that lead descriptive social norms andscrutiny to produce the group diversity thresholdeffects we document could be illuminating. Pastresearch has suggested that norms may be particu-larly relevant in the context of group diversity de-cisions because of ambiguity about how muchdiversity is enough and the fear of being singledout from peers (Ahmadjian & Robinson, 2001;Festinger, 1954; Sherif, 1936; Zavyalova et al.,2012). Future research isolating the specific mech-anisms through which descriptive social normsexert their influence would be valuable and couldhelp identifypotent interventions for changing salientnorms. Future research testing new interventions toreduce the reliance on descriptive social norms and


make other norms more salient would also be ex-tremely valuable.

CONCLUSION

Our work highlights the important roles that de-scriptive social norms, goal setting, scrutiny, andvisibility play in shaping decisions about group di-versity, while answering questions about how in-dividuals assess whether a group is diverse and howgroups respond to scrutiny around their diversitylevels. We find empirical evidence that descriptivesocial norms and threshold effects lead to an over-abundance of groups with exactly the same level ofdiversity in an important organizational context,providing evidence of a previously unexploredphenomenon that may contribute to the underrepre-sentation of women and minorities in many organi-zational groups. By shedding light on novel factorsthat influence group diversity decisions, we illumi-nate potential new avenues for increasing the di-versity of groups.

REFERENCES

Aguilera, R. V., Rupp, D. E.,Williams, C. A., &Ganapathi,J. 2007. Putting the S back in corporate social re-sponsibility: Amultilevel theory of social change inorganizations. Academy of Management Review,32: 836–863.

Ahmadjian, C. L., & Robinson, P. 2001. Safety in numbers:Downsizing and the deinstitutionalization of perma-nent employment in Japan. Administrative ScienceQuarterly, 46: 622–654.

Allen, E. J., Dechow, P. M., Pope, D. G., & Wu, G. 2016.Reference-dependent preferences: Evidence frommarathon runners.Management Science, 63: 1657–1672.

Angrist, J. D., & Pischke, J.-S. 2008. Mostly harmlesseconometrics: An empiricist’s companion. Prince-ton, NJ: Princeton University Press.

Avery, D. R. 2003. Reactions to diversity in recruitmentadvertising—are differences black andwhite? Journalof Applied Psychology, 88: 672–679.

Avery, D. R., & McKay, P. F. 2006. Target practice: Anorganizational impression management approach toattracting minority and female job applicants. Per-sonnel Psychology, 59: 157–187.

Bainbridge, S. M., & Henderson, M. T. 2014. Boards-r-us:Reconceptualizing corporate boards. Stanford LawReview, 66: 1051–1120.

Bell, J. M., & Hartmann, D. 2007. Diversity in everyday dis-course: The cultural ambiguities and consequences of“happy talk.”AmericanSociological Review, 72: 895–914.

Bolino, M. C., Kacmar, K. M., Turnley, W. H., & Gilstrap,J. B. 2008. A multi-level review of impression man-agement motives and behaviors. Journal of Manage-ment, 34: 1080–1109.

Brammer, S., & Millington, A. 2006. Firm size, organiza-tional visibility and corporate philanthropy: An empiri-cal analysis. Business Ethics: A European Review, 15:6–18.

Brands, R. A., & Fernandez-Mateo, I. 2017. Leaning out:How negative recruitment experiences shape women’sdecisions to compete for executive roles. Administra-tive Science Quarterly, 62: 405–442.

Buckley, C. 2016. Another Oscar year, another all-whiteballot. New York Times, January 15. Retrieved fromhttps://www.nytimes.com/2016/01/16/movies/oscar-ballot-is-all-white-for-another-year.html.

Campbell, J. L. 2007. Why would corporations behave insocially responsible ways? An institutional theory ofcorporate social responsibility. Academy of Man-agement Review, 32: 946–967.

Castilla, E. J., & Benard, S. 2010. The paradox of meritoc-racy in organizations. Administrative Science Quar-terly, 55: 543–576.

Chiu, S.-C., & Sharfman, M. 2011. Legitimacy, visibility,and the antecedents of corporate social performance:An investigation of the instrumental perspective.Journal of Management, 37: 1558–1585.

Cialdini, R.B. 2003.Craftingnormativemessages toprotectthe environment. Current Directions in Psychologi-cal Science, 12: 105–109.

Cialdini, R. B. 2007. Descriptive social norms as un-derappreciated sources of social control. Psychome-trika, 72: 263–268.

Cialdini, R. B., Kallgren, C. A., & Reno, R. R. 1991. A focustheory of normative conduct: A theoretical refinementand reevaluation of the role of norms in human be-havior. Advances in Experimental Social Psychol-ogy, 24: 201–234.

Cialdini, R. B., Reno, R. R., & Kallgren, C. A. 1990. A focustheory of normative conduct: Recycling the concept ofnorms to reduce littering in public places. Journal ofPersonality and Social Psychology, 58: 1015–1026.

Cialdini, R. B., & Trost,M. R. 1998. Social influence: Socialnorms, conformity and compliance. In D. T. Gilbert,S. T. Fiske, & G. Lindsey (Eds.), The handbook ofsocial psychology, vol. 2: 151–192. Boston, MA:McGraw-Hill.

Coffman, L. C., Featherstone, C. R., & Kessler, J. B. 2017.Can social information affect what job you choose and


https://www.nytimes.com/2016/01/16/movies/oscar-ballot-is-all-white-for-another-year.html

https://www.nytimes.com/2016/01/16/movies/oscar-ballot-is-all-white-for-another-year.html

keep? American Economic Journal. Applied Eco-nomics, 9: 96–117.

Desai, V. M. 2011. Mass media and massive failures: De-termining organizational efforts to defend field legiti-macy following crises. Academy of ManagementJournal, 54: 263–278.

Dezs}o,C. L., Ross,D.G., &Uribe, J. 2016. Is there an implicitquota on women in top management? A large‐samplestatistical analysis. Strategic Management Journal,37: 98–115.

DiMaggio, P., & Powell, W. W. 1983. The iron cage revis-ited: Collective rationality and institutional isomorphismin organizational fields. American SociologicalReview, 48: 147–160.

Dobbin, F., Schrage, D., & Kalev, A. 2015. Rage against theiron cage: The varied effects of bureaucratic personnelreforms on diversity.AmericanSociological Review,80: 1014–1044.

Elsbach,K.D., &Sutton,R. I. 1992.Acquiringorganizationallegitimacy through illegitimate actions: A marriage ofinstitutional and impression management theories.Academy of Management Journal, 35: 699–738.

Elsbach,K.D., Sutton, R. I., & Principe,K. E. 1998.Avertingexpected challenges through anticipatory impressionmanagement: A study of hospital billing. Organiza-tion Science, 9: 68–86.

Festinger, L. 1954. A theory of social comparison pro-cesses. Human Relations, 7: 117–140.

Fombrun, C. 1995. Reputation: Realizing value from thecorporate image. Brighton, MA: Harvard BusinessSchool Press.

Fombrun, C., & Shanley, M. 1990. What’s in a name?Reputation building and corporate strategy.Academyof Management Journal, 33: 233–258.

Forbes, D. P., & Milliken, F. J. 1999. Cognition and corpo-rate governance: Understanding boards of directors asstrategic decision-making groups. Academy of Man-agement Review, 24: 489–505.

Gardberg, N. A., & Fombrun, C. J. 2006. Corporate citi-zenship: Creating intangible assets across institutionalenvironments. Academy of Management Review,31: 329–346.

Gino, F., & Pierce, L. 2010. Robin Hood under the hood:Wealth-based discrimination in illicit customer help.Organization Science, 21: 1176–1194.

Goldstein, N. J., Cialdini, R. B., & Griskevicius, V. 2008. Aroomwith aviewpoint:Using social norms tomotivateenvironmental conservation in hotels. Journal ofConsumer Research, 35: 472–482.

Greve, H. R. 2008. A behavioral theory of firm growth:Sequential attention to size and performance goals.Academy of Management Journal, 51: 476–494.

Heath, C., Larrick, R. P., & Wu, G. 1999. Goals as referencepoints. Cognitive Psychology, 38: 79–109.

Highhouse, S., Brooks, M. E., & Gregarus, G. 2009. An or-ganizational impression management perspective onthe formation of corporate reputations. Journal ofManagement, 35: 1481–1493.

Kallgren, C. A., Reno, R. R., & Cialdini, R. B. 2000. A focustheory of normative conduct: When norms do and donot affect behavior. Personality and Social Psychol-ogy Bulletin, 26: 1002–1012.

Kanter, R. M. 1977. Some effects of proportions on grouplife: Skewed sex ratios and responses to tokenwomen.American Journal of Sociology, 82: 965–990.

King, B. G. 2008. A political mediation model of corporateresponse to social movement activism. Administra-tive Science Quarterly, 53: 395–421.

Konrad, A. M., Kramer, V., & Erkut, S. 2008. Criticalmass: The impact of three or more women oncorporate boards. Organizational Dynamics, 37:145–164.

Leary, M. R., & Kowalski, R. M. 1990. Impression man-agement: A literature review and two-componentmodel. Psychological Bulletin, 107: 34–47.

Lee, J. C. 2017. Trump’s cabinet so far is more white andmale than any first cabinet since Reagan’s. New YorkTimes, January 13. Retrieved from https://www.nytimes.com/interactive/2017/01/13/us/politics/trump-cabinet-women-minorities.html.

Leslie, L. M., Manchester, C., & Dahm, P. 2017. Why andwhen does the gender gap reverse? Diversity goalsand the pay premium for high potential women. Acad-emy of Management Journal, 60: 402–432.

Locke, E. A., & Latham, G. P. 2002. Building a practicallyuseful theory of goal setting and taskmotivation: A 35-year odyssey. American Psychologist, 57: 705–717.

McDonnell, M. H., & King, B. 2013. Keeping up appear-ances reputational threat and impression manage-ment after social movement boycotts.AdministrativeScience Quarterly, 58: 387–419.

Merchant, N. 2013. Viewpoint: Twitter’s all-male boardspells failure. Time, October 7. Retrieved from http://ideas.time.com/2013/10/07/viewpoint-twitters-all-male-board-spells-failure/.

Miller, C. C. 2013. Curtain is rising on a tech premierewith(as usual) a mostly male cast. New York Times, Oc-tober 4. Retrieved from http://www.nytimes.com/2013/10/05/technology/as-tech-start-ups-surge-ahead-women-seem-to-be-left-behind.html.

Miller, C. C. 2014. Women on the board: Quotas havelimited success. New York Times, June 19. Retrievedfrom https://www.nytimes.com/2014/06/20/upshot/women-on-the-board-quotas-have-limited-success.html.


https://www.nytimes.com/interactive/2017/01/13/us/politics/trump-cabinet-women-minorities.html



http://ideas.time.com/2013/10/07/viewpoint-twitters-all-male-board-spells-failure/



http://www.nytimes.com/2013/10/05/technology/as-tech-start-ups-surge-ahead-women-seem-to-be-left-behind.html



https://www.nytimes.com/2014/06/20/upshot/women-on-the-board-quotas-have-limited-success.html



Moskowitz, T., & Wertheim, L. J. 2011. Scorecasting:The hidden influences behind how sports areplayed and games are won. New York, NY: CrownArchetype.

Nolan, J. M., Schultz, P.W., Cialdini, R. B., Goldstein, N. J.,& Griskevicius, V. 2008. Normative social influenceis underdetected.Personality and Social PsychologyBulletin, 34: 913–923.

Oliver, C. 1991. Strategic responses to institutional pro-cesses. Academy of Management Review, 16: 145–179.

Paluck, E. L., & Shepherd, H. 2012. The salience of socialreferents: A field experiment on collective norms andharassment behavior in a school social network.Journal of Personality and Social Psychology, 103:899–915.

Pierce, L., Snow,D. C., &McAfee, A. 2015. Cleaning house:The impact of information technology monitoring onemployee theft and productivity. Management Sci-ence, 61: 2299–2319.

Pollock, T. G., & Rindova, V. P. 2003. Media legiti-mation effects in the market for initial publicofferings. Academy of Management Journal, 46:631–642.

Pope, D., & Simonsohn, U. 2011. Round numbers as goalsevidence from baseball, SAT takers, and the lab. Psy-chological Science, 22: 71–79.

Prentice, D. A., & Miller, D. T. 1993. Pluralistic ignoranceand alcohol use on campus: Some consequences ofmisperceiving the social norm. Journal ofPersonalityand Social Psychology, 64: 243–256.

Rubinstein, R. Y., &Kroese, D. P. 2011.Simulation and theMonte Carlo method, vol. 707. Hoboken, NJ: JohnWiley & Sons.

Ryan, P. 2016. #OscarsSoWhite controversy: What youneed to know. USA Today, February 2. Retrievedfrom https://www.usatoday.com/story/life/movies/2016/02/02/oscars-academy-award-nominations-diversity/79645542/.

Schultz, P.W., Nolan, J. M., Cialdini, R. B., Goldstein, N. J.,&Griskevicius, V. 2007.The constructive, destructive,and reconstructive power of social norms. Psycho-logical Science, 18: 429–434.

Shaffer, L. S. 1983. Toward Pepitone’s vision of a norma-tive social psychology:What is a social norm? Journalof Mind and Behavior, 4: 275–293.

Shemla, M., Meyer, B., Greer, L., & Jehn, K. A. 2016. Areview of perceived diversity in teams: Does howmembers perceive their team’s composition affectteam processes and outcomes? Journal of Organiza-tional Behavior, 37: S89–S106.

Sherif, M. 1936. The psychology of social norms. NewYork, NY: Harper & Brothers.

Simon, M., & Houghton, S. M. 2003. The relationship be-tween overconfidence and the introduction of riskyproducts: Evidence from a field study. Academy ofManagement Journal, 46: 139–149.

Smale, A., & Miller, C. C. 2015. Germany sets genderquota in boardrooms. New York Times, March 6. Re-trieved from https://www.nytimes.com/2015/03/07/world/europe/german-law-requires-more-women-on-corporate-boards.html.

S&P Dow Jones Indices. 2015. S&P U.S. indices method-ology, October 2015. Retrieved from https://us.spindices.com/indices/equity/sp-500.

Staats, B. R., Dai, H., Hofmann, D., & Milkman, K. L. 2016.Motivating process compliance through individualelectronic monitoring: An empirical examination ofhand hygiene in healthcare. Management Science,63: 1563–1585.

Staw, B. M. 1991. Dressing up like an organization: Whenpsychological theories can explain organizational ac-tion. Journal of Management, 17: 805–819.

Sutton, R. I., &Galunic, D. C. 1996. Consequences of publicscrutiny for leaders and their organizational imageand its management. Research in OrganizationalBehavior, 18: 201–250.

Swim, J. K., & Hyers, L. L. 1999. Excuse me—what did youjust say?!: Women’s public and private responses tosexist remarks. Journal of Experimental Social Psy-chology, 35: 68–88.

Tankard, M. E., & Paluck, E. L. 2017. The effect of a Su-premeCourt decision regarding gaymarriage on socialnorms andpersonal attitudes.Psychological Science,28: 1334–1344.

Torchia, M., Calabro, A., & Huse, M. 2011. Women di-rectors on corporate boards: From tokenism to criticalmass. Journal of Business Ethics, 102: 299–317.

Totenberg, N. 2016. Donald Trump unveils new, more di-verse Supreme Court short list. NPR, September 23.Retrieved from http://www.npr.org/2016/09/23/495216645/donald-trump-unveils-new-more-diverse-supreme-court-short-list.

Unzueta,M.M., &Binning,K. R. 2010.Which racial groupsare associated with diversity? Cultural Diversity &Ethnic Minority Psychology, 16: 443–446.

Unzueta, M. M., & Binning, K. R. 2012. Diversity is in theeye of the beholder: How concern for the in-group af-fects perceptions of racial diversity. Personality andSocial Psychology Bulletin, 38: 26–38.

Unzueta,M.M., Knowles, E. D., & Ho, G. C. 2012. Diversityis what you want it to be: How social-dominance mo-tives affect construals of diversity. PsychologicalScience, 23: 303–309.

Zavyalova, A., Pfarrer, M. D., Reger, R. K., & Shapiro, D. L.(2012). Managing the message: The effects of firm


https://www.usatoday.com/story/life/movies/2016/02/02/oscars-academy-award-nominations-diversity/79645542/



https://www.nytimes.com/2015/03/07/world/europe/german-law-requires-more-women-on-corporate-boards.html



https://us.spindices.com/indices/equity/sp-500

https://us.spindices.com/indices/equity/sp-500

http://www.npr.org/2016/09/23/495216645/donald-trump-unveils-new-more-diverse-supreme-court-short-list



actions and industry spillovers on media coverage fol-lowing wrongdoing. Academy of Management Jour-nal, 55: 1079–1101.

Edward H. Chang ([email protected]) is aPhD candidate at the Wharton School of the Universityof Pennsylvania. His research interests include diversity,discrimination, and behavior change.

KatherineL.Milkman ([email protected]) isthe EvanCThompsonEndowedTermChair for Excellencein Teaching and a Professor at the Wharton School of theUniversity of Pennsylvania. Her research relies on fielddata and field experiments to understand the forces thatshape people’s decisions. She received her PhD in com-puter science and business from Harvard University.

Dolly Chugh ([email protected]) is an Associate Pro-fessor in the Management and Organizations group at theNewYorkUniversity SternSchool ofBusiness. She receivedher PhD in social psychology and organization behaviorfrom Harvard University. Her research interests includebounded ethicality and unconscious bias. Dolly’s first book,The Person You Mean to Be: How Good People Fight Bias,was released by HarperCollins in September 2018.

Modupe Akinola ([email protected]) is an Associ-ate Professor ofManagement at ColumbiaBusiness School.She received her PhD in organizational behavior fromHarvard University. Her research explores how stress af-fects workplace performance. She also examines thestrategies organizations employ to increase diversity, aswell as the biases that affect the recruitment and retentionof women and minorities in organizations.


mailto:[email protected]




Copyright of Academy of Management Journal is the property of Academy of Managementand its content may not be copied or emailed to multiple sites or posted to a listserv withoutthe copyright holder's express written permission. However, users may print, download, oremail articles for individual use.

Date post:	10-Jan-2022
Category:	Documents
Upload:	others
View:	1 times
Download:	0 times

Diversity Thresholds: How Social Norms, Visibility, And ...

Documents