from bench to board

FROM BENCH TO BOARD: GENDER DIFFERENCES INUNIVERSITY SCIENTISTS PARTICIPATION IN CORPORATE

SCIENTIFIC ADVISORY BOARDS

WAVERLY W. DINGUniversity of Maryland

FIONA MURRAYMassachusetts Institute of Technology

TOBY E. STUARTUniversity of California, Berkeley

This article examines the gender difference in the likelihood that male and femaleacademic scientists will join corporate scientific advisory boards (SABs). We assess (i)demand-side theories that relate the gap in scientists rate of joining SABs to theopportunity structure of SAB invitations, and (ii) supply-side explanations that attri-bute that gap to scientists preferences for work of this type. We statistically examinethe demand- and supply-side perspectives in a national sample of 6,000 life scientistswhose careers span more than 30 years. Holding constant professional achievement,network ties, employer characteristics, and research foci, male scientists are almosttwice as likely as females to serve on the SABs of biotechnology companies. We do notfind evidence in our data supporting a choice-based explanation for the gender gap.Instead, demand-side theoretical perspectives focusing on gender-stereotyped percep-tions and the unequal opportunities embedded in social networks appear to explainsome of the gap.

The gap between men and women in labor mar-ket outcomes has been gradually decreasing for de-cades, but the underrepresentation of women inhigh-level corporate positions stubbornly persists.Among Fortune 500 companies, for example,women currently hold only 16.6 percent of allboard seats and 14.3 percent of all executive-levelpositions (Catalyst, 2013).Scholars seeking to explain this gap generally

have approached the issue from one of two distincttheoretical approaches. Demand-side perspectivesdeveloped in sociology and social psychology fo-cus on the mechanisms that limit opportunities forwomen to gain access to jobs of certain types. Bycontrast, supply-side perspectives developed ineconomics have emphasized that individual differ-ences in preferences lead to worker-level choicesthat drive labor market outcomes. According to the

latter view, skill and preference differences be-tween the gendersrather than systematic biasesin the workplaceexplain gender-based differ-ences in career outcomes.Vibrant research streams underlie both perspec-

tives. Findings from them, however, imply quitedifferent remedies to the persistent gender gap.Moreover, to the best of our knowledge, there havebeen no longitudinal studies that empirically com-pare, side-by-side, the relative importance of thetwo broad families of theories. Most archival de-mand-side studies that have investigated discrimi-nation in the evaluation of women for top positions(e.g., Gorman, 2005; Gorman & Kmec, 2009; West-phal & Stern, 2006) implicitly assume that all eli-gible candidates are motivated to pursue the posi-tions of interest. Conversely, supply-side studiesoften regress career outcomes (e.g., compensation)on a collection of explanatory variables that arebelieved to be proxies for human capital andworker preferences. In such studies, then, the re-sidual, unexplained variance in pay or promotionmay be attributed to discrimination (e.g., Bertrand& Hallock, 2001; Goldin & Katz, 2008).

We thank seminar participants at Stanford, Duke, Co-lumbia, Harvard, MIT, and Maryland for their helpfulcomments. Financial support from the Kauffman Foun-dation Center for Entrepreneurial Research and the Cam-bridge-MIT Institute is gratefully acknowledged.

1443

Academy of Management Journal2013, Vol. 56, No. 5, 14431464.http://dx.doi.org/10.5465/amj.2011.0020

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 holders expresswritten permission. Users may print, download, or email articles for individual use only.

We believe that an integrated empirical test in-corporating both supply- and demand-side per-spectives is of considerable value. On the one hand,assuming away supply-side explanations under-cuts the validity of research that has found system-atic discrimination in the opportunity structure forwomen to ascend to the top of a corporation. In thissense, studies that account for supply-side explan-atory factors are necessary for accurate estimates ofthe extent of gender-based discrimination. Ofcourse, the same logic applies conversely: the per-suasiveness of evidence for supply-side theories isdiminished in the absence of accounting for work-place biases. Therefore, integrative investigationsare necessary to begin to adjudicate between thetwo families of explanations and to better pinpointthe factors that contribute most to the gender gap insenior-level corporate roles.Undoubtedly, a primary explanation for the pau-

city of research that jointly considers both the sup-ply and demand sides of the equation is the chal-lenge of collecting appropriate data. In manyempirical settings, it can be difficult even to definethe pool of candidates for open positions. In situa-tions in which the candidate pool can be approxi-mately bounded (e.g., Gorman et al. [2005, 2009] forpartnerships at law firms; Westphal and Stern[2006] for board positions at Fortune 500 firms),researchers still face the daunting task of collectingdetailed career histories for members of the pool.The empirical approach in this article overcomes

some of these limitations. First, we study the gen-der gap in representation on scientific advisoryboards (SABs) in the biotechnology industry. Theimportance of greater female representation onthese boards is certainly one reason for this choice.Another attractive feature of SABs is that the pro-cess of appointing individuals to them largely re-sembles that for a corporate board of directorsinboth contexts, new members join by invitation (i.e.,one typically does not volunteer for such roles),and a variety of social mechanisms are invoked inthe vetting process (Khurana, 2002). More impor-tantly, we chose SAB members in the life sciencesindustry, our study context, from the pool of uni-versity professors in relevant fields of science. Thisallowed us to identify the candidate pool and toconstruct a data set of eligible recruits. Likewise,because our study population consists of academicscientists, there is a wealth of publicly availableinformation about their career histories and profes-sional achievements, and with a little creative li-cense, we also can construct indicators of the ex-

tent to which a given academic scientist hasexpressed interest in working with private sectorcompanies. Together, the career histories and theproxies for interest in opportunities in SAB posi-tions enable us to simultaneously assess demand-and supply-side perspectives.Theoretically, we build most heavily from four

streams of existing literature to develop our hy-potheses about gender differences in participa-tion in SABs: (i) expectation-state theory thatconsiders when gender is used by an evaluator toinfer a candidates competence (Berger, Cohen, &Zelditch, 1972; Ridgeway & Erickson, 2000); (ii)role congruity theory (Eagly & Karau, 2002; Eagly,Karau, & Makhijani, 1995; Eagly, Makhijani, &Klonsky, 1992) and lack-of-fit theory (Heilman,1983, 2001), which argue that the disadvantageswomen face in the workforce arise from an incon-gruity between the qualifications for a position andcharacteristics that a female candidate is conven-tionally expected to possess; (iii) social networktheory, which often finds gender-based differencesin network structures, with implications for accessto the professional opportunities that are allocatedacross networks; and (iv) a supply-side perspectivethat attributes the gender gap to individual careerchoices and preferences and maintains that the re-sulting accumulation of human capital determinescareer outcomes (Becker, 1985; Bertrand, Goldin, &Katz, 2010).Before developing our hypotheses, we first pres-

ent a brief overview of the role of SABs. We thenformulate predictions regarding gender differencesin the likelihood of joining SABs. These hypothe-ses are then tested in a case-cohort data archivecontaining career histories of a national sample of6,000 university-employed life scientists.

SCIENTIFIC ADVISORY BOARDS

Functions of SABs

Despite current research interest in commercialendeavors in academic settings, SABs have re-ceived little attention in the literature to date and,in consequence, their function is not well under-stood. As is evident from our interviews of scien-tists who have been SAB members, SABs have nei-ther fiduciary responsibility nor a formal place infirms governance structures.1 Nevertheless, they

1 As part of a broader research program that examinesthe underpinnings of the transition to commercial sci-

1444 OctoberAcademy of Management Journal

have become a near-ubiquitous feature of biotech-nology companies. Typically, a SAB is formed by afounding team very early in the development of abiotechnology company and consists of five to tenmembers who join by invitation. The founding sci-entist(s) of the company and its investors identifysuitable SAB candidates through their contact net-works or their knowledge of the experts in thecompanys field of research. Once formed, SABsremain quite stable, and replacement of members isuncommon. Board meetings typically occur quar-terly, during which SAB members review theirfirms key research projects. SAB members are typ-ically paid a fee and granted stock options that canrun as high as 2 percent of a start-up companysequity.The scientists we interviewed believe that SABs

perform three primary functions. First, they pro-vide expertise ranging from very specific tacitknowledge to general advice on broad scientificstrategies and experimental designs. In addition tooffering expertise, a second function of SABs is tosignal scientific quality to external investors. Ourinterviewees often likened advisory boards towindow dressing. In effect, prestigious academicscientists lend their reputations to the early-stagefirms they advise, which aids firms in attractingfinancial resources and recruiting scientific talent.A third obligation is that SAB members are ex-pected to share their social networks with thefirms: they assist in identifying other academicsthat might provide a critical resource through col-laborative research, and they locate suitable stu-dents to be hired by the firms.

Why Study SABs?

Research design considerations aside, our deci-sion to study SABs is motivated by the substantiveimportance of these advisory boards. First, we notethat among the small number of studies that haveinvestigated SABs (Audretsch & Stephan, 1996;Ding & Choi, 2011; Louis, Blumenthal, Gluck, &Stoto, 1989; Stuart & Ding, 2006), none has ad-dressed the gender gap in participation rates.

Second, Whittington and Smith-Doerr (2005) ar-gued that female involvement in commercial sci-ence may be aptly characterized by a gender-basedpipeline analogy (Berryman, 1983). As the degreeof commercial involvement increases and the levelof time and identity commitment intensifiesforinstance, as one moves across a continuum fromconsulting to patenting to SABmember to foundinga companyone would expect a decreasing num-ber of women scientists. Given that the time com-mitment and financial and reputational rewards forSAB involvement are more substantial than thosefor episodic commercial activities such as patent-ing and consulting, in the pipeline analogy, SABmembership is nearer to the pinnacle of commer-cial engagement than other widely studied prac-tices, most notably patenting.Third, we believe that the gender composition

of SABs has implications for female representa-tion in biopharmaceutical firms and possibly forthe career outcomes of women employees inthese firms. To the extent that in-group favorit-ism and social network ties influence recruiting,having more women in key corporate positionssuch as board, advisory board, and senior execu-tive ranks should create opportunities for otherwomen to ascend to senior-level positions (Beck-man & Phillips, 2005; Cohen, Broschak, & Have-man, 1998; Gorman, 2005; Gorman & Kmec,2009). In addition, research on gender effects insupervisory and mentoring relationships (Tsui &OReilly, 1989) has suggested that the presence ofwomen at the top of an organization has a posi-tive, trickle-down effect on the career experi-ences of female employees.Fourth, SAB membership has important implica-

tions for the gender gap in compensation in aca-demia. A recent survey of newly public biotechnol-ogy companies (those going public in 2004) foundthat in half of the firms, university faculty had largeenough equity holdings to be listed in Securitiesand Exchange Commission (SEC) filings, with amedian value of $5.6 million (Edwards, Murray, &Yu, 2006). Apart from direct financial returns, SABmembership also bolsters scientists industry repu-tations, which may translate into lucrative consult-ing work. Therefore, opportunities for extramuralwork such as SABs now meaningfully influenceearnings differences among scientists, and any gen-der differences in participation on SABs may be-come a growing source of earnings inequality.

ence among academic scientists, we conducted extensiveinterviews with a gender-matched sample of life scien-tists at one of the universities that has been most active inthe commercialization of academic science. For furtherdetails, please see Murray and Graham (2007); Ding,Murray, and Stuart (2010).

2013 1445Ding, Murray, and Stuart

HYPOTHESES

Building on the literature on gender and careers,we formulate hypotheses that specify demand-sideand supply-side explanations for gender differ-ences in faculty participation on SABs.

Overall Gender Gap

Our first objective was to determine whetherthere is a gender gap in the participation of univer-sity scientists in for-profit companies SABs, and ifso, to gauge its magnitude. There is ample reason toanticipate that such a gap may exist. First, a longtradition of research examines the effect of genderon different aspects of scientific careers. With fewexceptions, this work has concluded that femalescientists who are otherwise comparable to theirmale colleagues experience less successful careersby the standard metrics of professional attainment(Fox, 2001; Haberfeld & Shenhav, 1990; Long &Fox, 1995). Specifically, women scientists are lessproductive than men (Cole & Zuckerman, 1984;Long, 1990; Reskin, 1978; Xie & Shauman, 1998);they are less likely to be on the faculties of eliteinstitutions (Long & Fox, 1995); they advancethrough ranks at a slower rate than men (Farber,1977; Long, Allison, & McGinnis, 1993; NationalScience Foundation [NSF], 2005); and a salary gapcontinues to exist (Haberfeld & Shenhav, 1990;NSF, 2005).Some evidence suggests a gender gap exists in

commercial science as well. In earlier research(Ding, Murray, & Stuart, 2006), we found that overa 30-year period, men patented at about 2.5 timesthe rate of women. Whittington and Smith-Doerr(2005) reached a similar conclusion in studies ofNational Institutes of Health grant awardees. In astudy of 4,500 faculty members, Thursby andThursby (2005) showed that women are less likelythan men to disclose potentially patentable inven-tions to university technology transfer offices. Cor-ley and Gaughan (2005) found that women facultyengage in less consulting than men. Finally, Rosaand Dawson (2006) found that only 12 percent ofthe founders of all companies spun out of UK uni-versities were female.Because SABs typically are created early in the

life of a new company, the broader evidence con-cerning gender differences in other forms of entre-preneurship in society at large may be informativein this context. For instance, there is known to be awide gender gap in company founding rates. Sta-

tistics indicate that men start new businesses atapproximately twice the rate of women (U.S. SmallBusiness Administration, 2001). Moreover, the dis-parity between the genders in rates of businessfounding and in occupancy of high-level manage-rial positions appears to increase with the techno-logical intensity of the sector (Baron, Hannan, Hsu,& Kocak, 2008). For example, a study of venturecapital funding activity in 2000 found that just6 percent of the $69 billion funds dispensed by theindustry was invested in companies with a femalechief executive officer (Brush, Carter, Gatewood,Greene, & Hart, 2001).Given the sizable gender gap in academic career

outcomes, the documented gap in other forms ofcommercial science, and the underrepresentationof women in technology-sector entrepreneurship,we posit as a baseline expectation that women ac-ademic scientists are less likely than men to jointhe SABs of biotechnology firms.

Hypotheses from the Demand-Side Perspective

Assuming a gender gap in SAB membership,what are the factors that might contour it? Theliterature focuses on two broad classes of mecha-nisms underlying gender differences in career out-comes: demand-side biases in the workplace thatcause women whose qualifications are equal tothose of male colleagues to obtain fewer opportu-nities to join advisory boards, and supply-side ex-planations that emphasize the exercise of individ-ual choices that shrink the pipeline or diminish theinterest of eligible women candidates with the rightqualifications for positions of this type. In demand-side perspectives, the dominant logic is one ofblocked mobility because of prevalent workplacebiases. Conversely, supply-side arguments arebased on differences in individuals choices to in-vest in human capital or to supply labor.We begin with demand-side perspectives. In con-

sidering these theories, we follow the general ap-proach in Petersen and Saporta (2004), who assumethat gender-based discrimination is widespread inorganizations, and then investigate the circum-stances under which the prevalence of discrimina-tion is expected to vary. Similar to Petersen andSaporta (2004), we do not directly observe discrim-ination. Rather, our research strategy asks, if dis-crimination is commonplace, what does this implyin terms of contextual factors that theory predictswill moderate its effect? We derive hypothesesfrom the implications of the demand-side theories


that suggest amplifiers or dampeners of the base-case level of the observed gender gap.We focus on three groups of theory concerning

how gender may influence an evaluators assess-ment of an applicants qualifications for a job: (i)expectation states theory, (ii) role congruity andlack-of-fit theories, and (iii) in-group favoritismand social networks. No doubt, these theories com-prise just a subset of the theoretical possibilitiesthat could explain the gender gap. For instance, wedo not address demographic perspectives (cf. Beck-man & Phillips, 2005; Cohen et al., 1998), jobmatching theory (Reskin & Roos, 1990), or theoriesof the impact of specific organizational practices(cf. Dencker, 2008; Roth, 2006). In formulating ourargument, our theoretical lenses were focused bythe relevance of theories to our research contextand by what we could measure in our empiri-cal data.Expectation states theory. Expectation states

theory describes how status beliefs are constructedduring the process of social interaction. Ridgewaydefined status beliefs as widely held cultural be-liefs that link greater social significance and gen-eral competence, as well as specific positive andnegative skills, with one category of a social dis-tinction (e.g., men) compared to another (e.g.,women) (2001: 638). Research has found that inmany contexts, gender is a prominent social cate-gory that is associated with anticipated perfor-mance levels, even when there is no actual relation-ship between gender and relevant outcomes (e.g.,Berger et al., 1972; Ridgeway & Erickson, 2000). Inother words, an individuals gender may elicit pre-sumptions about his or her ability to perform aparticular task or roleand therefore influencewhether that person is presented with opportuni-ties to execute the task in the first instanceincontexts in which there are gender-based assump-tions about ability. This process is documented tooccur in many situations in which gender is unre-lated to ability.In the context of SABs, we know from the general

function of advisory boards that there is a premiumplaced on selecting advisors who are perceived tobe of high status within the scientific community.Moreover, we know that success in the resourcemobilization process for new ventures hinges onthe legitimacy and the status of the entrepreneursattempting to attract resources (e.g., Shane & Stuart,2002; Stuart, Hoang, & Hybels, 1999). Even when awoman occupies the same formal position as a man(e.g., professor at the same university), the fact that

women are atypical occupants of such positionsmay lead SAB selectors to form gendered statusbeliefs that hold women as less competent thanmen at performing the tasks of a SAB member(Ibarra, 1992; Kanter, 1977; Lucas, 2003; Ridgeway& Smith-Lovin, 1999).Expectation states theory contrasts with supply-

side explanations of gender differences in opportu-nity structures of the type we study, in that it positsthat actors often hold different, gender-based per-ceptions of candidates competence; supply-sideperspectives instead attribute the difference to can-didates competence per se. While supply-side per-spectives attribute the gap to womens lack of in-terest in or qualifications for SAB positions,expectations states theory points us to the differ-ences in SAB selectors subjective interpretation ofcandidates qualifications based on gender or othercategorical differences.Assuming that subjective frameworks rooted in

perceptions of gender differences affect evaluationsof potential SAB members, expectation states the-ory implies two possible outcomes. First, a gap mayexist between a candidates true qualifications andan evaluators perception of them, and this gapvaries between genders. To the extent that beingfemale elicits a presumption of inferior status,women scientists are likely to be disadvantaged inthe SAB vetting process, even when they hold thesame qualifications as their male counterparts. Sec-ond, if subjective evaluative frameworks play astrong role in SAB selection, then the theory im-plies that the degree of gender bias is likely to varywith how and which subjective frameworks areinvoked in the evaluation process. When multipleframeworks jointly influence the process, the de-gree of negative perceptions of a womans compe-tence can be mitigated (or aggravated), dependingon whether there are competing frameworks thatweaken (or strengthen) the impact of gender-disad-vantageous status beliefs. Lacking direct informa-tion on SAB selectors perceptions, we focus ourhypothesis on the second of these two possibilities.Social psychologists have shed light on howmul-

tiple status categories operate together. Researchshows that, although it is often salient, gender isone of many ways in which individuals are catego-rized. As information on additional social catego-ries becomes available, it cognitively nests withingender categories and may alter the interpretationof gender identities (Stangor, Lynch, Duan, & Glass,1992). Ridgeway and Correll (2000) observed thatother identities, particularly those based on insti-


tutional roles (e.g., CEO, university professor), mayoperate in the foreground, while gender operates asa background factor in evaluative settings. Theirresearch suggests that, in certain situations, the in-vocation of other social categories as bases of eval-uation may modify gender-based assumptions ofcompetence.Among the kinds of information likely to be in-

voked in the vetting process for prominent corpo-rate positions, employment affiliations loom large.For example, Crane (1965) famously found that sci-entists gain greater recognition from an affiliationwith a prestigious university than they do evenfrom their own high productivity. In the context ofselection of board members for early stage technol-ogy companies, the prestige of a scientists employ-ing university itself lends legitimacy to a venture(Sine, Shane, & Gregorio, 2003; Stuart et al., 1999).To the extent that affiliation-based social categori-zation occurs in the SAB evaluation process, ahigh-prestige university affiliation might diminishreliance on gender as a cue that evaluators use tojudge an individuals qualifications for board mem-bership. If this occurs, we would expect:

Hypothesis 1. An affiliation with a high-pres-tige institution has a greater effect on the like-lihood of women scientists becoming SABmembers than it has on the likelihood for men.

Role congruity and lack-of-fit theories.Next, weconsider role congruity (Eagly et al., 1992; Eagly &Karau, 2002) and lack-of-fit (Heilman, 1983, 2001;Lyness & Heilman, 2006) theories of gender-baseddisadvantages faced by women aspiring to leader-ship positions. Like theory on expectation states,these theories hold that biases are grounded ingender-based stereotypes; however, there is a dif-ferent nuance to how gender stereotypes influenceevaluations. Specifically, prejudice is not causedby a generalized, negative view of women. Rather,it emerges when evaluators view a woman as pos-sessing attributes that are incongruent with thoserequired for a specific position to which she as-pires. These theories propose that women will bedisadvantaged when the leadership position atstake involves traits that are culturally associatedwith men.In forming an SAB, biotechnology founders often

value qualities in a candidate that range from his orher depth of knowledge in specific areas of researchinterest to the firm to the quality of her reputationand network, to whether or not the candidate pos-sesses business savvy. Indeed, both scientific dis-

tinction and business acumen may be qualifica-tions that are incongruent with commonly heldstereotypes about women faculty. For example, inhis famously controversial speech at a 2005 Na-tional Bureau of Economic Research workshop,Larry Summers posited that the underrepresenta-tion of women in tenured positions in science andengineering at top universities might result fromdifferent availability of aptitude [among men andwomen] at the high end (Bombarderi, 2005). Fur-thermore, the numerical dominance of men inother areas of the translation of university scienceinto commercial products (e.g., filing invention dis-closures, patenting, and company founding) sug-gests there are particularly strong gender stereo-types in the set of roles that link academic scienceto the marketplace. As a result, gender incongruitybias may exist in SAB selection.Like expectation states theory, role congruity and

lack-of-fit theories differ from supply-side explana-tions of the gender gap in that they draw a theoret-ical distinction between scientists objective quali-fications and social perceptions of them. Roleincongruitybased biases arise from SAB selectorsbeing unable to correctly evaluate certain qualitiesof women scientists when these qualities are notconventionally associated with the female gender.If, as role congruity and lack-of-fit theories hold,gender-unfavorable social perceptions are thesource of gender gap in SAB memberships, then analteration in how a SAB evaluation is performedmight lead to variation in perceptions and in thedegree of gender-based biases.Indeed, Heilman (2001) showed that gender-ste-

reotype-based biases are more likely to occur insettings where evaluations are informal or informa-tion can be easily distorted during the vetting ofcandidates. For the context we study, this impliesthat an objective record of performance may mattermore for women than for men, as it can reduce thegap between perceived feminine (or masculine)qualities and the true qualifications of candidates.Evidence from Lyness and Heilmans (2006) studycorroborates this point. The authors examined per-formance evaluations and promotion rates of 448managers and found that performance records,which provide quasi-objective evidence that dem-onstrates a managers fitness for a position, aremore strongly associated with promotion forwomen managers than for men.Examples from our field interviews with SAB

scientists also support this conjecture. In describ-ing the flow of opportunities to join SABs, a few


women interviewees highlighted the impact of vis-ible administrative appointments on the arrival ofsuch opportunities. In contrast to their male col-leagues, who more often described a steady flowof opportunities throughout their careers, severalwomen commented, the phone started to ringwhen I became [Dean, provost, director]. Onewoman scientist described to us that about 20invitations [to SABs] followed my getting this new[administrative] position from companies big andsmall. . . . I am not sure why . . . certainly my lablooks at broad problems across many fieldsits anunusually diverse labbut this is not new! I sup-pose with the new job I have achieved a level ofstature or position that people think is interesting.For women, such appointments confirmed theirscientific and management capabilities and therebyhelped them to contend for SAB positions. We thusexpect:

Hypothesis 2. Evidence of scientific or manage-ment capabilities has a greater effect on thelikelihood of women scientists becoming SABmembers than it has on the likelihood for men.

In-group favoritism and social networks. Athird, widely studied demand-side factor that mayinhibit womens ascent to senior corporate posi-tions is in-group favoritism. In her field study of themanagers at a large company, Kanter (1977) ob-served that the nature of managerial tasks, oftencharacterized by high degrees of uncertainty anddiscretion, creates pressure for homogeneity andconformity in hiring and promotion. Candidateswho are members of a minority group (e.g., women)thus face advancement hurdles in such a bureau-cratic kinship system that reproduces the majoritygroups dominance of the senior ranks and author-ity structures in corporations (Moore, 1962).Research in social psychology explains why in-

group members are favored in highly uncertain taskenvironments. Managers enjoy a great amount ofdiscretion in their work, and the level of discretiontypically increases in the level of an organizationalhierarchy. For this reason, existing staff membersprefer to fill any vacant, senior positions with can-didates that are deemed trustworthy. In general,individuals perceive others who share similar so-cial backgrounds, characteristics, or experiences tobe more trustworthy, cooperative, and competent(Fiske, Cuddy, Glick, & Xu, 2002; Insko, Schopler,Hoyle, Dardis, & Graetz, 1990). Any predispositionto favor in-group members among an already en-franchised, majority group will lead to homoso-

cial reproduction of this dominant coalition, espe-cially in employment situations in which a highdegree of trust is required.In-group trust can come from two different types

of homogeneity: similarity in social backgroundsand characteristics (e.g., gender), or proximity inrelational space (Kanter, 1977). This implies thatwhile gender-based homophily is likely to be atplay in the selection process for senior leadershippositions, proximity in social networks may serveas an alternative, and possibly counterbalancing,route for winning trust from the members of aninner circle. Indeed, network researchers have longdocumented that interpersonal ties generate cohe-sion and social obligations (Coleman, 1988). In par-ticular, strong ties through which individuals inter-act with higher frequency and intensity caneffectively convey trust (Reagans & McEvily, 2003),as does comembership in a network that is richwith referrals and that can serve as an informalenforcement mechanism encouraging conformanceto social norms (Macaulay, 1963; Raub & Weesie,1990). In an empirical examination of these argu-ments in an employment context, Seidel, Polzer,and Stewart (2000) found that the presence of so-cial ties to an employer elevate a job applicantsappraised trustworthiness and performance poten-tial in the eyes of the employers recruitingpersonnel.At the interface of industry and academe in the

life sciences, network ties have been found to playa strong role in entrepreneurial processes. For ex-ample, Stuart and Ding (2006) showed that if anacademic life scientist undertook a collaborativeresearch project with a commercially active coau-thor, he or she becomes much more likely thanother faculty members to affiliate with for-profitcompanies. In keeping with previous research onthe role of social networks in facilitating matchesbetween workers and jobs (Fernandez, Castilla, &Moore, 2000; Granovetter, 1973), biotech foundersin our interviews describe tapping into their con-tact networks to identify individuals who would bestrong candidates to join a SAB. However, the maleand female scientists in our interviews describeddifferent network pathways through which theyobtained invitations to SABs. While the male sci-entists mostly mobilized an eclectic and far-flungreferral network made of strong and distal ties, thefemale faculty tended to rely on referrals from alimited circle of close colleagues, collaborators,and students they have advisedstrong, direct ties


through which they had shared research projectsrather than more distant social connections.Combining insights from our field interviews, the

theory of the effect of social network ties on cohe-sion and trust, and the theory of in-group favorit-ism, we conjecture that when a woman scientisthas close ties with insiders in the networks of com-mercial science, these insiders, who have firsthandinformation on the woman scientists qualifica-tions, may help other SAB selectors to overcomethe (male) in-group tendency and develop trust inthe female faculty candidate. In this process, strongtie networks may partially or fully offset in-groupbiases against women. We thus propose:

Hypothesis 3. Location in a strong, direct tienetwork conducive to SAB opportunities has agreater effect on the likelihood of women sci-entists becoming SAB members than it has onthe likelihood for men.

Hypotheses from the Supply-Side Perspective

In the theories above, the gender gap in SABparticipation is caused by a paucity of opportuni-ties available to women. A contending set of expla-nations holds that any observed gap results fromwomens preferences and career choices, ratherthan from a bias-based dwindling in the set of op-portunities available to them.Life cycle. First, there may be life-cycle-related

factors that affect the timing of womens pursuit ofSAB appointment. Studies find that, relative tomen, women in the full-time workforce assumegreater family responsibilities (e.g., Hochschild &Maschung, 1989; Robinson, 1996). Although thereare documented high rates of nonparenting andnonmarriage among women faculty, survey datasuggest that women faculty (but not men) with chil-dren at home work slightly fewer hours per weekthan their male peers (Jacobs & Winslow, 2004). Inaddition, a large fraction of married female facultymembers have spouses with full-time jobs (Jacobs,2004). Therefore, female faculty may have bothgreater nonprofessional time commitments andhigher household incomes than do male faculty,which may dampen women scientists interest inallocating their time to SABs, particularly duringthe early career years. This supply-side, family-career trade-off does not suggest that women willshun commercial opportunities altogether. It does,however, imply that female and male scientists willjoin SABs at different career stages: because they

(on average) have greater family and parenting ob-ligations, women scientists will join SABs at olderages than will male scientists. This rightward shiftin the distribution of age at first joining of SABswill occur if women are interested in working inthese roles but are relatively more time constrainedthan men in the years in which they are likely tohave young children at home. If this choice-basedexplanation holds, we would expect to observe:

Hypothesis 4. Relative to men, women (on av-erage) become SAB members at greaterprofessional ages.

Research preference. A second choice-basedmechanism that would generate a gender gap inSAB participation is whether male and female sci-entists exhibit gender differences in the actual con-tent of research. It is naturally the case that certainareas of scientific research have greater commercialvalue than do others. Moreover, it is likely thatdifferences in commercial relevance often areknowable at the time that research projects are ini-tiated. If women scientists on average are less in-terested than men in opportunities in industry, wewould expect to observe a division of scientificeffort: men will focus more of their research efforton questions that have potential commercial value,while women will not.If female life scientists develop research streams

that are of less interest to commercial enterprise,then the estimated gender gap may not imply thatthe genders face differences in opportunities. Spe-cifically, as long as we assume that, in some casesat least, faculty members are aware ahead of timewhether a research program may have commercialvalue, then the relative degree of commercial ori-entation in a scientists research can be consideredto be a measure of the scientists revealed prefer-ence for commercial-sector opportunities. If, assupply-side theories suggest, the gender gap iscaused by the fact that women are, for any reason,less interested than men in engaging with compa-nies, we should expect to observe:

Hypothesis 5. With the commercial orientationof research choices controlled for, the magni-tude of the estimated male-female gender gapin the likelihood of becoming a SAB memberdeclines.

In particular, we would find support for Hypoth-esis 5 if women faculty were less interested thanmen in roles in industry and if the two genders setsome aspects of their respective research agendas


according to this base difference in preference.When we fail to control for the gap in level ofcommercial interest, it will load onto the parameterestimate for a scientists gender. If this explanationholds, when we explicitly condition our statementon the difference in interestand hence the will-ingness to supply effort for commercial workweshould anticipate a decline in the estimated gen-der gap.

DATA AND METHODS

We have assembled a data archive with careerhistories of approximately 6,000 life scientists toempirically gauge the gender gap in SAB member-ships. As we discuss next, because there are a largenumber of academic life scientists and a relativelysmall number of SAB-joining events, we employeda sampling procedure known as case cohort design.This method was developed by biostatisticians(Prentice, 1986; Self & Prentice, 1988) and is com-monly used in epidemiological research.

Sample and Statistical Estimator

Case cohort design applies well in research con-texts where there are few events in a large popula-tion, rendering it costly to draw a random samplecontaining enough events (or failures in biosta-tistics parlance) for statistical estimation. To sam-ple in this way, a researcher first compiles theevent histories of some or all of the members of apopulation that have experienced the event underexamination. Next, the researcher randomly drawsa comparison sample, known as the subcohort,from the population. The observations in the sub-cohort are then weighted in the estimation to mir-ror the distribution of events and nonevents in thepopulation.To construct our data set, we first collected in-

formation about scientific advisors of all biotech-nology firms that had filed an initial public offering(IPO) prospectus (Form S-1, SB-2, or S-18) with theSEC by 2002. A total of 533 dedicated biotechnol-ogy firms headquartered in the United States hadfiled IPO prospectuses between 1972, when thefirst biotechnology firm went public, and January2002, when we concluded our data collection. For511 of these companies, we were able to retrievefilings from which we obtained biographicalsketches of SAB members. In this analysis, we re-tained only those individuals who held a Ph.D.degree and were in the employ of a US-based uni-

versity or research institution at the time theystarted or joined the biotech company. We identi-fied 720 unique SAB scientists. Their first transi-tions to SAB membership constitute the events weanalyze.The next step was to create a comparison set (the

subcohort) of scientists who were eligible to be-come SAB members. We did this by drawing astratified, random sample of 13,000 doctoral degreeholders listed in the UMI Proquest Digital Disser-tation database, which reports the name, academicdiscipline, and date of all US Ph.D. program grad-uates. The subcohort was constructed so that itsdisciplinary composition and Ph.D. year distribu-tion matched those of the event set (e.g., 15 percentof biotechnology company advisors hold doctoratesin biochemistry, so the random sample contains15 percent Ph.D.s in biochemistry). We stratified onthese two dimensions so that the individuals in thecomparison cohort hailed, in exact proportions,from the specific disciplines responsible for theknowledge base of the life-sciences-relatedindustries.The members of this sample were prospectively

followed from the time they earned a Ph.D. degree.We created publication histories for all scientists inour database by querying the ISI (Thomson Reuters)Web of Science database. We then used the affilia-tions listed on papers to identify each scientistsemployer and, assuming frequent enough publica-tions, to track job changes.Approximately 2,000 of the 13,000-person ran-

dom sample were deleted because they did notappear in the Web of Science in any year afterearning their doctoral degrees. We further deletedthose who (i) published exclusively under corpo-rate affiliations, (ii) had zero publications for a pe-riod of five consecutive years, or (iii) exited aca-demia during the early stage of their career (thosewho stopped publishing within five years after re-ceiving their Ph.D. were likely to have only heldpostdoctoral positions before exiting academia).The final matched sample contains 5,946 scientistsin the randomly drawn subcohort, augmented bythe 720 event cases (SABmembers), yielding a ratioof 8 matched, subcohort sample members to 1 SABevent case. It has been demonstrated that a cohort-to-event ratio of 5 to 1 or higher results in little lossof efficiency in estimations (Breslow, Lubin,Marek, & Langholz, 1983; Self & Prentice, 1988).We structured our data as individual-level career

histories and modeled the rate of first-time transi-tion to SAB membership. Each scientist was con-


sidered to be at risk of being appointed to a SAB inhis/her Ph.D. year or in 1961 (when the first bio-technology company was founded), whichever wasthe later. All individuals known to be in academiabut not yet appointed to an SAB are right-censored(i.e., we stopped observing these individuals) at theend of January 2002 or the (assumed) age of 65. Inthe estimations, we used a modification of Coxs(1972) proportional hazards model that adjusts forthe case cohort sampling design. This estimatorclosely resembles rare events logit (Manski & Ler-man, 1977). More details of the statistical proper-ties of the estimator can be found in Stuart andDing (2006).

Variables

Gender and control variables. We consulted anumber of data sources to create covariates at theindividual, network, and university levels. Alltime-changing variables were updated annuallyand included in the regressions as one-year-laggedvariables.The gender of each scientific advisor and mem-

ber of the matched sample was coded by his or herfirst name. Gender is the primary characteristicchoosers seek to convey in the selection of givennames (Alford, 1988; Lieberson & Bell, 1992). Twoindividuals independently coded the gender of sci-entists in our sample, and intercoder reliability isnearly perfect. We were able to confidently identifygender for 95 percent of the scientists in our data,either from first names or from web searches. Weassumed that all scientists with androgynous firstnames were male. Our results were similar when wedropped scientists with gender-ambiguous names.We include three levels of control variables: in-

dividual, organizational and temporal. First, fol-lowing previous studies of scientists commercialactivities (Azoulay, Ding, & Stuart, 2007; Shane &Khurana, 2003; Stuart & Ding, 2006), at the individ-ual level we include three time-changing measuresof scientists human capital: total publicationcount, total citation count,2 and inventor on pat-

ents. Each of these is annually updated, and thepublication and citation count variables are cumu-lative values for a scientist prior to time t. Thepatent covariate is an indicator variable that turnsfrom 0 to 1 at the point when a scientist is firstlisted as an inventor on a patent (based on thepatent application date) and remains 1 thereafter; itis defined as 0 for the full event histories of scien-tists who never patent. Previous studies haveshown that commercial science is the province ofaccomplished researchers, and we expect that all ofthese measures will increase SAB participation.Also at the individual level, we control for cohorteffects and research-field effects. For the former, weinclude a scientists Ph.D. year. For the latter, weinclude dummy variables for Ph.D. field.3

At the university level, to control for the degreeof institutional support for commercial activities,we constructed a time-changing dummy variable,employer has TTO, coded 1 in each year that theuniversity employing a focal scientist had an activetechnology transfer office (TTO). Information onfounding dates for all university TTOs is availablefrom the Association of University TechnologyManagers surveys. The regressions also include aperiod effect (prior to 1980), coded 1 if a focal yearis before 1980. This cut point was chosen because itwas a watershed year for the development of thebiotech industry, in large measure due to the verysuccessful IPO of Genentech, an industry pioneer.4

2 Because the citation distribution is truncated and wewere only able to gather current-day citation totals foreach paper tallied in 2001 rather than annual counts, itwas necessary for us to impute the time path of citationsto annualize the data. We did so assuming that the arrivalof citations follows an exponential distribution with haz-ard rate (i.e., inverse mean) equal to 0.1. The bibliometricliterature suggests that citations accumulate according to

an exponential distribution (Redner, 1998), and this istrue of the typical paper in our database. We identifiedthe specific parameter, 0.1, by manually coding 50 ran-domly selected papers in each of three publication years,1970, 1980, and 1990, and then choosing the parameterthat yielded the best fit to the actual time path of citationsto these randomly chosen papers.

3 The sample is stratified on Ph.D. year and field, somost of the heterogeneity in graduation cohort and sci-entific field is eliminated by enforcing a match betweenthe SAB subsample to the random sample on these twovariables. However, because individuals subsequentlyleave the sample at different rates, residual, cohort, andscientific field effects on the rate of joining SABs arepossible.

4 The period effect is coarse-grained because we al-ready include Ph.D. year in the regressions. The propor-tional hazards model uses a scientists professional age(the difference between current calendar year and Ph.D.year) as the clock. With an age-based clock and Ph.D.year included in the models, we cannot include a fine-grained period control variable (e.g., continuous calendaryear or calendar year dummies). In unreported robust-


Covariates for demand-side hypotheses. Hy-pothesis 1 proposes that the prestige of a scientistsuniversity employer has a stronger, positive effecton a womans likelihood of joining an SAB than ithas for a man. We collected Gourman Report rank-ings of biochemistry departments for all employinginstitutions in the data set, as this discipline hasspawned the greatest number of commercial lifescientists (and hence it is the modal discipline inthe data). To capture the prestige of a scientistsemployer, we collapsed the scale and measuredemployer prestige as a dummy variable, coded 1 ifa university was among the top 20.5

Hypothesis 2 states that documented scientificand management capabilities has a greater effect onSAB participation for female scientists. Our fieldinterviews particularly highlighted the impact ofprominent administrative posts. In the archivaldata, however, we were not able to gather time-changing measures of administrative positions,such as deanships. The nearest proxy we couldproduce for the full span of the data is the (time-changing) proportion of a focal scientists papersfor which he or she was the last author, or percent-age of last-authored publications. By convention inthe life sciences, the principal investigator andhead of a research group is the last author on paperspublished by the group. Although last authors in-tellectual contributions to joint research often areless than those of first authors, project leaders raisethe resources that are expended in collaborativeendeavors (Kempers, 2002; Shapiro, Wenger, &Shapiro, 1994). In fact, productive labs are small

enterprises unto themselves: each is a hierarchystaffed by technicians, graduate students, postdoc-toral fellows, and at the helm, a principal investi-gator. Thus, scientists with many last-authored pa-pers will possess credible management experience.We expect these scientists will elicit more SABopportunities, and that having a high proportion oflast authored publications will have a particularlystrong effect on the transition rate for womenscientists.Hypothesis 3 posits a differential effect of a sci-

entists network ties on mens and womens SABparticipation. A scientists professional networkcan include ties to coauthors, thesis advisors, stu-dents, mentors, and others. We follow prior re-search on social influence among academic scien-tists (Azoulay, Zivin, & Wang, 2010; Moody, 2004)by measuring scientists coauthorship networks.Assuming that coauthorships are strong ties, thesegauge the potential for endorsements for commer-cial opportunities within each scientists network.Although the coauthorship network admittedly isan incomplete representation of a scientists port-folio of connections, it does capture an importantcomponent of his/her strong tie network, and it alsooffers the primary benefits of being traceable back-ward in time and available for the full populationof academic scientists.We create two measures from the coauthorship

network. The first is an authors degree score, or thecount of coauthors. Individuals with higher degreescores are more likely to have direct contacts thatcan match them to advisory opportunities. Second,we count the number of academic entrepreneursindividuals who have previously (prior to a givenyear) made the transition to found or advise a bio-technology firmwith whom a focal scientist hasone or more coauthored publications. We label thiscovariate count of coauthorship ties to academicentrepreneurs and assume that strong connectionsto scientists who have already entered the commer-cial sphere will abet the transition of a focal scien-tist. Hypothesis 3 states that, relative to men, awoman faculty members rate of joining a SAB willbe more sensitive to the presence of a direct tienetwork that is conducive to commercialopportunities.Covariates for supply-side hypotheses. Hypoth-

esis 4 asserts that if supply-side explanations haveempirical leverage in explaining SAB member-ships, we are likely to observe that womens age oftransition to first SAB membership is higher thanthat of men. We tested this hypothesis by calculat-

ness tests, we ran the regressions with calendar yearinstead of Ph.D. year as a control in the models. Thisspecification produces results that are almost identical tothe findings reported here.

5 In unreported regressions, we performed robustnesstests with different cut points. The variations included:(i) using overall graduate school rankings instead of bio-chemistry department rankings; (ii) using top 10, 15, 25,30, or 50 as the cutoff for high prestige; (iii) using cate-gories such as 120, 2150, and 51 and below or otherforms of category groupings; and (iv) using continuousranking for the top 50 employers and a dummy indicat-ing an employer was below 50 (Gourman only ranks thetop 50 institutions). As one would expect, the magnitudeof the prestige effect in boosting the probability of anSAB appointment decreases as the employer prestigedummy becomes less exclusive (i.e., top 25, 30, or 50versus top 20). However, interaction effects with thefemale variable remain largely unchanged from those wereport.


ing and comparing the distributions of the timeuntil transition to SAB membership for eachgender.Next, Hypothesis 5 posits that in regressions that

account for the commercial focus of a scientistsresearch, the magnitude of the estimated gendergap will decline. To measure the commercial ori-entation of a research agenda, we utilized the pub-lication data to generate a fine-grained gauge of thelatent research commercializability of each scien-tists research program. Specifically, we used thewords in the titles of scientists publications toidentify the areas in which they had conductedresearch6 and then applied weights to these areas,using an (endogenous-to-the-sample) measure ofthe extent to which other scientists working inthese same areas previously have patented theirscientific discoveries. Intuitively, we used the pub-lications of scientists who have already applied forpatent rights to define the benchmark for commer-cializable research and then compared the researchof each scientist in our data set to this benchmark togenerate a scientist-specific research commercializ-ability score for each scientist-year. Formally, theresearch commercializability score for scientist i inyear t is defined as:

Research commercializablityit j 1

J

wj, t 1i

nijtknikt

,

where j 1 . . . J indexes each of the scientifickeywords appearing in the titles of the journal ar-

ticles published by scientist i in year t. The numer-ator, nijt, is the number of times each of the key-words j has appeared in scientist is articlespublished in year t, and the denominator sums upthe frequency with which all keywords have ap-peared in scientist is articles published in thatyear. The term is a weight for each keyword thatmeasures the frequency with which word j is usedin the titles of articles published by scientists whohave patented in year t or earlier, relative to its usein the articles of scientists who have not patentedbefore year t. This weight measures the degree ofpatentability of the research keyword j. More de-tails of this measure are described in Azoulay,Ding, and Stuart (2009).7

RESULTS

Descriptive Statistics

Notably, only 49 women are listed as scientificadvisors, representing just 6.8 percent of the totalnumber of academic scientists in this role. In com-parison, almost 20 percent of the members of thematched sample are female. A log-rank test estab-lishes that the survivor functions for men andwomen are unequal (p .00001).Table 1 describes the gender composition in the

random, matched subcohort and in the SAB sam-ple, broken out by Ph.D. cohort. Consonant with

6 We relied on title words in journal articles instead ofjournal- or author-assigned keywords, because the Webof Science database did not begin to include keyworddescriptors until 1992. However, the titles of biomedicalresearch papers typically indicate the research area andthe methodology used in the paper. We find high overlapbetween title words and keywords in the papers forwhich both are available.

7 A potential alternative to research commercializabil-ity is to use patents themselves as proxy for commercialinterest. We chose to use the commercializability scorerather than a patent-based variable because we believethe former better captures a scientists choice to investi-gate research topics of commercial relevance. In general,scientists choose the research questions they explore. Bycontrast, patenting is affected by factors beyond the in-dividual scientists control, including the view of his/heruniversitys TTO on whether a particular scientific dis-covery merits the costs of patent protection.

TABLE 1Number of Ph.D.s Granted in the Life Sciences by Gender and Cohort

Ph.D. Cohort

Random Matched Sample SAB Samplea

Female Male Female Male

194160 54 (11.7%) 408 (88.3%) 2 (2.1%) 95 (97.9%)196170 169 (13.6%) 1,070 (86.4%) 10 (4.5%) 210 (95.5%)197180 388 (18.0%) 1,764 (82.0%) 30 (10.2%) 264 (89.8%)198195 419 (30.5%) 954 (69.5%) 7 (6.4%) 102 (93.6%)

a The SAB sample is a sample of scientific advisory board members.


published statistics (CPST, 1996; NSF, 1996), theproportion of Ph.D. degrees earned by women inthe random sample increases significantly overtime, from 12 percent in the cohort of Ph.D.s before1960 to 30 percent in the post-1980 cohort. In all thescientist cohorts, however, there are none in whichwomen exceed 10.2 percent of the SAB sample.8

Table 2 reports means for the human and socialcapital variables at five different cross-sections ofscientists tenure, broken out by gender and againincluding only members of the random sample.The gender gap in performance metrics in Table 2would increase substantially if we presented thesestatistics for the overall data set, instead of just forthe random subcohort. As we demonstrate shortly,outstanding professional achievement is highlypredictive of the transition to commercial science,

and men comprise the vast majority of the SABsubsample. Including SAB members would there-fore drive up the mean values of all measures ofcareer success for men.

Overall Gender Gap

Multivariate regression results are presented inTable 3. Model 1 includes Ph.D. field dummy con-trols, time period (prior to 1980), Ph.D. year, and adummy for employer TTO. In addition, the modelsimplicitly control for a scientists professional age(current year minus Ph.D. degree year). In this base-line model, gender has a strong, negative effect onthe hazard of joining an SAB. Womens rate ofjoining an SAB is estimated to be 45 percent that ofmens. Model 2 adds four covariates characterizinga scientists research and patenting activity: totalpublication count, total citation count, percentageof last-authored publications, and inventor on pat-ents. Not surprisingly, all four are strong, positivepredictors of the likelihood of joining an SAB. Inkeeping with the findings of past studies (e.g.,Zucker, Darby, & Brewer, 1998), the emergent pic-ture is that commercial science is concentratedamong the scientific elite.Model 3 adds two network variables: count of

coauthors and count of coauthorship ties to aca-demic entrepreneurs, the latter of which specifi-cally gauges the extent of a scientists connectionsto commercially active peers. Scientists who have agreater number of coauthors are substantially morelikely to become SAB members. In addition, indi-viduals who have coauthored with an academic

8 Table 1 suggests that women made inroads into SABmembership between the first, second, and third cohortsin the data, and then their participation dropped off inthe final (198195) cohort. However, the apparent de-cline in female participation in the table may be mislead-ing. Because women entered the sample later (on aver-age), given their gradual progress in obtaining facultyappointments, the average woman in the 198195 cohortwas younger than the average male scientist in that co-hort. Because the peak of the transition rate is at profes-sional ages of 15 years and beyond and our data concludein 2002, the age difference between women and men mayaccount for the apparent decline in female participationin SABs in the last cohort. In unreported regressions, weestimate the interaction effect of a scientists gender andher Ph.D. year and do not find evidence for a lower rateof transition for more recent cohorts.

TABLE 2Mean Values of Human and Social Capital Covariates at Five Professional Tenure Cross-Sections, by Gendera

Variables

5th Year 10th Year 15th Year 20th Year 25th Year

Male Female Male Female Male Female Male Female Male Female

Total publication count 4.69 3.82 12.97 10.72 23.45 20.21 34.63 33.31 48.49 45.38Citation count per paper 8.51 8.31 14.72 14.85 18.74 19.56 21.17 21.12 22.48 22.22Percentage of last-authoredpublication

0.12 0.10 0.22 0.17 0.29 0.22 0.34 0.27 0.38 0.30

Count of coauthors 14.66 12.66 22.68 20.79 30.45 28.22 38.62 36.18 45.10 43.58Count of coauthorship ties toacademic entrepreneurs

0.05 0.04 0.10 0.08 0.17 0.15 0.30 0.21 0.32 0.27

Patent count 0.03 0.02 0.08 0.05 0.21 0.07 0.38 0.11 0.48 0.16Inventor on patentb 0.02 0.01 0.05 0.03 0.09 0.04 0.13 0.08 0.16 0.09Research commercializability 0.58 0.48 0.55 0.48 0.52 0.48 0.48 0.44 0.46 0.46n 4,153 1,021 3,156 653 2,498 456 1,949 320 1,346 199

a Values are for scientists in our random matched cohort. Year denotes time elapsed since the year of a scientists Ph.D.b 1 yes.


entrepreneur transition at a rate about 1.31 times( exp[0.27]) higher than those who lack connec-tions to academic entrepreneurs. Model 4 adds themain effect of employer prestige, which performsas expected. Holding a position at a university witha top-20 biochemistry department accelerates therate of joining an SAB by a factor of 2.32( exp[0.84]).We now turn to the effect of gender. First, recall

that the descriptive statistics in Table 2 show that

women scientists have fewer patents, papers, last-authored papers, citations, and coauthors at eachcareer stage than do men. Even after controlling forthese variables and characteristics of scientists em-ployers, we find a large gender difference in thehazard: estimates of the effect of the gender dummyvariable ranges between 0.80 in the uncondi-tional results in model 1 to 0.49 in model 4,which controls for research and patenting activity,network ties, and employer characteristics. This

TABLE 3Case-Cohort-Adjusted Cox Regression Models of Transition to SABa

Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10

Ph.D. field fixed effectincluded

Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Year is prior to 1980 4.46 4.80 5.09 5.09 12.79 12.98 12.62 12.61 12.50 12.28(0.53)** (0.55)** (0.56)** (0.55)** (3.48)** (3.48)** (3.43)** (3.44)** (3.44)** (3.39)**

Ph.D. year 0.03 0.06 0.07 0.07 0.03 0.04 0.03 0.03 0.03 0.01(0.01)** (0.01)** (0.01)** (0.01)** (0.01)* (0.01)* (0.01)* (0.01)* (0.01)* (0.01)

Employer has TTO 0.64 0.55 0.53 0.31 0.25 0.23 0.26 0.26 0.25 0.61(0.11)** (0.11)** (0.11)** (0.12)** (0.13) (0.13) (0.13)* (0.13)* (0.13) (0.11)**

Female 0.80 0.59 0.51 0.49 0.35 1.16 0.77 0.77 0.60 0.85(0.18)** (0.20)** (0.20)* (0.19)* (0.26) (0.39)** (0.23)** (0.25)** (0.22)** (0.19)**

Total publication count 0.01 0.001 0.002 0.001 0.001 0.001 0.001 0.001(0.001)** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)

Total citation count 0.02 0.02 0.01 0.01 0.01 0.01 0.01 0.01(0.002)** (0.002)** (0.002)** (0.002)** (0.002)** (0.002)** (0.002)** (0.002)**

Inventor on patents 1.26 1.25 1.24 1.30 1.30 1.29 1.29 1.30(0.14)** (0.14)** (0.14)** (0.14)** (0.14)** (0.14)** (0.14)** (0.14)**

Percentage of last-authoredpublication

2.42 2.59 2.61 2.55 2.43 2.53 2.55 2.56(0.24)** (0.23)** (0.23)** (0.24)** (0.25)** (0.24)** (0.24)** (0.24)**

Count of coauthors 0.004 0.004 0.004 0.004 0.004 0.004 0.004(0.001)** (0.001)** (0.001)** (0.001)** (0.001)** (0.001)** (0.001)**

Count of coauthorship tiesto academic entrepreneurs

0.27 0.27 0.25 0.25 0.25 0.24 0.25(0.06)** (0.05)** (0.05)** (0.05)** (0.05)** (0.05)** (0.05)**

Employer prestigea 0.84 0.88 0.84 0.84 0.83 0.83(0.12)** (0.13)** (0.13)** (0.13)** (0.13)** (0.13)**

Researchcommercializability

1.34 1.35 1.33 1.33 1.32 1.27(0.19)** (0.19)** (0.18)** (0.18)** (0.18)** (0.20)**

Female employer prestige 0.60(0.46)

Female percentage of last-authored publication

1.52(0.73)*

Female count ofcoauthors

0.003(0.001)**

Female count ofcoauthorship ties toacademic entrepreneurs

0.29(0.16)

Log-likelihood 8,673.48,223.5 8,093.4 8,046.3 7,885.5 7,883.7 7,881.0 7,884.5 7,887.4 8,513.72 280.70 553.62 634.81 673.23 696.52 694.29 968.20 700.02 694.17 200.40Model df 12 16 18 19 21 21 21 21 20 13

a Time at risk 110,461 scientist-years; n (individuals) 5,946; n (events) 720. Robust standard errors in parentheses. SAB isscientific advisory board; TTO is technology transfer office.

b Employer prestige equals 1 if the scientist is in the top 20 of the department. p .10* p .05

** p .01


translates to a per unit time hazard rate for malescientists that ranges between 1.63 ( 1/exp[0.49]) and 2.23 ( 1/exp[0.80]) times thetransition rate for otherwise comparable women.

Tests of Demand-Side Hypotheses

Models 58 estimate the effect of hypothesizedmoderators of the gender gap according to demand-side theories. We test our first three hypotheses byincluding interaction terms between the femaledummy variable and four covariates: employerprestige (Hypothesis 1), percentage of last-authoredpublications (Hypothesis 2), count of coauthors(Hypothesis 3), and count of coauthorship ties toacademic entrepreneurs (Hypothesis 3).Hypothesis 1 proposes that a prestigious univer-

sity affiliation partially offsets gender-based biasesin inferences about competence, and thus it has agreater effect on female than on male scientistslikelihood of becoming SAB members. We testedthis hypothesis in model 5. The results fail to con-firm Hypothesis 1: the interaction between femalegender and employer prestige (top-20 department)is not statistically significant. This null result per-sists in unreported robustness tests with alternativespecifications of the university prestige covariate.We had expected to find that a high-status uni-

versity affiliation conveys legitimacy to scientistswishing to participate in the commercial sector andthat this certification is more important for women.The finding suggests that to the extent that anysuch certification process is at work, forces operat-ing in the reverse direction offset it. One potentialfactor suggested by our interviewees is that a pro-cess of cumulative disadvantage for women mayhave developed alongside the gain in momentum ofcommercial science in academic circles (cf. Cole &Zuckerman, 1984). Although the members of a coregroup of male faculty have become central actors inthe network of commercial science, women haveremained on the periphery of this social structure.Because male-dominated commercial networkshave been centered at elite universities, men atthese institutions may have enjoyed access to manyopportunities distributed throughout this network.In other words, Hypothesis 1 may be rejected, be-cause any legitimacy-based benefit that accrues tofemale faculty at high-status universities is metwith an equal benefit that men obtain from havingprestigious employers, which is based on their fa-vorable access to the networks of commercial sci-

ence that are now anchored by senior faculty atelite universities.Hypothesis 2 posits that evidence of scientific or

management capabilities is more positively associ-ated with the likelihood that women will join SABsthan it is for men. We tested this hypothesis inmodel 6 and found support for it: there is a posi-tive, statistically significant interaction effect be-tween female gender and percentage of last-au-thored publications. This evidence is consistentwith role incongruity and lack-of-fit argumentsabout gender biases. If it is the case that gender-based stereotyping causes women to be held to ahigher standard than men for recruitment to posi-tions for which women are atypical (Lyness &Heilman, 2006), we would expect to observeaswe dothat a track record of scientific achieve-ment and managerial experience will be more im-portant for women than for men.We also hypothesized that a broad direct tie net-

work may help women more than men in counter-ing in-group favoritism. Models 7 and 8 in Table 3test the differential effect of network ties on mensand womens rates of joining SABs. First is aninteraction in model 7 between the female dummyand our primary proxy for a scientists direct tieprofessional networks, the cumulative count of co-authors the scientist has accrued. The positive, sig-nificant coefficient supports Hypothesis 3: thenumber of direct professional network ties has astronger effect on the likelihood of joining SABs forwomen than for men. The next finding of note isthe large, though weakly (p 0.10) significant in-teraction effect in model 8 between femaleness andhaving previously coauthored papers with an aca-demic entrepreneur (count of coauthorship ties toacademic entrepreneurs). Having one more coau-thor with a scientist who has previously founded oradvised a company increases a male scientists like-lihood of joining an SAB by 27 percent( exp[0.24]). In comparison, a coauthor who is anacademic entrepreneur elevates a female scientistslikelihood by 70 percent ( exp[0.24 0.29]).Overall, these results support Hypothesis 3: beingin a direct tie network conducive to generatingreferrals to commercial science is particularly im-portant for creating opportunities for womenfaculty.

Tests of Supply-Side Hypotheses

Turning our attention to supply-side perspec-tives on the gender gap, we consider whether the


evidence suggests that part of the gender gap maybe explained by the fact that women choose to beinfrequent participants in commercial science, ver-sus the fact that they are interested in involvementbut are excluded from it by limited opportunities.First, issues of work-family balance may leadwomen to choose to avoid nonessential work dur-ing the years in which their family commitmentsare largest. Second, we consider the possibility thatany factor leads women to choose, on average, to doresearch of less commercial relevance than do men.Hypothesis 4 posits that if women choose to al-

locate more time to their families, we should expectthem to delay participation in commercial oppor-tunities until a later stage of life, when less time iscommitted to child care. If this occurs, we willobserve that among male and female scientists whohave made the transition, there will be differentdistributions of the age at first transition to a SAB.The typical scientist in our sample does not en-

gage in commercial science until relatively late inhis or her career. For scientists who do join SABs,Table 4 presents the distribution, by gender, of theprofessional age group at which individuals jointheir first SAB. Among the 49 female SAB mem-bers, 42 transitioned 11 or more years after theyobtained their Ph.D. with the hazard peaking inapproximately the 20th year after the Ph.D., thesame peak year for the male SAB scientists in oursample. Assuming that life scientists obtain theirdoctoral degrees at an average age of 31 (Jacobs &Winslow, 2004), this suggests that the times ofhighest risk are between 46 and 56 years of age. Ourdata further indicate that transition times are indis-tinguishable by genderthe mean transition age formale and female SAB scientists is 19.2 and 19.0,respectively. In addition, a two-sample Kolmogo-

rov-Smirnov test for equality of the distribution ofage at time of first SAB does not permit rejection ofthe null hypothesis that male and female scientistsjoin SABs at similar career stages (D 0.12, p 0.54). We take this as suggestive evidence againstthe argument that women lack interest in pursuingSAB opportunities due to family-work balanceconsideration.While the between-gender similarity of the age

distributions at the time of first transition is sug-gestive, it is not conclusive. In particular, if thefemale sample is split between women who are andwho are not interested in SABs and if this split istilted more toward the no interest group amongwomen than among men, it is possible that the ageof transition will be identical, but that, on average,women still are less interested in commercial sci-ence. To further investigate whether the gender gapin SAB membership is a matter of choice, we moveto test Hypothesis 5, which posits that the esti-mated gender gap will decline when the regres-sions account for gender differences in choicesabout the commercial orientation of research agen-das. To address this hypothesis, we introduce theresearch commercializability covariate. In Table 2we have presented comparison of mean values ofthis covariate between men and women over crosssections of professional tenure. There is a moder-ate-in-magnitude but statistically significant differ-ence in mean research commercialization betweenthe genders.9 The gap persists until the 20th year ofprofessional tenure, though it narrows and disap-pears altogether for senior faculty.We test Hypothesis 5 in regression models 9 and

10 in Table 3. First, research commercializabilityhas the expected, positive effect on the likelihoodof joining an SAB. An increase of one standarddeviation in the covariate ( 0.69) is associatedwith a 2.49 times ( exp[1.32 0.69]) increase inthe likelihood of joining an SAB. This level ofsignificance and magnitude indicates that the coef-ficient works as we intended to capture differencesbetween scientists in choices about whether to fo-cus their research on topics that are relevant tocommerce. Model 9 also shows that when researchcommercializability is included in the regression,the magnitude of the gender gap does not decline.In fact, the estimated male-to-female ratio of thehazard of joining an SAB actually increases from

9 Results of t-tests of the mean comparisons are notreported in Table 2 but are available upon request.

TABLE 4Distribution of First SAB Transitiona

Years since Ph.D. Male Female

15 23 1610 86 6

1115 126 91620 155 132125 132 112530 74 73135 51 23540 24 0

a A two-sample Kolmogorov-Smirnov test for equality of thetenure distribution across gender indicates statisticalequivalence.


1.63 ( 1/exp[0.49]) in model 4 to 1.82 ( 1/exp[0.60]) in model 9, when we include researchcommercializability. In model 10, we reestimateour regression excluding all the individual, net-work, and university prestige covariates. Compar-ing the magnitude of the coefficients on femalebetween models 1 and 10, we find that the esti-mated gender gap modestly increases from a male-to-female ratio of 2.23 ( 1/exp[0.80]) to 2.34( 1/exp[0.85]). Therefore, our data do not sup-port Hypothesis 5: there is no evidence that thegender gap declines in regressions that account forgender differences in the revealed preference forcommercial sector work.

DISCUSSION AND CONCLUSION

Our research documents a gender gap and theninvestigates mechanisms that may account for it inuniversity scientists participation in private-sectorscientific advisory boards. The mechanisms we ex-amine are informed by demand-side theories em-phasizing biases that create unequal opportunitystructures for men and women, and supply-sidetheories that attribute different outcomes to hetero-geneities in human capital or career preferencesbetween the genders. Our goal has been to providean integrated test of both families of explanationsin a context in which we can identify the candidatepool for senior-level corporate positions and inwhich we can construct detailed, longitudinal mea-sures of the qualifications of the individuals in thesample.Our analyses show that women scientists are

much less likely than men to join the advisoryboards of for-profit biotechnology companies. Withcontrols for scientists professional accomplish-ments, social networks, employer characteristics,and proxies for their interest in commercial sci-ence, we find that male scientists are almost twiceas likely to join SABs. This gender gap is roughlycomparable to that of the rate of patenting amongscientists found in previous studies (Ding et al.,2006; Whittington & Smith-Doerr, 2005). Thoughthis gap is large and in need of further investiga-tion, our findings, in one way, offer hope. Whitting-ton and Smith-Doerr (2005) hypothesized that ascommercial activities become more demanding ofscientists time and identity, female participationwill decline (i.e., in the progression from patentingto advising to founding companies, we should an-ticipate a winnowing in female participation ateach juncture). We find that the predicted gender

ratio in SAB participation is roughly comparable toestimates of the gap in rates of patenting amongacademic scientists, which is notable because of apivotal distinction between the two activities: thedecision to patent (conditional on the filing of aninvention disclosure) is solely at the discretion ofthe universitys technology transfer office, whereasSAB membership depends on an invitation from acompany. Despite this, the evidence suggests nofemale attrition in the progression from patentingto SAB membership.What general lines of theory account for the large

gender gap we document? We positioned this arti-cle as an effort to compare demand-side theoriesthat posit biases in the workplace against supply-side theories that locate causality in individualchoice. We believe that the preponderance of ourevidence supports demand-side theories. First, thefinding that past leadership positions have a muchstronger effect for women than for men possiblyindicates perceived incongruities between being fe-male and leadership roles in commercial science.Likewise, the result that having many coauthorshas a stronger effect on the transition rate forwomen faculty suggests that well-structured, strongtie networks help women to overcome traditionalout-group biases. Second and relatedly, our archi-val analyses show that accounting for the revealedpreference of scientists to work on commerciallyrelevant research does not mitigate the estimate ofthe gender gap. These results suggest that the con-ditions under which the gender gap arises are morecompatible with a constraint-based explanation, al-beit one that may be tempered by some differencesin interest in commercial science on the part offemale faculty.What do our findings suggest for policy makers?

First, a salient message is the lack of evidence tosupport interest-related accounts of the gender gap.Second, it is highly doubtful that the gap can beexplained by the fact that male and female scien-tists sort into different research areas that are ofvarying relevance to commercial firms. Therefore,we believe that policy interventions are most likelyto succeed if they are directed toward the condi-tions that affect perceptions of womens qualifica-tions to hold senior corporate roles, especially inthe eyes of the decision makers who allocate suchopportunities.In the short run, our findings point to some spe-

cific areas in which university administrators mayhave some leverage to remedy the gender gap. Ev-ery woman we interviewed who had held a senior


administrative role believed that the visibility ofthe office led to consulting and SAB opportunitiesand bolstered her legitimacy in the commercial sec-tor. The quantitative evidence also accords withthis perspective. Therefore, for women faculty whoare willing to take on senior administrative assign-ments such as deanships, we believe that activeuniversity policies to match women to these posi-tions will help to create opportunities for women incommercial science.In addition, most research universities have

TTOs. When they function well, these offices arereservoirs of relationships between members of theuniversity community and industry. Our findingssuggest that direct contact networks matter morefor women than for men in generating opportuni-ties in commercial science; men often attract offersbased on generalized reputations, while womenstill may depend on referrals from close ties tomatch to commercial-sector opportunities. Activepolicies at TTOs to promote the research of womenfaculty and to broker connections between themand influential members of the entrepreneurshipand investment communities will help women gainentry to the set of heretofore male-dominated net-works in science- and technology-based industries.We also must note a number of limitations of this

research. First, in testing demand-side explana-tions of the allocation of SAB opportunities, ourapproach aims to assess biases in the allocationprocess by examining the conditions under whichthe degree of discrimination varies (cf. Petersen &Saporta, 2004). This research design has been fre-quently used in archival studies of workplace in-equities (e.g., Lyness & Heilman, 2006). However,econometric identification of hypothesized effectsbased on theoretically implied contingencies is nat-urally less definitive than directly observing (in ourcase) the perceptions of SAB selectors. Should di-rect measures of evaluation records ever becomeavailable (e.g., ratings of applicants, as in Petersen,Saporta, and Seidel [2005]), researchers may ac-quire better empirical leverage by which to adjudi-cate between demand- and supply-side theoreticalexplanations.Second, in our tests of supply-side theories, we

examine the timing of joining SABs and facultymembers tastes for commercial science as re-vealed by their bibliographic records. There are,however, many other supply-side factors thatmay influence the gender gap in SAB appoint-ments but which we were unable to explore inour archival, historical data. For example, our

test of the distribution of SAB events across sci-entists career cycles reveals no evidence thatwomen faculty pursue commercial interest at alater career stage than do men. This empiricalresult rests on the logic that family obligationsmay restrict womens interest in commercial en-gagements during certain periods of their profes-sional lives. In reality, it is likely that somewomen are deterred from entering commercialscience altogether, given the oftentimes dauntingtask of balancing the time demands of university-related work and family commitments. Our em-pirical approach cannot rule out this possibility,since it merely compares the transition timesamong scientists who chose to become SAB mem-bers. Once again, a direct measure of scientistswillingness to supply effort may provide a moredefinitive empirical test, though an analyst mustbe sensitive to the fact that women may expressless interest in commercial work simply becausethey (correctly or incorrectly) perceive that thereare biases in the selection process that lead themto perceive a diminished chance of securing po-sitions on SABs.Indeed, whether a faculty member seeks to join

an SAB will depend on many factors, ranging fromhis or her research tastes, desire for remuneration,career concerns, perception of the opportunitystructure, and so on. The proxy we constructedresearch commercializabilitycaptures the prox-imity of a scientists research program to the corpusof commercial knowledge. This covariate tells usthat in terms of the production of scientific knowl-edge, men and women select research topics thatare nearly equivalent, at least with regard to thelatent commercial potential of their ideas. Theproxy, however, does not incorporate other factorsin the labor supply pipeline, such as scientistswillingness to promote the commercial value oftheir discoveries to industry. For this reason, re-search commercializability, too, is an imperfectproxy for scientists desire to pursue SAB opportu-nities. In short, we suggest caution in interpretingthe finding that there is a lack of support for sup-ply-side explanations of the SAB gender gap inthese data.Future research should strive to construct sup-

ply-side indicators that combine objective data onprofessional histories with information that revealsthe career aspirations of members of the candidatepool. For instance, the Scientists and EngineersStatistical Data System (SESTAT), a longitudinalsurvey administered by the NSF, collects informa-


tion about respondents education, work experi-ence, and employer characteristics. In addition, re-spondents to the SESTAT are surveyed about theirjob satisfaction and reasons for job changes. If theseeducational and career histories were supple-mented with self-reported motivational informa-tion, they could enable cleaner estimation of gen-der differences in candidates interest in supplyingeffort to commercial ventures. Of course, the chal-lenge remains that individuals interest in pursuingpositions is shaped by their perception of the op-portunity structure. Specifically, if women scien-tists feel that their odds of securing a position arelow, they may be deterred from expressions of in-terest. Therefore, surveys must be designed to ac-count for these kind of subtleties, and in an idealscenario, candidates interest in supplying effortcan be appropriately matched with information onhow evaluators assess them to fully adjudicate be-tween demand- and supply-side explanations ofthe gender gap in career outcomes.

REFERENCES

Alford, R. D. 1988. Naming and identity. New Haven,CT: HRAF Press.

Association of University Technology Managers. 19912001. A survey summary of technology licensing(and related) performance for U.S. and Canadianacademic and nonprofit institutions, and patentmanagement firm, Northbrook, IL: AUTM.

Audretsch, D. B., & Stephan, P. E. 1996. Company-scien-tist locational links: The case of biotechnology.American Economic Review, 86: 641652.

Azoulay, P., Ding, W. W., & Stuart, T. E. 2007. Thedeterminants of faculty patenting behavior: Demo-graphics or opportunities? Journal of Economic Be-havior and Organization, 63: 599623.

Azoulay, P., Ding, W.W., & Stuart, T. E. 2009. The impactof academic patenting on the rate, quality and direc-tion of (public) research output. Journal of Indus-trial Economics, 57: 637676.

Azoulay, P., Zivin, J. S. G., & Wang, J. L. 2010. Superstarextinction. Quarterly Journal of Economics, 125:549589.

Baron, J. N., Hannan, M. T., Hsu, G., & Kocak, O. 2005.Gender and the organization-building process inyoung, high-tech firms. In M. F. Guilln, R. Collins,P. England, & M. Meyer (Eds.), The new economicsociology: Developments in an emerging field:245273. New York: Sage.

Becker, G. S. 1985. Human-capital, effort, and the gender

division of labor. Journal of Labor Economics, 3:S33S58.

Beckman, C. M., & Phillips, D. J. 2005. Interorganization-al determinants of promotion: Client leadership andthe attainment of women attorneys. American Soci-ological Review, 70: 678701.

Berger, J., Cohen, B. P., & Zelditch, M., Jr. 1972. Statuscharacteristics and social interaction. American So-ciological Review, 37: 241255.

Berger, J., Cohen, B. P., & Zelditch, M., Jr. 1972.

Berryman, S. E. 1983. Who will do science? New York:Rockefeller Foundation.

Bertrand, M., Goldin, C., & Katz, L. E. 2010. Dynamics ofthe gender gap for young professionals in the finan-cial and corporate sectors. American EconomicJournal: Applied Economics, 2: 228255.

Bertrand, M., & Hallock, K. F. 2001. The gender gap intop corporate jobs. Industrial and Labor RelationsReview, 55: 321.

Bombarderi, M. 2005. Summers remarks on womendraw fire. Boston Globe, January 17.

Breslow, N. E., Lubin, J. H., Marek, P., & Langholz, B.1983. Multiplicative

from bench to board

Documents