Does the Gender Composition of Scientific Committees Matter?∗
Manuel Bagues † Mauro Sylos-Labini ‡ Natalia Zinovyeva § October 1, 2016
Abstract
We analyze how a larger presence of female evaluators affects committee decision-making using information on 100,000 applications to associate and full professorships in Italy and Spain. These applications were assessed by 8,000 randomly selected evaluators. A larger number of women in evaluation committees does not increase either the quantity or the quality of female candidates who qualify. Information from individual voting reports suggests that female evaluators are not significantly more favorable towards female candidates. At the same time, male evaluators become less favorable towards female candidates as soon as a female evaluator joins the committee.
Keywords: scientific committees, gender discrimination, randomized natural experiment.
JEL Classification: J71, J16.
∗This paper combines information that was previously reported in two separate working pa- pers: Bagues, Sylos-Labini and Zinovyeva (2014) and Zinovyeva and Bagues (2011). We would like to thank Olympia Bover, Irma Clots-Figueras, Sara de la Rica, Gemma Derrick, Juan Jose Dolado, David Dorn, Silvia Dorn, Berta Esteve-Volart, Luis Garicano, Marco Giarratana, Elena Mart ́ınez, Nic Morgan, Javier Ruiz-Castillo and participants in numerous presentations for their useful comments. We also acknowledge the financial support of the Social Sciences and Humanities Research Council of Canada, Instituto de la Mujer (research grant 36/12) and the Spanish Min- istry of Science and Technology (research grants ECO2008-06395-C05-05 and ECO2008-01116). All remaining errors are our own.
†Aalto University, Helsinki, Finland; email: manuel.bagues@aalto.fi ‡University of Pisa, Italy; email: mauro.syloslabini@unipi.it
§Aalto University, Helsinki, Finland; email: natalia.zinovyeva@aalto.fi
1
1 Introduction
The underrepresentation of women in academia remains a cause for concern among universities and policy makers around the world. In Europe, women account for 47% of PhD graduates, 37% of associate professors and only a mere 21% of full professors (European Commission 2016). Similar patterns may be observed in the US and the gender imbalance is even larger in Japan (National Research Council 2009, Abe 2012).
Several explanations may account for the lack of women in high-level positions. According to the pipeline theory, once women have entered the lower rungs of the academic career it is mainly a matter of time that they would move their way through a metaphorical pipeline to reach high-level jobs. However, in most disci- plines, the share of women among faculty members remains low even after decades of improved recruitment of women at the undergraduate and the doctoral level (Ginther and Kahn 2004, 2009). Gender differences in promotion rates might also reflect differences in productivity, perhaps due to the existence of gendered roles at the household level or the lack of female mentors and role models (Blau, Currie, Cro- son and Ginther 2010). Some women may also devote excessive time to tasks that are socially desirable but which are not taken into account in promotion decisions (Vesterlund, Babcock and Weingart 2014). Furthermore, some authors have pointed out that women are less likely to apply for promotions (Bosquet, Combes and Garcia- Pen ̃alosa 2013; De Paola, Ponzo and Scoppa 2015), perhaps due to the existence of gender differences in the preference for competitive environments (Niederle and Vesterlund 2007; Buser, Niederle and Oosterbeek 2014) or in bargaining abilities in the labour market (Babcock, Gelfand, Small and Stayn 2006; Blackaby, Booth and Frank 2005).
Beyond these supply-side explanations, the slow progress made by women has been sometimes attributed to the lack of female evaluators in the committees which decide on hiring and promotions. In this paper we examine whether the presence of women in scientific committees might help to increase the chances of success of female candidates and to improve the quality of the evaluations. There are several reasons for considering this hypothesis. First, there is evidence of gender segre- gation across different scientific subfields (Dolado, Felgueroso and Almunia 2012, Hale and Regev 2014). If men and women tend to do research in different sub- fields and evaluators overrate the importance of their own types of research, the lack of female evaluators might be detrimental for female candidates (Bagues and Perez-Villadoniga 2012, 2013). Second, research networks tend to be gendered (Bos- chini and Sjo ̈gren 2007, Hilmer and Hilmer 2007).1 If evaluators are mostly male,
1Boschini and Sj ̈ogren (2007) show that coauthoring is not gender neutral in Economics. Hilmer 2
male candidates might have a better chance to be acquainted with committee mem- bers and could perhaps benefit from these connections (Zinovyeva and Bagues 2015; Bagues, Sylos-Labini, Zinovyeva 2015). Third, men might hold more negative stereo- types of women than other women do or they may be biased against women reaching high-level positions. For instance, according to the World Value Survey, around 25% of US males believe that men make better political leaders and 16% think than men make better business executives. Women are half as likely to hold such views. A similar pattern is observed in Europe.2 According to some authors, similar biases arealsopresentintheacademicworld.3 Fourth,thepresenceofwomeninevaluation committees might also improve the quality of the evaluation. It has been argued that group performance is positively correlated with the proportion of women in a group (Woolley, Chabris, Pentland, Hashmi and Malone 2010). The presence of women in scientific boards might not only help to achieve gender balance in the academic profession, but it can also make science more meritocratic and invigorate its progress.
These arguments seem to have reached policy makers. A number of countries have introduced quotas requiring that scientific committees be at least 40% female (and male) and many universities and scientific institutions have their own internal guidelines ensuring the presence of both genders in committees.4 However, despite the increasing popularity of gender quotas in scientific committees, there are con- cerns about their effectiveness. Quotas are costly for senior female researchers, as
and Hilmer (2007) observe that in the US 55% of the Economics PhD students being advised by women are female, while only 18% of Economics PhD students advised by men are female.
2World Value Survey Wave 6: 2010-2014. Official aggregate v.20140429. World Values Survey Association (www.worldvaluessurvey.org).
3Gender discrimination in academia remains a controversial issue. According to meta-analyses by Ceci and Williams (2011) and Ceci, Ginther, Kahn, and Williams (2014), the more recent em- pirical evidence fails to provide any clear support to the assertions of discrimination in manuscript reviewing, interviewing, and hiring. However, other studies find that female researchers might still receive lower evaluations than male researchers with identical characteristics (Steinpreis, Anders and Ritzke 1999, Moss-Racusin, Dovidio, Brescoll, Graham and Handelsman 2012). Some experts in gender studies have also argued that male evaluators discriminate against female candidates. For instance, in a report commissioned by the European Commission, the expert group Women In Research Decision Making concludes that “(a)t the very least, having male only committees risks replicating stereotypes and bias, both regarding applicants and issues in research” (European Commission 2008, page 27). Another expert report on the situation of women researchers in Spain asserts that “there are prejudices about women among those who co-opt, promote or have the key to promotion. The bodies which control this are mostly male and, even if they are not totally conscious of it, they see an academic woman first as a woman and secondly as a colleague.” (Fun- daci ́on Espan ̃ola para la Ciencia y la Tecnolog ́ıa 2005, page 48). Other researchers have voiced similar views (Bagilhole 2005, Barres 2006, Smith et al. 2015).
4Gender quotas in scientific committees were introduced in 1995 in Finland (amendment of the Act on Equality between Women and Men, Act No. 624/1992 and No. 206/1995.), in 2007 in Spain (Constitutional Act 3/2007 of 22 March for Effective Equality between Women and Men) and in 2014 in France (decree No. 2014-997, September 2 2014). The European Commission has also committed to reaching a target of 40% female participation in its advisory structures for Horizon 2020, the European Union’s research and innovation programe for 2014-20.
3
they increase disproportionately the amount of time that they have to devote to evaluation committees (Vernos 2013). Furthermore, a larger presence of women in committees may not necessarily benefit female candidates. Both men and women have developed their careers in an academic environment dominated by men, and both genders may tend to associate important academic positions, and the fea- tures they require, with men, not with women (M ́endez and Busenbark 2012). And even if women are relatively more sympathetic towards female candidates, they may not have equal levels of voice and authority in deliberation processes (Karpowitz, Mendelberg and Shaker 2012, Brescoll 2011). The presence of women in the commit- tee can also induce male evaluators to be less favorable towards female candidates. Past research on group dynamics suggests that men might not respond favorably to the presence of gender diversity, particularly in domains that men have historically dominated (Crocker and McGraw 1984). Female evaluators can also contribute to strengthen male identity (Akerlof and Kranton 2000) or they can trigger a licensing effect (Monin and Miller 2001, Khan and Dhar 2006).
A better understanding of the impact of scientific committees’ gender composi- tion on recruitment and promotion decisions is crucial in order to determine whether quotas are desirable. The empirical evidence has been so far inconclusive and typi- cally based on small samples. Sometimes researchers seem to benefit from the pres- ence of evaluators who share the same gender (Casadevall and Handelsman 2013, De Paola and Scoppa 2015), sometimes they seem to obtain relatively better evalua- tions from opposite-sex evaluators (Broder 1993; Ellemers, Heuvel, de Gilder, Maass and Bonvini 2004), and in some other cases gender does not seem to play any (sta- tistically) significant role (Abrevaya and Hamermesh 2012; Jayasinghe, Marsh and Bond 2003; Milkman, Akinola and Chugh 2015; Moss-Racusin, Dovidio, Brescoll, Graham and Handelsman 2012; Steinpreis, Anders and Ritzke 1999; Williams and Ceci 2015). A brief summary of these studies is available in Appendix A.5 It is unclear whether these mixed findings reflect the idiosyncrasies of the different sit- uations and samples analyzed in each study, or simple random sampling variation. The literature also does not shed light on the mechanisms through which a higher presence of female evaluators in committees may benefit female candidates. From a policy perspective, the lack of more extensive and clear evidence is disappointing.6
In this paper we analyze the role of evaluators’ gender in academic evaluations
5All appendix material can be found in the Online Appendix.
6A related literature also analyses the role of evaluators’ gender in non-academic occupations (Bagues and Esteve-Volart 2010, Bertrand, Black, Jensen and Lleras-Muney 2014, Booth and Leigh 2010, Kunze and Miller 2014), in sport activities (Sandberg 2014) or in the lab (Bohnet, van Geen and Bazerman 2015). In general, in these studies evaluators’ gender is not relevant, with the exception of Bagues and Esteve-Volart (2010) who document that female applicants to the Spanish judiciary have lower chances of being hired when they are randomly assigned to an evaluation committee including women.
4
using the exceptional evidence provided by two large-scale randomized natural ex- periments in two different countries, Italy and Spain. The representation of women in Italian and Spanish universities is similar to their representation in other Euro- pean countries and the US. Despite having achieved parity at the lower rungs of the academic ladder, women are still underrepresented in top academic positions. They account for approximately half of new PhD graduates, one third of associate profes- sors but only one fifth of full professors in both countries.7 The Spanish and Italian institutional arrangements offer several unique features. In order to be either pro- moted or hired by a university at the level of associate or full professor, researchers are required to first obtain a qualification granted by a centralized committee at the national level. In these nation-wide examinations, which are performed periodically in all disciplines in both countries, evaluators are selected from a pool of eligible professors using a random draw. This allows us to consistently estimate the causal effect of committees’ gender composition on evaluations. We also observe extensive and detailed information on evaluators’ and candidates’ research production, aca- demic connections and their subfield of specialization. We exploit this information to explore the different mechanisms suggested by the theory about the role of com- mittees’ gender composition. Each country also offers some comparative advantage in terms of data availability. We use individual voting reports, available in Italy, to study the voting behavior of male and female evaluators within each committee. In Spain, we can observe the future productivity of promoted candidates. We use this information to examine the quality of the assessments granted by committees with different gender compositions. As we explain in more detail in section 2, there exist also a number of interesting institutional differences between the evaluation processes in the two countries. Having data for the two different institutional ar- rangements allows us to cross-validate the findings and to explore their robustness.
Our database includes information on all qualification exams that were con- ducted in Italy in years 2012-2014 and in Spain in years 2002-2006. Overall, these evaluations involved approximately 100,000 applications and 8,000 evaluators in all disciplines. Evaluation committees, which include five members in Italy and seven members in Spain, are composed mostly by men. Approximately one third of eval- uation committees do not include any women, in one third there is just one female evaluator, and in one third of committees there are two or more women, but very rarely we observe a female majority.
7In Italy, women account for 54% of new PhD graduates, 35% of associate professors, and 21% of full professors (Ministry of Education, University and Research, year 2014). In Spain, women account for 49% of new PhD graduates, 40% of associate professors, and 21% of full professors (Ministry of Education, Culture and Sports, year 2014). According to information from individuals who obtained a PhD in the 90s in Spain, female graduates are half as likely to attain full professorship than male graduates (S ́anchez de Madariaga, de la Rica and Dolado 2011).
5
In both countries male applicants tend to be more successful than female appli- cants. In Italy, approximately 38% of men receive a positive evaluation, compared to 35% of women. In Spain, 12% of male applicants qualify, while the success rate among female applicants is equal to 11%. When we take into account candidates’ observable productivity, the remaining gender gap is equal to 1.5 percentage points (p.p.) in Italy and 1.4 p.p. in Spain, and it is statistically significant in both countries. We find no empirical support, neither from the average in the two coun- tries nor from the majority of subsamples analyzed, to suggest that the presence of women in evaluation committees decreases the gender gap in a statistically or economically significant way. On the contrary, in Italy gender-mixed committees exhibit a significantly larger gender gap than committees composed only of male evaluators. An extra woman in a committee of five members increases the gender gap by somewhere between 0.4 and 3.3 p.p., considering a 95% confidence interval. In the Spanish case, we can reject any sizable impact. An additional woman in a committee of seven members may decrease the gender gap by at most 0.5 p.p. or it might also increase it by up to 1 p.p.
We also examine whether committees with a relatively larger proportion of women promote better candidates, using as a proxy of candidates’ quality their research output before the evaluation and, in the case of Spain, also their research output during the following five years. We do not observe any significant differ- ence in the past or future observable quality of candidates who have qualified in committees with different gender compositions.
Evidence from 300,000 individual voting reports, available in the case of Italy, suggests that there are two main factors that explain why female candidates do not benefit from a larger presence of women in committees. In mixed gender commit- tees, female evaluators rate female applicants higher than their male colleagues, but the difference is small and statistically non significant. At the same time, the pres- ence of female evaluators in committees makes male evaluators tougher upon female candidates, perhaps reflecting a licensing effect or male identity priming.
To gain a better understanding of why female evaluators do not exhibit a stronger same-sex preference and also to determine the validity of our findings in other con- texts, we explore why none of the standard theories predicting that a larger presence of women in committees helps female candidates plays a major role in this context. First, we consider the gendered networks hypothesis. As expected, we find that research networks tend to be gendered in both countries. Female professors are significantly more likely to have an advisor, a colleague or a coauthor of the same gender. We also observe that committees tend to favor connected candidates. How- ever, the likelihood of having a connection in a national committee is relatively low and, therefore, networks have only a limited effect on the evaluation outcomes. Sec-
6
ond, we examine the role of gender segregation across research subfields. At the level at which evaluations were conducted, around 200 different fields, gender segregation turns out to be relatively small. As a result, while evaluators tend to prefer candi- dates with a similar research profile, the impact of gender segregation on evaluations is negligible. Third, we study gender stereotypes. Stereotypes are expected to be more relevant when evaluators cannot observe accurately the quality of candidates, for instance because evaluators and candidates are specialized in different subfields of research. The influence of stereotypes on evaluation outcomes seems to increase, not decrease, when there are women in the committee. Finally, we also examine sep- arately evaluations for high-level positions. Male evaluators might have prejudices against women being promoted to full professorships, but not to positions at lower levels of the career ladder. Results are mixed: we find support for this hypothesis in the case of Spain, but not in the case of Italy.
Our study contributes to the literature in several ways. We provide the first large-scale assessment of the causal impact of the gender composition of scientific committees. There is no evidence suggesting that, in the two evaluation systems considered in this study, female candidates benefit from the presence of a larger share of women in evaluation committees. We also examine explicitly the relevance of the different theoretical arguments that have been proposed in the literature in favor of increasing the share of women in committees. This analysis helps to assess the external validity of our findings and, as we discuss in detail in the final section of the paper, it provides a better understanding of when gender quotas might be desirable. Finally, we open the black box of committee decision-making and we analyze the voting behavior of individual committee members. Our findings suggest that interactions within committees might exacerbate the impact of gender stereotypes.
2 Institutional background
Several European countries have national evaluation systems which are meant to guarantee the academic quality of professors in public universities. The evidence presented in this paper is based on an analysis of two variants of such systems: the Italian system known as Abilitazione Scientifica Nazionale, which was introduced in 2012, and the Spanish system known as Habilitacio ́n, which was in place between 2002 and 2006.
Both systems require candidates for associate and full professorships to qualify in national evaluations held by an academic board in the appropriate discipline. In each country, there are nearly two hundred legally defined academic disciplines, each corresponding to a certain area of knowledge. Successful candidates can then apply
7
for a position at a given university. The time line of evaluations has the following steps. First, a call for applicants is announced in which candidates can apply for multiple disciplines and positions. When the list of applicants is settled, committee members are randomly selected from the list of eligible evaluators in the correspond- ing discipline. Once the committees are formed, the evaluation process begins and once this is over, the evaluation results are made public. Rostered evaluators can potentially resign at any point of the process, something that happens in 2% of cases in Spain and in 8% of cases in Italy. Resigned evaluators are substituted by randomly selected evaluators.
The procedure has also distinctive features specific to each country. In Spain, evaluations involve oral presentations by the candidates, while in Italy evaluations are based only on candidates’ CVs and publications. In Spain qualification leads almost automatically to promotion, while in Italy the chances to get promoted con- ditional on obtaining qualification are much lower. The Italian system is relatively more transparent and exposed to public scrutiny. Nonetheless, in both systems there seems to be room for subjectivity. For instance, Zinovyeva and Bagues (2015) and Bagues, Sylos-Labini and Zinovyeva (2015) document that the presence of a coau- thor or a colleague in the evaluation committee has a significant positive impact on candidates’ chances of success in both countries.
We describe in detail the main features of each system below. This information is also summarized in Appendix B.
2.1 Abilitazione Scientifica Nazionale
In Italy, four out of five committee members are selected through a random draw from the pool of ‘Italian’ eligible evaluators and the remaining evaluator is drawn out of the pool of ‘foreign’ eligible evaluators. The former pool consists of full professors affiliated to Italian universities who volunteered to be members. The latter pool consists of professors affiliated to universities from OECD countries, who also voluntarily participate in Italian evaluations. The randomization procedure is subject to one important constraint: no university can have more than one evaluator within a single committee.
The eligibility of evaluators is decided in the following way. In science, tech- nology, engineering, mathematics, medicine and psychology (STEMM), evaluators are required to have a research output above the median for full professors in the discipline in at least two of the following three dimensions: (i) the number of articles published in scientific journals, (ii) the number of citations, (iii) and the H-index. In the social sciences and the humanities (SSH), the research performance of evaluators has to be above the median in at least one of the following three dimensions: (i)
8
the number of articles published in high quality scientific journals (in what follows, A-journals),8 (ii) the overall number of articles published in any scientific journals and book chapters, and (iii) the number of published books. ‘Foreign’ eligible eval- uators have to satisfy the same requirements. While ‘Italian’ evaluators work pro bono, ‘OECD’ evaluators receive e16,000 for their participation.
Evaluations are based solely on the material provided in candidates’ application packages consisting of CVs and recent publications. Committees have full autonomy regarding the criteria to be used in the evaluation and the number of qualifications to be granted. Each evaluation committee is required to draft and publish on- line a document describing the general criteria to be used in providing a positive assessment. Candidates may withdraw their application up until two weeks after evaluation criteria are publicized. A positive assessment of the candidate requires a qualified majority of four out of five votes. Once granted, qualifications are only valid for four years, while a negative evaluation means that candidates are excluded from participating in further national evaluations during the following two years.
An important feature of the Italian system is its extreme transparency: all the relevant information – including candidates’ and evaluators’ CVs, as well as indi- vidual evaluation reports – is published online. An independent evaluation agency appointed by the ministry also collects and publicizes information on the research output of final candidates in the ten years preceding the evaluation, as measured by the three bibliometric indicators described above. The evaluation agency com- pared the research productivity of candidates in each of these three dimensions with the research productivity of professors in the category to which they applied, and committees were asked to take this information into consideration.
2.2 Habilitacio ́n
In Spain, committees are composed of seven members. In evaluations for full profes- sorships, all evaluators are full professors based in Spanish universities or research institutes. In evaluations for associate professorships, three committee members are full professors and four evaluators are associate professors. No more than one non- university researcher is allowed to be selected as a member of the committee for a given exam. Similarly, no more than one emeritus professor may be selected as a member of a given committee.
In order to be eligible, evaluators are required to satisfy some minimum research level which is assessed by the Spanish education authority.9 This requirement is
8An evaluation agency determined with the help of several scientific committees the set of journals to be considered as high quality in each field.
9The Spanish education authority determines professors’ eligibility according to the number of sexenios completed. Sexenios are granted periodically by the ministry on the basis of applicants’
9
satisfied by approximately 81% of full professors and 70% of associate professors. Unlike the Italian system, where participation is voluntary, in Spain all eligible professors can be selected to serve in committees.
Candidates for evaluation are required to make several oral presentations in front of a committee. For candidates to full professorships, these exams have two qualifying stages. In the first stage, each candidate presents the CV and then, in the second stage, an example of his or her research work. Exams for the position of associate professor, in addition to these two stages, have an intermediate stage where candidates give a lecture on a topic randomly chosen from a syllabus proposed by the candidate. In each stage evaluations are made on a majority basis. Qualifications have unlimited validity once they have been granted. The number of qualifications conceded at the national level is very limited and being accredited is, in most cases, equivalent to being promoted.
3 Data
We use data on all evaluations from the first edition of the Italian Abilitazione Sci- entifica Nazionale (years 2012-2014) and on all evaluations from the Spanish Habil- itaci ́on (years 2002-2006). In Italy, the data includes information on 184 committees, one per each academic discipline. Each committee assessed both applications to as- sociate and to full professorships. In Spain, there are in total 967 committees in 174 disciplines, of which 502 are committees evaluating candidates for full professorships and 465 evaluating candidates for associate professorships.
The dataset includes information on eligible and actually selected evaluators, applicants, and the final outcome of the evaluation. In addition to demographic characteristics and a number of productivity measures, we have also gathered infor- mation on research networks and research specialization. In Appendix C we provide detailed information on how this information was collected, and how each variable was constructed. Below we briefly summarize the main features of the dataset.
3.1 Evaluators
In Italy, 39% of Italian female full professors and 41% of Italian male full professors volunteered and were considered eligible to sit in evaluation committees. The list of eligible evaluators includes 5,876 professors based in Italian universities and 1,365 evaluators based in OECD universities. In the average field, the pool of eligible
research output in any non-interrupted period of a maximum of six years. Eligible associate professors are required to have held at least one sexenio while eligible full professors are required to have held at least two sexenios.
10
evaluators includes 32 ‘Italian’ professors and eight ‘foreign’ professors. While ap- proximately 20% of ‘Italian’ evaluators are women, the ‘foreign’ pool is less feminized and only 12% of ‘foreign’ evaluators are women. Taking into account the composi- tion of both pools, the expected share of women in the committee is around 18%, whichissimilartotheinitialshareofwomeninactualcommittees.10 Approximately one out of every thirteen evaluators resigned and was replaced by another eligible evaluator. These replacements slightly increased the share of women in committees to 19%, but the difference is not statistically significant. 41% of committees include no women at all, in 35% there is one woman, in 16% there are two women, and only 8% of committees have a majority of female evaluators.
Table C1 provides descriptive information on eligible evaluators based in Italy.11 On average, they have been in a full professor position for 13 years. They list 131 publications in their CVs, of which just over half are articles in scientific jour- nals, and the rest are books, book chapters, publications in conference proceedings, patents, etc. Around half of these publications were published during the previous ten years. To assess the quality of research output, in STEMM disciplines we com- pute their total Article Influence Score, summing up the Article Influence Score of all publications; in SSH disciplines we use the number of articles in A-journals.12 About 28% of eligible professors are based in the South of Italy.
In columns 2-4, we compare characteristics of male and female evaluators. For this comparison, we normalize all variables at the discipline level. Female evaluators have significantly shorter tenure than their male counterparts and they also have lower research output in almost all dimensions. They are less likely to be based in the South, but this difference is not significant.
In Spain, the lists of eligible evaluators include 49,199 full professors and 61,052 associate professors.13 Women constitute 35% of eligible associate professors, but only 14% of full professors are women. Taking into account the composition of both pools, the expected share of women in the committee is around 19%. This figure is similar to the share of women in the initial set of committees selected by random draw and is unaffected by the resignation of 2% of evaluators. Overall, 32%
10We have calculated the expected gender composition of committees using a simulation with 1,000,000 draws, taking into account that the lottery that decided committee composition was subject to the constraint that committees cannot include more than one member from the same university.
11Unfortunately, we were unable to gather systematic information on ‘foreign’ evaluators. In their case, the official CVs are not in a standardized format and they are often incomplete.
12Article Influence Score is available for all journals in the Thomson Reuters Web of Knowledge. It is related to Impact Factor, but it takes into account the quality of the citing journals, the propensity to cite across journals and it excludes self-citations.
13The Spanish data covers information from several evaluation waves, so many professors appear in the lists several times. In total, there are 7,963 individual full professors and 21,979 individual associate professors in these lists.
11
of committees are composed by only male evaluators, 29% of committees have one woman on board, 22% include two women, 11% three women, while only 6% have more women than men.
We collect information on the research outcomes of Spanish researchers from several sources. We observe their publications in international journals covered by Web of Knowledge and their articles and books in the Spanish language included in the database Dialnet, as well as patents in the European Patent Office in which these researchers are listed as inventors. We also have information on their activity as Ph.D. advisors and as members of dissertation committees. We compare female and male eligible evaluators, normalizing their characteristics at the level of exam and category. Results are very similar to the ones observed for the Italian academia (see columns 6-8 and 10-12 of Table C1). Female eligible evaluators are younger, have shorter tenure, and on average they published less than male researchers in the same discipline and rank. They have also lower accumulated quality-adjusted scientific production, they tend to participate less in advising and evaluating doctoral students, and they are relatively less likely to come from universities located in the southern regions of the country.14
3.2 Candidates
There were 69,020 applications in Italy. On average, there were 375 applications per field, with 117 of them participating in evaluations for full professor positions and 258 participating in evaluations for associate professor positions. Some candi- dates applied to more than one position: the average candidate participated in 1.5 evaluations.
As shown in the upper panel of Table C2, 31% of applications for the position of full professor and 41% of applications for the position of associate professor were submitted by women. Candidates for a full professorship are about 49 years old and candidates for an associate professorship are six years younger. About half of the applicants for associate professorships hold a permanent contract and about three fourths of applicants for full professorships do. Candidates mainly apply for an evaluation in the field in which they currently hold a permanent contract.
Female applicants tend to be younger among applicants for associate professor- ships, and they are of a similar age as their male counterparts in evaluations for full professorships (columns 3-5 and 8-10). In both cases the publication record of female candidates is significantly weaker. The only dimension in which women seem to be achieving better results than men is in publishing conference proceedings. In addition to information on productivity coming from candidates’ CVs, we observe
14In Spain, we define A-journals following the journal rank developed by Dialnet, which catego- rizes journals in four groups according to their prestige.
12
the order in which candidates submitted their applications. In principle, the timing of the application might reflect both candidates’ self-confidence and quality. We normalize this variable uniformly between 0 and 1. We observe that female candi- dates for the post of full professor apply a bit later than their male counterparts, but no similar gender difference can be observed among candidates for associate profes- sor positions. In Italy, approximately 14% of applications were withdrawn once the identity and the criteria of evaluators were made public. Withdrawals were more common among female applicants. Overall, approximately 38% of applications by male candidates and 35% of applications by female candidates were successful.
As explained above, the evaluation agency of the Ministry of Education published detailed information regarding the research production of the final set of applicants in the 10 previous years. Around 38% of candidates were above the median in each of the three corresponding bibliometric dimensions. Performance according to these indicators is strongly correlated with success. Among those candidates whose quality was below the median in every dimension there was a success rate of only 4%, while among those who excelled in every dimension there was a success rate of 63%.
In addition to the final decision of the committee, we also collected informa- tion on the individual evaluation reports, available in the case of Italy. Overall we observe around 300,000 individual reports. 45% of these reports were favorable to the candidate and most of the time decisions were taken unanimously (in 86% of the cases). Unanimity is relatively more frequent when applicants are below the median in each of the three corresponding bibliometric dimensions (93%) and when applicants are above the median in all three dimensions (86%), and it is lower when applicants are above the median only in one (84%) or two dimensions (82%).
In Spain, overall there were 13,444 applications for full professorships and 17,799 applications for associate professorships (lower panel of Table C2). The gender ratios among applicants are very similar to the ones in Italy: around 27% of applicants to full professor are women and there are around 40% of women among applicants to associate professor. Once again, male applicants seem to have stronger research records than their female counterparts. They also tend to be slightly more successful in evaluations.
Finally, for the candidates who qualified in Spain, we collected information on their individual research productivity in a five-year period following the national evaluations and on their performance in future evaluations for promotion to full professor. This information allows us to assess the quality of selection not only in terms of candidate characteristics easily observable at the moment of the exam, but also in terms of dimensions that are difficult to observe but that are nevertheless important determinants of future productivity.
13
3.3 Connections
We identify professional links between candidates and eligible evaluators. We con- sider all the possible interactions within each discipline, around 2.5 million possible pairs in Italy and 5.5 millions in Spain. As shown in Table C3, the probability that a candidate and an eligible evaluator are affiliated to the same institution is around 3% in Italy and 5% in Spain. The probability that they have coauthored a paper is smaller: 1.4% in Italy and 0.4% in Spain.
In the case of Spain, we also observe if there was a student-advisor relationship or if the candidate and the eligible evaluator have participated in the same thesis committee.15 These links are relatively rare: in 0.2% of the cases the eligible evalu- ator is the PhD thesis director of the candidate and in 1.3% they have participated in the same thesis committee.
Male candidates tend to have more coauthors among eligible evaluators and they are more likely to have interacted with an eligible evaluator previously in a thesis committee (Table C3, columns 3-5).
3.4 Research similarity
We also collect information on the overlap of research interests between candidates and eligible evaluators. Due to data availability, there are some differences in how we define research similarity in the two countries. In the case of Italy, we have information on the field and the subfield where researchers with a permanent con- tract in an Italian university are officially registered. There are 184 fields (settore concorsuale ) and approximately 370 subfields (settore scientifico-disciplinare ).16 In about 60% of the cases the candidate and the eligible evaluator belong to the same subfield (Table C3).
In the case of Spanish researchers, we infer their research interests using infor- mation on their participation in doctoral dissertations, either as authors, advisors, or committee members. In Spain, all doctoral theses are classified in more than two thousand categories.17 Economics, for example, is divided into one hundred different research fields (e.g.: Labor Economics). We construct a measure of the overlap of the research interests of candidates and evaluators based on the subfield
15We consider three possible interactions: (i) the evaluator was a member of candidate’s thesis committee, (ii) one of them had invited the other to sit in her students’ thesis committee, or (ii) both of them sat in the same student thesis committee.
16Historically, each Italian researcher was a assigned to certain settore scientifico-disciplinare. More recently, upon the introduction of the new system of competitive exams, researchers were assigned also to a settore concorsuale. The correspondence between the two classifications is not always unique, in some cases researchers belonging to the same settore scientifico-disciplinare may be assigned to different settore concorsuale.
17The author of the dissertation selects the subfield using the International Standard Nomen- clature for Fields of Science and Technology, a system developed by Unesco.
14
of every dissertation where they have been involved. In the spirit of Jaffe (1986) and Bloom, Schankerman and Van Reenen (2013), we measure research proximity between individuals i and j as the angular separation of the vectors Si = (S1i…SCi) and Sj = (S1j…SCj), where SCi is the share of dissertations in category C in which individual i has been involved:
S i S j′
Overlapij = (SiSi′)1/2(SjSj′)1/2. (1)
This index takes value one if two individuals have participated in dissertations in the same subfields in the same proportion and value zero if there is no overlap. On average, in our sample the degree of overlap between candidates and evaluators is equal to 0.20. As shown in Table C3, female candidates are slightly more likely than male candidates to share their research interests with eligible evaluators.
4 Empirical analysis
We start our analysis by providing descriptive information on the average success rate of male and female applicants, unconditional and conditional on their observ- able research productivity. Then we investigate how the gender composition of committees affects the success rate of male and female candidates, candidates’ de- cision to withdraw their application and the quality of male and female applicants who qualify. To achieve a better understanding of the observed patterns, we use the information provided by individual voting reports to examine how male and female evaluators vote within the same committee. We examine whether male and female evaluators vote differently depending on the gender of applicants, and we also in- vestigate whether the presence of women in a committee affects the voting behavior of male evaluators. Finally, we explore the relevance of the main theories according to which evaluators’ gender may be relevant.
4.1 Gender gap
We estimate the gender gap separately for the applicants in the two countries using the ordinary least squares (OLS) method:
Yie =β0 +β1Femalei +Xiβ2 +μe +εie (2)
where Yie is a dummy variable that takes value one if candidate i qualifies in eval- uation e and takes value zero if the candidate receives a negative evaluation or withdraws the application before receiving the evaluation. Each evaluation e refers to the examination that was conducted in a given field and position (e.g. qualifi-
15
cation for an associate professorship in Applied Economics in Spain in year 2005). F emalei is a dummy variable indicating the gender of the candidate and Xi includes all (normalized) productivity indicators and individual characteristics listed in Table C2. We allow the effect of productivity indicators to vary across disciplinary groups, and the effect of age and contract type to vary across disciplinary groups and levels of promotion. Evaluation fixed effects (μe) control for any differences across evalu- ations that might affect the success rate of male and female candidates in a similar way. Throughout the analysis, we cluster standard errors at the committee level.
In Italy, the success rate of female candidates is 2.8 p.p. lower than male candi- dates in the same exam, unconditional on any measure of quality (Table 1, column 1, upper panel). In Spain, the unconditional gender gap is equal to 2.2 p.p. (column 1, lower panel). In both countries, approximately half of the gender gap can be ex- plained by the differences in observable characteristics (column 2). The remaining conditional gender gap is equal to 1.5 p.p. in Italy (4% relative to the success rate of men) and 1.4 p.p. in Spain (12%) and it is statistically significant in both countries.
It is unclear whether the remaining gap should be attributed to evaluators biases or to differences in unobservable characteristics. There may be substantial differ- ences in the quality of male and female candidates which are not fully captured by our controls. Furthermore, the individual proxies of quality that we use in our analysis, such as position, affiliation or publications might also be the outcome of discriminatory processes, which would further hinder the interpretation of β1.
4.2 The impact of committees’ gender composition on the chances of success of male and female candidates
We examine whether the gender composition of committees affects the success rate of male and female applicants. In order to obtain causal estimates, our analysis exploits the random assignment of evaluators to committees. We compare the performance of applicants who initially were expected to face an evaluation committee with the same gender composition but, due to the random draw, were assigned to committees with a different number of female evaluators. Given that a few of the evaluators who were initially selected eventually declined to participate and were substituted by other (randomly selected) evaluators, first we report results from an intention- to-treat analysis where our independent variable is the gender composition of the initial set of evaluators. Later on, we instrument the gender composition of the committee which actually evaluated applicants using the gender composition of the committee initially drawn.
16
4.2.1 Intention-to-treat analysis
We estimate the following equation on the pool of applicants using OLS:18
Y = β + β Female + β Femaleinitial + β Female ∗ Femaleinitial+
ie 0 1 i 2 e 3 i e
+ β Femaleexpected + β Female ∗ Femaleexpected + X β + ε (3)
4 e 5 i e i6ie
where Femaleinitial represents the share of female evaluators in the committee that
e
was initially randomly drawn, before any evaluator resigned, and Femaleexpected is e
the expected share of women in this committee, calculated based on the composition ofthepoolofeligibleevaluatorsandtherulesthatdeterminethedraw.19 Inorderto increase the accuracy of the estimation, we also include applicants’ predetermined characteristics (Xi) and, in some specifications, evaluation fixed effects (μe).
Coefficient β2 captures the causal effect of committees’ initial gender composition
upon the success rate of male candidates, and coefficient β3 shows how the gender
gap varies depending on the share of women in the committee. Since Femaleinitial e
is computed using the initial assignment of evaluators, coefficients β2 and β3 pro- vide intention-to-treat estimates. The causal interpretation of β2 and β3 relies on the assumption that the assignment was indeed random. The way in which the randomization was conducted in each country suggests that there was little room for manipulation.20 Nonetheless, before moving into the discussion of the impact of committees’ gender composition on candidates’ chances of success, we verify em- pirically that, conditional on the expected composition of the committee, its actual composition is uncorrelated with any observable predetermined factor. We estimate equation (3) using predetermined characteristics included in Xi as outcome variables instead of controls. As expected, the evidence is consistent with the assignment be- ing indeed random. Table 2 shows estimation results for the eleven predetermined variables that are common for Italian and Spanish databases. Out of forty four coefficients, only two are significantly different from zero at 5% level. A joint F-test cannot reject that quality of female and male candidates is similar across committees with different gender compositions.
We examine the causal impact of committees’ gender composition in column 3 of Table 1. In Italy, the proportion of women in committees has no significant impact on the success rate of male candidates and it has a significant negative impact on
18Results from probit estimations are very similar and are available upon request. We report the results for the linear probability model because interpreting the interaction effects is simpler.
19To ease the interpretation of coefficient β , we center Femaleexpected at zero by subtracting 1e
its sample mean.
20In Italy, a random sequence of numbers was drawn and was then applied to several disciplines.
In Spain, the random draw was carried out publicly on the same day for all disciplines and was certified by the notary.
17
the relative chances of success of female candidates (upper panel). An additional female evaluator decreases the relative chances of success of female candidates by approximately 1.8 p.p. (β2 = −0.092, ∆Femalee=1/5). In Spain, the share of female evaluators has a positive effect on the success rate of male candidates and a negative effect on the success rate of female candidates, though these effects are not significantly different from zero (lower panel).
To make estimates from Spain and Italy more comparable, it is useful to consider explicitly the upper and the lower bounds of a 95% confidence interval. In Italy, an additional woman in the committee decreases the success rate of female candidates relative to men by somewhere between 0.4 and 3.3 p.p. In Spain, an extra woman on the committee can lower it by at maximum 1.0 p.p. but she can also increase it by up to 0.5 p.p. In sum, the impact that women in committees have upon the relative success rate of female candidates is negative and statistically significant only in the Italian case, but we cannot reject that the effect is statistically similar in the two countries.
4.2.2 Instrumental variables estimates
To account for the resignation of some evaluators before the actual evaluation took place, we instrument the final gender composition of the committee using as an instrument the initial composition determined by the random draw. Specifically, we estimate the following equation using the instrumental variables (IV) method:
Y = β + β Female + β Femalefinal + β Female ∗ Femalefinal+ ie 0 1 i 2 e 3 i e
+ β Femaleexpected + β Female ∗ Femaleexpected + X β + ε (4) 4 e 5 i e i6ie
where Femalefinal represents the share of female evaluators in the committee that e
evaluated candidates, and Femalefinal and Female ∗Femalefinal are instrumented eie
using Femaleinitial and Female ∗ Femaleinitial. eie
The first stage results of the IV estimation show that there is a strong relation- ship between the initial and the final gender composition of committees (see Table D1). The IV estimates are slightly larger but very similar to the intention-to-treat estimates (column 4 of Table 1). To further increase the precision of these estimates, we also reestimate equation (4) including evaluation fixed effects. The estimates are slightly more accurate but they are (statistically) unchanged (column 5 of Table 1).
Female and male evaluators differ in a number of dimensions. As shown in Table C1, male evaluators tend to be relatively older, have longer tenure, and a longer publication record. They are also more likely to be based in the south of Italy and Spain. In order to check whether our results can be explained by these differences, we estimate equation (4) including the interaction between evaluators’
18
characteristics and candidates’ gender. The inclusion of these controls does not affect our previous estimates (Table 1, column 6).
The range of variation in gender composition that we exploit in our analysis is typically between committees with no women and committees with a minority of women. In Appendix E we also show that within this range there are no significant non-linearities.
4.3 Does the presence of women in the committee affect candidates’ decision to withdraw?
So far we have considered the initial sample of candidates. Some of these candidates dropped from the evaluation process after committees were formed, perhaps because they anticipated that they had only a small chance to qualify and they preferred to avoid the costs associated to failure. These candidates did not receive an evaluation from the committee.
Therefore, the above estimates may in principle capture the effect that the gender composition of a committee has upon candidates’ decision to self-select into the process. To examine this issue, we use data from Italy and estimate equation (4) using as the dependent variable the indicator for those candidates who did not withdraw their application. While relatively fewer women decided to go ahead with the application (-2.6 p.p.), these differences are not related to the share of female evaluators (Table 1, column 7). The evidence thus suggests that committees’ gender composition does not affect application decisions and its impact on the chances of success of candidates can be attributed to evaluations.
4.4 Does the presence of women in the committee affect the quality of promoted candidates?
An additional justification for increasing female representation in committees might be that female researchers help to reduce evaluation biases and select better can- didates, even though not necessarily more female candidates. To learn about the quality of the assessments, we compare the observable productivity of candidates who qualified in committees with different gender compositions:
q = β + β Femalefinal + β Femaleexpected + ε (5) ie 0 1 e 2 e ie
where qie is a proxy of candidate i’s quality, measured at the time of the evaluation or during the following five years. We estimate equation (5) for all qualified candi- dates, and then separately for females and males. We instrument the final gender composition of the committee (Femalefinal) using the original one (Femaleinitial).
19
We consider several proxies of quality. First, we consider the research output of successful candidates at the time of the evaluation. As shown in Table 3, candidates that were promoted by committees with a different gender composition are at the time of the evaluation statistically similar in terms of the number of papers that they have published, the quality of the journals, the number of students advised or their participation in theses committees.
Using the Spanish data, we also examine the research productivity of successful candidates during the five-year period following the evaluation. Additionally, for the candidates who qualified to positions of associate professor, we check whether they succeeded in obtaining a qualification for full professorship. Once again, we see no evidence that the quality of candidates who qualify is related to the number of women who sat on these candidates’ evaluation committees. Overall, we do not observe any indication that committees with more female evaluators select better or worse candidates.
4.5 Individual voting
We have documented that mixed-gender committees are not more favorable towards female candidates than all-male committees. This finding is consistent with several possibilities. It might be that female evaluators are not more favorable towards female candidates than their male counterparts. Alternatively, maybe female eval- uators are more sympathetic towards female candidates (or less unbiased) but their presence in the committee induces male evaluators to become less favorable towards female candidates. To shed light on this issue, we analyze the information provided by individual voting reports, available in Italy.
First, we compare the assessments of male and female evaluators sitting in the same committee. We estimate the following equation:
Vije = β0 + β1Femalej + β2Femalei ∗ Femalej + μie + εije, (6)
where Vije takes value one if evaluator j casted a positive vote for candidate i in evaluation e, and Femalei and Femalej are indicators that capture the gender of the candidate and the evaluator respectively. A vector of application fixed effects μie captures any differences in application characteristics that are observable to all evaluators.
The empirical results suggest that, if anything, female evaluators are more favor- able towards female candidates than male evaluators. Female candidates are 0.7 p.p. (1.6%) more likely to receive a positive vote from a female evaluator than from a male evaluator, although this difference is not statistically different from zero (Table 4, column 1). This estimate is likely to be a lower bound of the overall effect. Com-
20
mittee members share information and discuss their decision before casting their vote. A high fraction of committees reach unanimous decisions, suggesting that there may be less disagreement reflected in these final individual evaluations than there would have been at interim stages.
Another question that we would like to answer is whether the voting behavior of male evaluators changes when there are women on the committee. We estimate the following equation on the sample of assessments granted by male evaluators:
Vije = β0 + β1Femalei + β2Femalefinal + β3Femalei ∗ Femalefinal+ je je
+ β4F emaleexpected + β5F emalei ∗ F emaleexpected + Xiβ4 + εij , (7) je je
where Femalefinal and Femaleexpected stand respectively for the actual and the ex- je je
pected share of women in a committee including evaluator j.21 Coefficient β2 cap- tures how the probability that a male candidate receives a positive vote from a male evaluator varies depending on the gender composition of the committee. Similarly, coefficient β3 captures how the presence of women in the committee affects the probability that a female candidate receives a positive vote from a male evaluator, relative to a male candidate.
There are three possible threats to the consistency of our estimates. First, simi- larly to the analysis conducted in previous sections, the initial assignment of evalua- tors to committees should be random. As shown above, this assumption is satisfied. Second, we do not observe the assessments that would have been casted by evalua- tors who resigned (8% of initial evaluators). This might introduce a selection bias if resignations are related to the gender composition of the committee or to eval- uators’ gender biases. We examine this possibility in Appendix F. We do not find evidence suggesting that resignations are related to gender issues. Third, given that we only observe the evaluations received by candidates who did not withdraw their application (86% of applicants), a bias might arise if candidates’ withdrawal deci- sion somehow depends on committees’ gender composition or gender biases. Our previous analysis shows that the gender composition of committees does not affect application decisions (see section 4.3). As a robustness check, we also consider an additional specification where we impute a negative assessment to every withdrawn application.
According to our estimates, each additional female evaluator in the committee
increases the probability that a male candidate receives a positive vote from a male
evaluator by 0.3 p.p. (β = 0.017, ∆Femalefinal = 1/5) and it decreases the proba- 2e
bility that a female candidate receives a positive vote, relative to a male candidate,
21We compute these expectations separately for each evaluator using the outcomes of 1,000,000 simulated random draws that take into account the rules of the randomization.
21
by 0.8 p.p. (β = −0.042, ∆Femalefinal = 1/5), although these estimates are not 3e
significantly different from zero (Table 4, column 2). To increase the accuracy of
the estimation, we also include evaluation fixed effects. According to this specifica-
tion, each additional woman in the committee reduces the probability that a female
candidate receives a positive vote from a male evaluator by 1.2 p.p. (β3 = −0.061,
∆Femalefinal = 1/5), relative to the probability that a male candidate receives a e
positive vote (column 3). This effect is significant at the 5% level. The estimate is slightly larger, around 1.6 p.p., if we consider in our analysis also candidates who withdrew their application (column 4).
4.6 Mechanisms
The two large-scale randomized natural experiments provide a clear result: increas- ing the proportion of women in scientific committees does not increase the success rate of female candidates. The analysis of individual votes within the committee suggests that this is due to two factors. On the one hand, female evaluators are slightly more likely to vote in favor of female candidates than male evaluators, but this effect is not economically or statistically significant. On the other hand, the presence of women in the committee decreases the probability that female candi- dates receive a positive vote from male evaluators. Next, we analyze these two issues in more detail.
4.6.1 Why are women not more supportive of other women?
The literature has emphasized several theoretical arguments according to which evaluators are expected to favor same-sex candidates. The most prominent ones are the existence of gender segregation across research networks, gender segregation across subfields of research, gender stereotypes and discrimination against women attaining top positions. Next, we provide an in-depth examination of these theories and we try to understand why they do not play a more important role in our data.
Gender segregation across research networks One of the arguments behind gender quotas is the existence of ‘old boy networks’. If professional connections with committee members help to achieve success and, at the same time, these connec- tions are gendered, female candidates might be at a disadvantage when evaluation committees do not include women. The relevance of ‘old boy networks’ depends on three factors: (i) the extent to which networks are gendered, (ii) the likelihood that applicants are evaluated by a member of their network, and (iii) the magnitude of the connection premium.
First, we examine whether research networks in Spain and Italy are gendered. 22
We consider all possible pairs between candidates and potential evaluators within a given field and we analyze whether the probability of being linked varies with their gender:
Lij = β0 + β1Femalei + β2Femalej + β3Femalei ∗ Femalej + μe + εij (8)
where Lij stands for any of the observable links between candidate i and eligible evaluator j. Femalei and Femalej are indicators for female candidates and eligible evaluators, and μe are evaluation fixed effects.
As expected, links tend to be gendered. The β3 estimate is positive and sig- nificant in all specifications, indicating that, when a male eligible evaluator is sub- stituted by a woman, female candidates’ likelihood of being connected increases relatively more than male candidates’ in every dimension (Table 5). In Italy, the likelihood of observing a female professor with the same affiliation as a female candi- date is 0.6 p.p. (20%) larger than the likelihood of observing a similar link between a female professor and a male candidate.22 In Spain, female professors are 0.4 p.p. (8%) more likely to be in the same institution as a female candidate, relative to the probability of being affiliated to the same institution as a male candidate. Coau- thorships are also relatively more likely when individuals share the same sex. In Italy female professors are 0.3 p.p. (23%) more likely to coauthor with a female candidate than with a male one; in Spain the premium is equal to 0.1 p.p. (23%). Similarly, PhD supervisions and participation in PhD committees are also gendered. Female professors are 0.04 p.p. (33%) more likely to have a female advisee and 0.03 p.p. (3%) more likely to have participated in the same dissertation committee as a female candidate.
Another relevant factor is whether candidates benefit from the presence of a member of their network in an evaluation committee. Previous work by Zinovyeva and Bagues (2015) and Bagues, Sylos-Labini and Zinovyeva (2015) documents the existence of a substantial connection premium in qualification exams in Spain and Italy. However, while connections in evaluation committees might be useful, they are relatively rare in a context where evaluations are conducted at the national level. For instance, as pointed out in section 3.3, the probability that a candidate and an eligible evaluator are colleagues is around 3% in Italy and 5% in Spain. The probability that they are coauthors is even lower, around 1.4% in Italy and 0.4% in Spain. In sum, we observe a relative large degree of gender segregation across networks and also a substantial connection premium, but the impact of these two
22We have calculated this figure using the information reported in Table 5, column 1. The probability that a female professor and a female candidate in the Italian sample are affiliated to the same university is equal to 3.34% (0.0026+0.0017+0.0029+0.0262), and the probability that a female professor and a male candidate are colleagues is equal to 2.79% (0.0017+0.0262).
23
factors is likely to be attenuated by the scarcity of connections in committees.23 Next, we study whether taking into account connections between candidates and evaluators affects our estimates of the impact of committees’ gender composition.
We estimate the following equation:
Y =β +β Female +β Female ∗Femalefinal +L finalβ +
ie 0 1 i 2 i e ie 3
+ β Female ∗ Femaleexpected + Lexpectedβ + X β + μ + ε (9)
4 i e ie 5i6eie
where Liefinal is a vector including the different types of links between commit-
tee members and candidates. We also include as controls the expected proportion
of links in the committee Lexpected and we instrument the final composition of ie
the committee (Femalefinal, L final) using the outcome of the initial lottery draw e ie
(Femaleinitial, L initial). The vector of coefficients β provides information about eie 3
the causal impact of connections in the committee.
Table 6 reports the results of this analysis. In line with the findings of Zinovyeva
and Bagues (2015) and Bagues, Sylos-Labini and Zinovyeva (2015), we find that connections with evaluators are helpful for promotion. The presence of a colleague in the committee increases the success rate of connected candidates by 3.6 p.p. (10%) inItalyandby4.6p.p.(41%)inSpain.24 Theimpactofcoauthorsislarger:4.7p.p. (13%) in Italy and 12.8 p.p. (112%) in Spain. Candidates with an advisor in the evaluation committee also enjoy a premium of 9.0 p.p. (79%) and when an evaluator has interacted previously with the candidate in some thesis committee the premium is around 2.5 p.p. (22%). However, the inclusion of connections as controls in the analysis does not affect significantly our estimates of the effect of evaluators’ gender on candidates’ success rate (columns 1 and 5 vs. columns 2 and 6). As pointed out above, a plausible explanation for why connections, while being gendered, do not affect significantly our estimates may be related to their scarcity. For instance, in Italy the probability that a female candidate and a male evaluator are coauthors is around 1.4%. This probability increases by 0.1 p.p. when the evaluator is also female. Taking into account the premium associated to the presence of a coauthor in the committee (4.7 p.p.), replacing a male evaluator by a female one translates into an increase in the average success rate of female candidates by a mere 0.005 p.p. Moreover, as we show in Appendix G, evaluators’ support of connected candidates
23There are may be also weaker links between candidates and evaluators, such as the existence of a common a coauthor. Zinovyeva and Bagues (2015) show that these indirect links tend also to be gendered but they do not have a significant impact on evaluation outcomes.
24To calculate these figures we take into account the number of committee members in Italy
and Spain (5 and 7 respectively) and the average success rate in each country (37% and 11%).
For instance, in Italy the presence of a colleague in the committee has an impact of 3.6 p.p.
((β = 0.181, ∆Femalefinal = 1/5). Relative to an average success rate of 37%, this implies a 3e
10% premium.
24
does not depend on their gender.
Gender segregation across research subfields Another argument in favor of increasing the share of women in committees has been the potential existence of gender segregation across subfields. If committee members tend to prefer candidates with similar research interests and, at the same time, men and women are segregated across research subfields, the lack of women in committees might hinder the ability of female candidates to succeed.
The extent of gender segregation across subfields is likely to depend on the level of aggregation at which evaluations are held. Segregation is probably larger when applicants are grouped in a few broadly defined fields. In the nation-wide evaluations that we analyze in this paper, applicants were classified in approximately 200 differ- ent fields (e.g. Applied Economics). We check whether, at this level of aggregation, candidates are more likely to have the same research interests as eligible evaluators of the same gender. We estimate equation (8) using as the dependent variable the research similarity between candidates and eligible evaluators. We observe gender segregation across research subfields in both countries but its magnitude is relatively small. In Italy, a female eligible evaluator is 1.3 p.p. relatively more likely to be in the same subfield as a female candidate than in the subfield of a male candidate. In Spain, the overlap between a female eligible evaluator and a female candidate is 0.4 p.p. larger (Table 5, columns 3 and 8).
Research similarity with evaluators tends to increase candidates’ chances of suc- cess, but the effect of female evaluators on female candidates’ relative success rate is unchanged when we control in the estimation for research similarity (Table 6, columns 3-4 and 7-8). This is consistent with the relatively small level of gender segregation observed. In sum, gender segregation across research interests is too limited for female candidates to benefit significantly from more female evaluators in the committee.
Stereotypes An additional theoretical argument in favor of a higher female pres- ence in evaluation committees is that senior male researchers might have stereotypes against female candidates. If senior female researchers do not share these stereo- types, having more women on the committee might reduce the impact of gender prejudices.
Stereotyping might be stronger when evaluators are less informed about candi- dates’ quality. Given that it might be particularly difficult to assess the quality of candidates who do research in subfields that lie far away from evaluators’ knowledge, we divide evaluations in two groups based on the distance between evaluators’ and candidates’ research interests. The evidence suggests that information asymmetries
25
matter, but the presence of women in the committee does not contribute to elimi- nate potential gender biases. When candidates and evaluators work in similar areas, evaluators’ gender does not have a significant impact (Table 7, first row). However, when candidates do research in a different subfield, female candidates tend to per- form significantly worse when there are relatively more women in the committee. This pattern is observed in both countries.
It is also sometimes argued that stereotyping against women is stronger in sci- ences and mathematics-related disciplines (Reuben, Sapienza, and Zingales 2014). We compare the effect of female evaluators in STEMM and SSH disciplines, but we do not observe any significant differences between these two groups neither in Spain nor in Italy (Table 7, second row).
One might also expect prejudices against women to be stronger in disciplines that are less feminized and, therefore, offer fewer chances to interact with female researchers. We examine separately disciplines with a relatively low and a relatively high proportion of women among full professors. We do not find any evidence suggesting that evaluators in these two groups differ in terms of their preference for candidates of the same sex (Table 7, third row).25
High-level positions The impact of committees’ gender composition might also depend on the importance of the position at stake. Some male evaluators might be reluctant to see a female colleague at the top of the academic career ladder. They might hold negative stereotypes of women, for instance, regarding their leadership or other abilities specific to full professor positions. There might also be a problem of taste-discrimination.
We examine separately the effect of female presence upon the evaluation com- mittee for candidates to full and associate professor positions (Table 7, fourth row). We do not observe any significant differences between these groups of evaluations in Italy, but we do observe a significant difference between exams for full and associate professorships in Spain. Specifically, it appears that in Spain, in committees assess- ing candidates to full professor positions, a higher female presence has a positive impact on female candidates’ relative chances of success. However, the opposite is true in evaluations for promotion to more junior positions.
So, in the case of promotions to full professorships in Spain, but not in Italy, the result is consistent with the existence of stereotypes, or even of taste discrimination, against women by committees with low or no representation of women.
25In Table H1 in Appendix H we report results from an alternative specification of heterogeneity tests. Instead of splitting the sample in two groups based on the overlap of candidates’ and evaluators’ research interests and on the degree of feminization of the discipline, we estimate a model with triple interactions exploiting the full range of possible values of these variables. Results from these alternative specifications are in line with the findings discussed in this section.
26
Analysis by disciplinary groups Beyond these theories, it might be that the gender composition of committees matters in some specific fields. The previous empirical literature of evaluators’ gender does not provide a clear pattern. Two articles that study the role evaluators’ gender in Science and Economics find that evaluators tend to prefer candidates of the same sex (Casadevall and Handelsman 2013 and De Paola and Scoppa 2015), but in two other studies conducted in the same disciplines evaluators exhibit a preference for candidates of the other sex (Broder 1993, Ellemers et al. 2004). Six other articles in different fields do not find any significant relationship.26
Following the official classification of disciplines adopted by the Italian Ministry, we consider 16 different groups of disciplines: Industrial Engineering, Civil Engi- neering, Physics, Mathematics, Chemistry, Geology, Biology, Veterinary, Medicine, Psychology, Architecture, Economics and Business, Social Sciences, History, Lan- guages and Law. We estimate equation (4) separately for each group and each country, including evaluation fixed effects and instrumenting the final composition of the committee with the initial one. We report these estimates in Figure 1. Out of 32 coefficients, 28 are not significant, one is significantly positive and three are significantly negative. When we take into account in the calculation of standard errors that we are running multiple regressions using a Bonferroni correction none of the coefficients remains significant. Altogether, it is not possible to reject that the impact is similar to zero in any of the different samples. Similarly, we cannot reject the hypothesis that the effect is similar across different fields.
4.6.2 Why does the presence of women in the committee affect the voting behavior of male evaluators?
There are at least three potential explanations. The presence of women in the com- mittee might unleash a backlash against female candidates, particularly in fields that have been historically dominated by men (Crocker and McGraw 1984). While we cannot directly test this hypothesis, we do not observe any significant differ- ence in the impact of committees’ gender composition depending on the degree of feminization of the field (see Table 7, third row).
The presence of female evaluators might also induce a licensing effect (Monin and Miller 2001). In all-male committees, evaluators may feel that they have a moral obligation to worry about sexism and seek to overcome it by expressing more positive (and perhaps less discriminatory) views about female candidates. When there are women on a committee, men may feel licensed to express more honest opinions about female candidates. Furthermore, female evaluators might strengthen male identities
26See more details in Table A1.
27
within committees and hence weaken their support for female candidates (Akerlof and Kranton 2000). Unfortunately, we cannot disentangle empirically these two competing hypotheses, licensing effect and male identity priming.
5 Conclusions
A larger presence of women in scientific committees is frequently defended in policy discussions. This paper contributes to this debate by providing a comprehensive and systematic analysis of the impact of scientific committees’ gender composition. We exploit the exceptional evidence provided by qualification evaluations for full and associate professorships in every discipline in two different countries, Italy and Spain. These evaluations involved around 100,000 applications and 8,000 evaluators in all academic fields. The random assignment of evaluators to committees creates a setting of large-scale natural randomized experiments. We also take advantage of the availability of very detailed information about candidates, evaluators and the content of evaluations, in order to analyze explicitly the theoretical arguments that are usually employed in support of a higher representation of women in scientific committees.
In general, the presence of female evaluators in the committee neither increases the success rate of female candidates, nor does it alter the quality of selected can- didates. Strikingly, in all but one subsamples we observe the opposite pattern in success rates: committees with a higher women share tend to be relatively less favorable towards female candidates. The only exception refers to evaluations to full professorships in Spain, where female candidates have better chances of success when evaluated by a committee with more women.
Information from individual votes within committees suggests that there are two factors that explain why a larger presence of women does not increase the success rate of female candidates. First, while female committee members are slightly more favorable towards female candidates than their male colleagues, this effect is not economically or statistically significant. Second, male evaluators become less favor- able towards female candidates when women are present in the committee, perhaps due to a licensing effect or to male identity priming.
Two common arguments that are usually employed in support of a higher rep- resentation of women in scientific committees – gendered networks and segregation across subfields – do not play an important role in our data. We document the ex- istence of gender segregation across research networks in both countries. A female candidate is significantly more likely to be connected to a female evaluator, as mea- sured by coauthorships, affiliation, doctoral thesis supervision and participation in theses committees. We also observe that committees tend to favor connected can-
28
didates. However, in the nation-wide evaluations that we consider in this paper the likelihood of connections between candidates and evaluators is small and, therefore, the impact of gendered networks on evaluations is very modest. We also find that evaluators have a preference for candidates with similar research interests but the extent of gender segregation within each field is relatively small. As a result, the impact of gender segregation on evaluation outcomes is very limited. Another justi- fication for increasing the presence of women in committees is that male evaluators may hold stereotypes that have a negative effect upon female candidates. In order to explore the potential impact of gender stereotypes, we focus on cases where in- formation asymmetries are expected to be important. Our results indicate that the gender of evaluators only matters when evaluators are not familiar with candidates’ research. However, in this case gender-mixed committees are less favorable towards women than all-male committees.
It remains an open question how the specific institutional characteristics of the Italian and the Spanish promotion systems affect the role of committees’ gender composition. Overall, we cannot reject that the estimates for both countries are statistically similar, but we observe a significant difference in the behavior of com- mittees evaluating applications to full professor positions. In Italy, a larger presence of men in the committee increases the chances of success of female applicants. On the contrary, in Spain female applicants to full professorships tend to be relatively less successful when evaluated by an all-male committee. It is unclear whether this difference reflects random sampling or whether it captures some institutional or country-specific characteristic.27
Our analysis may be relevant for the design of policies aimed at increasing the representation of women in the academic career. Several countries, including Spain, have introduced quotas in scientific committees requiring the presence of a minimum share of male and female evaluators. According to our results, in general, a higher representation of women in scientific committees per se does not increase the number of promoted female candidates, nor does it help candidates who prove to be more productive in the future. Introducing gender quotas indiscriminately might also have unintended consequences. Quotas may be detrimental for senior female researchers, who would have to spend a disproportionate amount of time sitting on committees and, in some cases, for junior ones, whose chances of success may be hindered.
To be sure, gender quotas could be desirable in certain cases. The analysis suggests that the prevalence of gender segregation across subfields might be an im- portant determinant of whether female committee representation is likely to help
27Some authors have argued that the degree of transparency in an evaluation procedure can affect gender biases (van den Brink, Benschop and Jansen 2010). Hence, one possible explanation is that the higher level of transparency and public scrutiny of the Italian system deterred male evaluators from discriminating against female applicants to full professor positions.
29
female candidates. We expect gender segregation to play a more important role when evaluations are held at a more aggregate level than the one considered here.28 Another important factor is the potential existence of connections between evalu- ators and candidates. These connections, which tend to be gendered, are likely to be more relevant in committees at the university- or department-level. More em- pirical work is needed to understand the impact of gender quotas in those contexts. Moreover, there are certain features of gender quotas that are not captured by our analysis. Evaluators who are explicitly chosen to represent a minority might be- have differently, perhaps being more inclined to take a positive view of candidates belonging to their own group. The introduction of quotas may also affect the strate- gic incentives of evaluators. Nonetheless, keeping in mind these limitations, our results cast doubts on a generalized implementation of gender quotas in scientific committees.
References
Abe, Yukiko (2012), “The Academic Labor Market in Japan and the Presence of Women,” CSWEP Newsletter, Fall, pp. 9-10.
Abrevaya, Jason and Daniel S. Hamermesh (2012), “Charity and Favoritism in the Field: Are Female Economists Nicer (to Each Other)?” Review of Economic and Statistics, Vol. 94(1), pp. 202-207.
Akerlof, George A. and Rachel E. Kranton (2000), “Economics And Identity,” The Quarterly Journal of Economics, MIT Press, Vol. 115(3), pp. 715-753.
Babcock, Linda, Michele Joy Gelfand, Deborah Small, and Heidi Stayn (2006), “Propensity To Initiate Negotiations: A New Look At Gender Variation In Negotiation Behavior,” in David De Cremer, Marcel Zeelenberg, and J. Keith Murnighan (Eds.) Social Psychology and Economics. Mahwah, NY: Lawrence Erlbaum Associates, pp. 239-259.
Bagilhole, Barbara (2005) “Address at the NORFACE gender equality workshop,” Reykjavik.
Bagues, Manuel and Berta Esteve-Volart (2010), “Can Gender Parity Break the Glass Ceiling? Evidence from a Repeated Randomized Experiment,” Review of Economic Studies, Vol. 77(4), pp. 1301-1328.
Bagues, Manuel and Maria Jose Perez-Villadoniga (2012), “Do Recruiters Prefer Applicants With Similar Skills? Evidence From a Randomized Natural Ex- periment,” Journal of Economic Behavior and Organization, Vol. 82(1), pp. 12-20.
28The level of disaggregation at which scientific evaluations are held varies largely across countries and institutions. For instance, the European Research Council groups applications in 25 broadly defined areas (http://erc.europa.eu/evaluation-panels, accessed on September 1 2015), while in the National Institutes of Health (NIH), which considers only life sciences, grant applications are evaluated by 174 different “study sections” (http://public.csr.nih.gov/StudySections/ Standing/Pages/default.aspx, accessed on September 1 2015).
30
Bagues, Manuel and Maria Jose Perez-Villadoniga (2013), “Why Do I Like People Like Me?” Journal of Economic Theory, Vol. 148(3), pp. 1292-1299.
Bagues, Manuel, Mauro Sylos-Labini and Natalia Zinovyeva (2015), “Connec- tions in Scientific Committees and Applicants’ Self-Selection: Evidence from a Natural Randomized Experiment,” IZA Discussion Paper, No. 9594.
Bagues, Manuel, Mauro Sylos-Labini and Natalia Zinovyeva (2014), “Do Gender Quotas Pass the Test? Evidence from Academic Evaluations in Italy,” LEM Working paper 2014/14.
Barres, Ben (2006), “Does gender matter?” Nature, Vol. 442, pp. 133-136.
Bertrand, Marianne, Sandra E. Black, Sissel Jensen and Adriana Lleras-Muney (2014), “Breaking the Glass Ceiling? The Effect of Board Quotas on Female Labor Market Outcomes in Norway,” NBER Working Paper 20256.
Blackaby, David, Alison L. Booth and Jeff Frank (2005), “Outside Offers And The Gender Pay Gap: Empirical Evidence From the UK Academic Labour Market,” Economic Journal, Vol. 115(501), pp. F81-F107.
Blau, Francine D., Janet M. Currie, Rachel T. A. Croson and Donna K. Ginther (2010), “Can Mentoring Help Female Assistant Professors? Interim Results from a Randomized Trial,” American Economic Review, Vol. 100(2), pp. 348- 352.
Bloom, Nicholas, Mark Schankerman and John Van Reenen (2013), “Identifying Technology Spillovers and Product Market Rivalry,” Econometrica, Vol. 81(4), pp. 1347-1393.
Bohnet, Iris, Alexandra van Geen and Max Bazerman (2015), “When Perfor- mance Trumps Gender Bias: Joint Versus Separate Evaluation,” Management Science, Vol. 62(5), pp. 1225-1234.
Booth, Alison L. and Andrew Leigh (2010), “Do Employers Discriminate by Gen- der? A Field Experiment in Female-Dominated Occupations,” Economic Let- ters, Vol. 107(2), pp. 236-238.
Boschini, Anne and Anna Sj ̈ogren (2007), “Is Team Formation Gender Neu- tral? Evidence from Coauthorship Patterns,” Journal of Labor Economics, Vol. 25(2), pp. 325-365.
Bosquet, Cl ́ement, Pierre-Philippe Combes, Cecilia Garcia-Pen ̃alosa (2013), “Gender and Competition: Evidence from Academic Promotions in France,” CESifo Working Paper, No. 4507.
Brescoll, Victoria L. (2011), “Who Takes the Floor and Why: Gender, Power, and Volubility in Organizations,” Administrative Science Quarterly, Vol. 56(4), pp. 622-641.
Broder, Ivy E. (1993), “Review of NSF Economics Proposals: Gender and Insti- tutional Patterns,” American Economic Review, Vol. 83(4), pp. 964-970.
Buser, Thomas, Muriel Niederle and Hessel Oosterbeek (2014), “Do Women Shy away from Competition? Do Men Compete too Much?” Quarterly Journal of Economics, Vol. 129(3), pp. 1409-1447.
31
Casadevall, Arturo and Jo Handelsman (2014), “The Presence of Female Con- veners Correlates with a Higher Proportion of Female Speakers at Scientific Symposia,” mBio, Vol. 5(1), e00846-13.
Ceci, Stephen J., Donna K. Ginther, Shulamit Kahn, and Wendy M. Williams (2014), “Women in Academic Science: A Changing Landscape,” Psychological Science in the Public Interest, Vol. 15(3), pp. 75-141.
Ceci, Stephen J. and Wendy M. Williams (2011), “Understanding Current Causes of Women’s Underrepresentation in Science,” Proceedings of the National Academy of Sciences, Vol. 108(8), pp. 3157-3162.
Crocker, Jennifer, and Kathleen M. McGraw (1984) “What’s Good for the Goose is not Good for the Gander,” American Behavioral Scientist, Vol. 27 (3), pp. 357-369.
De Paola, Maria and Vincenzo Scoppa (2015), “Gender Discrimination and Evalu- ators’ Gender: Evidence from the Italian Academy,” Economica, Vol. 82 (325), pp. 162-188.
De Paola, Maria, Michela Ponzo and Vincenzo Scoppa (2015), “Gender Differ- ences in Attitudes Towards Competition: Evidence from the Italian Scientific Qualification,” IZA Discussion Paper, No. 8859.
Dolado, Juan Jose, Florentino Felgueroso and Miguel Almunia (2012), “Are Men and Women-Economists Evenly Distributed across Research Fields? Some New Empirical Evidence,” SERIEs: Journal of the Spanish Economic Associ- ation, Vol. 3(3), pp. 367-393.
Ellemers, Naomi, Henriette Van den Heuvel, Dick de Gilder, Anne Maass, and Alessandra Bonvini (2004), “The Underrepresentation of Women in Science: Differential Commitment or the Queen Bee Syndrome?” British Journal of Social Psychology, Vol. 43 (3), pp. 315-338.
European Commission (2008), Mapping the Maze: Getting More Women to the Top in Research, Luxembourg: Publications Office of the European Union.
European Commission (2016), She Figures 2015: Gender in Research and Inno- vation, Luxembourg: Publications Office of the European Union.
Fundaci ́on Espan ̃ola para la Ciencia y la Tecnolog ́ıa (2005), Mujer y Ciencia: La situaci ́on de las Mujeres Investigadoras en el Sistema Espan ̃ol de Ciencia y Tecnolog ́ıa, Madrid: FECYT.
Ginther, Donna K. and Shulamit Kahn (2004), “Women in Economics: Moving Up or Falling Off the Academic Ladder,” Journal of Economic Perspectives, Vol. 18(3), pp 193-214.
Ginther, Donna K. and Shulamit Kahn (2009), “Does Science Promote Women? Evidence from Academia 1973-2001,” NBER chapters in: Science and Engi- neering Careers in the United States: An Analysis of Markets and Employ- ment, pp. 163-194.
Hale, Galina and Tali Regev (2014), “Gender Ratios at Top PhD Programs in Economics,” Economics of Education Review, Vol. 41, pp. 55-70.
Hilmer, Christiana and Michael Hilmer (2007), “Women Helping Women, Men Helping Women? Same-Gender Mentoring, Initial Job Placements, and Early
32
Career Publishing Success for Economics PhDs,” American Economic Review, Vol. 97(2), pp. 422-426.
Jaffe, Adam B. (1986), “Technological Opportunity and Spillovers of R&D: Evi- dence from Firms’ Patents, Profits, and Market Value,” American Economic Review, Vol. 76(5), pp. 984-1001.
Jayasinghe, Upali W., Herbert W. Marsh and Nigel Bond (2003), “A Multilevel Cross-Classified Modelling Approach To Peer Review Of Grant Proposals: The Effects Of Assessor And Researcher Attributes On Assessor Ratings,” Journal of the Royal Statistical Society: Series A (Statistics in Society), Vol. 166(3), pp. 279-300.
Karpowitz, Christopher F., Tali Mendelberg and Lee Shaker (2012), “Gender Inequality in Deliberative Participation,” American Political Science Review Vol. 106(3), pp. 533-547.
Khan, Uzma and Ravi Dhar (2006), “Licensing Effect in Consumer Choice,” Jour- nal of Marketing Research, Vol. 43(2), pp. 259-266.
Kunze, Astrid and Amalia R. Miller (2014), “Women Helping Women? Evidence from Private Sector Data on Workplace Hierarchies,” IZA Discussion Paper, No. 8725.
M ́endez, Mar ́ıa J and John R. Busenbark (2015), “Shared Leadership And Gender: All Members Are Equal… But Some More Than Others”, Leadership and Organization Development Journal, Vol. 36(1), pp. 17-34.
Milkman, Katherine L., Modupe Akinola and Dolly Chugh (2015), “What Hap- pens Before? A Field Experiment Exploring How Pay and Representation Differentially Shape Bias on the Pathway Into Organizations,” Journal of Ap- plied Psychology, Vol. 100(6), pp. 1678-1712.
Monin, Benoˆıt and Dale T. Miller (2001), “Moral Credentials and the Expression of Prejudice,” Journal of Personality and Social Psychology, Vol. 81(1), pp. 33-43.
Moss-Racusin, Corinne A., John F. Dovidio, Victoria L. Brescoll, Mark J. Gra- ham and Jo Handelsman (2012), “Science Faculty’s Subtle Gender Biases Fa- vor Male Students,” Proceedings of the National Academy of Sciences, Vol. 109(41), pp. 6474-16479.
National Research Council (2009), Gender Differences at Critical Transitions in the Careers of Science, Engineering, and Mathematics Faculty, Washington D.C.: The National Academies Press.
Niederle, Muriel and Lise Vesterlund (2007), “Do Women Shy Away from Compe- tition? Do Men Compete Too Much?” Quarterly Journal of Economics, Vol. 122 (3), pp. 1067-1101.
Reuben, Ernesto, Paola Sapienza, and Luigi Zingales (2014), “How Stereotypes Impair Women’s Careers in Science,” Proceedings of the National Academy of Sciences, Vol. 111 (12) 4403-4408, doi: 10.1073/pnas.1314788111.
Sa ́nchez de Madariaga, In ́es, Sara de la Rica and Juan Jos ́e Dolado (coord.) (2011), “White Paper on the Position of Women in Science in Spain,” Ministry of Science and Innovation.
33
Sandberg, Anna (2014), “Competing Biases: Effects of Gender and Nationality in Sports Judging,” Stockholm School of Economics, mimeo.
Smith, Kristin A., Paola Arlotta, Fiona M. Watt, The Initiative on Women in Science and Engineering Working Group, and Susan L. Solomon (2015), “Seven Actionable Strategies for Advancing Women in Science, Engineering, and Medicine,” Cell Stem Cell, Volume 16(3), pp. 221-224.
Steinpreis, Rhea E. and Katie A . Anders, and Dawn Ritzke (1999), “The Impact of Gender on the Review of the Curricula Vitae of Job Applicants and Tenure Candidates: A National Empirical Study,” Sex Roles, Vol. 41(7/8), pp. 509- 528.
van den Brink, Marieke, Yvonne Benschop and Willy Jansen (2010), “Trans- parency in Academic Recruitment: A Problematic Tool for Gender Equality?” Organization Studies, Vol. 31 (11), pp. 1459-1483.
Vernos, Isabelle (2013), “Research Management: Quotas Are Questionable,” Na- ture, Vol. 495(7439), pp. 39.
Vesterlund, Lise, Linda Babcock, and Laurie Weingart (2014), “Breaking the Glass Ceiling with “No”: Gender Differences in Declining Requests for Non- Promotable Tasks,” mimeo.
Williams, Wendy M. and Stephen J. Ceci (2015), “National Hiring Experiments Reveal 2:1 Faculty Preference for Women on Stem Tenure Track,” Proceedings of the National Academy of Sciences, Vol 112 (17), pp. 5360-5365.
Woolley, Anita Williams, Christopher F. Chabris, Alex Pentland, Nada Hashmi, and Thomas W. Malone (2010), “Evidence for a Collective Intelligence Factor in the Performance of Human Groups,” Science, Vol. 330, pp. 686-688.
Zinovyeva, Natalia and Manuel Bagues (2011), “Does Gender Matter for Aca- demic Promotion? Evidence from a Randomized Natural Experiment,” IZA Discussion Paper, No. 5537.
Zinovyeva, Natalia and Manuel Bagues (2015), “The Role of Connections in Aca- demic Promotions,” American Economic Journal: Applied Economics, Vol. 7(2), pp. 264-292.
34
Table 1: The causal impact of committees’ gender composition
Dependent variable:
Female candidate
Share of women in committee
Female candidate* Share of women in committee
Adj. R-squared Number of observations
Mean dep. var. (for men)
Female candidate
Share of women in committee
Female candidate* Share of women in committee
Adj. R-squared Number of observations
Mean dep. var. (for men)
Controls for both panels:
Candidate characteristics ExamFE
Expected share of women Female candidate* Expected share of women
Committee characteristics
1234567 Qualified Applied
OLS OLS ITT IV IV IV IV
-0.028*** -0.015*** -0.004 (0.006) (0.005) (0.009)
Italy
0.001 0.008 0.009 -0.026*** (0.011) (0.008) (0.007) (0.006) -0.0004— (0.071)
-0.116** -0.128*** -0.132*** -0.025 (0.050) (0.035) (0.036) (0.026)
0.000 (0.059) -0.092** (0.036)
0.245 69020
0.38
-0.022***
(0.004) (0.004) (0.007)
0.001 69020
0.38
0.240 69020
0.38
0.245 69020
0.38
Spain
-0.009
(0.007) (0.007) (0.007)
0.012 – – (0.018)
-0.019 -0.016 -0.029 (0.027) (0.028) (0.028)
0.001 31243
0.12
0.036 31243
0.12
0.011 (0.017) -0.018 (0.026)
0.039 31243
0.12
Yes
Yes Yes
0.039 31243
0.12
Yes
Yes Yes
0.005 0.005 31243 31243
0.12 0.12
Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Yes
0.236 69020
0.38
0.236 69020
0.075 69020
0.38 0.87
-0.014***
-0.009
-0.011
-0.009
Yes Yes Yes
Notes: Candidate characteristics include all individual predetermined characteristics listed in Table C2. Committee characteristics include the interaction between candidates’ gender and the average tenure of evaluators (Italy only), their age (Spain only), their quality-adjusted productivity during the previous 10 years, and the proportion of com- mittee members based in the South. The first-stage results for the IV estimations reported in columns 4 and 5 are available in Table D1. Standard errors are clustered by committee.
* p < 0.10, ** p < 0.05, *** p < 0.01.
35
36
Dependent variable:
All Publ.
Articles
Books
Chapters
Patents
Total A-journal AIS articles
Coauthors per article
Prop. first-author
Prop. last-author
Age
Share of women in committee
0.005
Female candidate*Share of women in committee
-0.027 (0.079)
-0.001 (0.071)
Share of women in committee
-0.029
Female candidate*Share of women in committee
0.015 0.038 (0.077) (0.078)
0.014
(0.034) (0.031)
-0.023 (0.031) 0.059 (0.066)
-0.020 (0.028) 0.048 (0.063)
0.019 (0.021) -0.040 (0.049)
-0.005 (0.030) 0.018 (0.064)
0.038 (0.027) -0.087 (0.061)
0.017 (0.031) -0.044 (0.067)
-0.040* (0.023) 0.093* (0.053)
-0.011 (0.041) 0.031 (0.088)
-0.065* (0.035) 0.150* (0.080)
-0.019
(0.030) (0.031)
0.004 (0.022) -0.017 (0.054)
0.002 (0.022) -0.010 (0.055)
0.024 (0.015) -0.062 (0.038)
-0.068** (0.027) 0.152** (0.068)
-0.023 (0.022) 0.043 (0.057)
-0.040 (0.030) 0.103 (0.076)
-0.020 (0.031) 0.045 (0.078)
-0.023 (0.032) 0.042 (0.080)
0.034 (0.034) -0.093 (0.086)
Table 2: Randomization check
1 2 3 4 5 6 7 8 9 10 11
Notes: OLS estimates. All regressions include also the variables Female candidate, Expected share of women in committee, and the interaction between the two. Standard errors are clustered by committee.
* p < 0.10, ** p < 0.05, *** p < 0.01.
Italy
Spain
Dep. var.:
All Women Men
All Women Men
All Women Men
Publications
0.017 (0.088) -0.044 (0.112) 0.029 (0.101)
0.022 (0.145) 0.210 (0.206) -0.124 (0.193)
0.016 (0.132) 0.345 (0.213) -0.187 (0.182)
Citations
0.130 (0.117) 0.139 (0.119) 0.098 (0.150)
0.072 (0.223) 0.469 (0.370) -0.242 (0.291)
-0.060 (0.218) -0.009 (0.356)
Total AIS A-journal PhD students articles advised
A. Italy, before the evaluation
PhD thesis committees
-0.147 (0.132) 0.053 (0.220) -0.303* (0.168)
-0.086 (0.136) -0.117 (0.231) -0.134 (0.186)
Success in future evaluations
Table 3: Quality of qualified candidates
1234567
-0.055 (0.157) 0.154 (0.170) -0.208 (0.211)
-0.135 (0.255) -0.102 (0.317) -0.213 (0.251)
B. Spain, before the evaluation -0.088 -0.200 0.125
(0.244) (0.237) (0.136) -0.004 -0.142 0.580** (0.399) (0.329) (0.229)
-0.215 -0.219 -0.170 (0.301) (0.333) (0.176)
C. Spain, after the evaluation -0.098 -0.173 0.175
(0.227) (0.181) (0.135) -0.102 0.170 0.119 (0.376) (0.288) (0.212)
0.042 (0.052) 0.001 (0.054) 0.019 (0.077)
-0.140 -0.247 -0.266 0.080
(0.281)
(0.284) (0.252) (0.191)
Notes: OLS estimates for the sample of qualified candidates. Each coefficient corresponds to an independent regression for a given sample and dependent variable. In panels A and B the dependent variables are measured at the time of the evaluation. In panel C the dependent variables refer to the output in the five- year period following the evaluation. Success in future evaluations takes value one if a candidate who obtained a qualification for an associate professorship in our sample, qualifies in the evaluation for full professorship by year 2013. The dependent variables in columns 1-6 are normalized to have zero mean and unit variance for candidates within each exam. Citations and Article Influence Score are only available for candidates in science, technology, engineering, mathematics, medicine and psychology. Information on publications in A-journals is only provided for candidates in social sciences and humanities. All regressions include non- parametric controls for expected share of women in the committee, disciplinary area*rank, and age. Standard errors are clustered by committee.
* p < 0.10, ** p < 0.05, *** p < 0.01.
37
Female candidate
Female evaluator
Female candidate * Female evaluator
Share of women in committee
Female candidate*Share of women in committee
Controls:
Application FE
Expected share of women
Female candidate*Expected share of women Candidate characteristics
Exam FE
Number of observations
-0.0004 0.008 (0.009) (0.006)
-0.003 (0.006)
Table 4: Individual voting
Notes: OLS estimates. The dependent variable is an indicator that takes value one if the evaluator casted a positive vote for a given candidate. Column 1 includes information from all individual evaluations, columns 2-4 include information only on evaluations by male evaluators. In column 4 we also include applications that were withdrawn after committee composition was announced, imputing a negative assessment to these applications. Candidate characteristics include all predetermined characteristics listed in Table C2. Standard errors are clustered by committee.
* p < 0.10, ** p < 0.05, *** p < 0.01.
38
1234
All evaluators -
-0.001 (0.007) 0.007 (0.005)
Yes
294,656
Male evaluators
0.017 - - (0.079)
-0.042 -0.061** (0.043) (0.030)
-0.078*** (0.030)
Yes Yes Yes Yes Yes Yes Yes Yes Yes
Yes Yes
240988 240988
281289
39
Female candidate Female evaluator
0.0026*** (0.0004) 0.0017* (0.0009) 0.0029*** (0.0007)
0.0007** (0.0003) -0.0015*** (0.0004) 0.0022*** (0.0005) 0.0140*** (0.0001)
0.0209*** (0.0060) -0.0067 (0.0075) 0.0133*** (0.0045) 0.5897*** (0.0029)
-0.0012 (0.0014) 0.0006 (0.0014) 0.0043*** (0.0016) 0.0453*** (0.0007)
-0.0003*
-0.0001 (0.0001) -0.0013*** (0.0002) 0.0005*** (0.0002) 0.0025*** (0.0000)
-0.0010*** (0.0003) -0.0047*** (0.0006) 0.0013*** (0.0005) 0.0142*** (0.0002)
0.0065** (0.0028) -0.0110*** (0.0017) 0.0042* (0.0022) 0.1959*** (0.0010)
Female candidate*
Female evaluator
Constant 0.0262***
(0.0002) -0.0015*** (0.0002) 0.0010*** (0.0002) 0.0045*** (0.0000)
Observations 2,555,839
2,555,839
1,373,825
5,445,067
5,445,067
5,445,067
5,445,067
4,711,621
Colleague
Coauthor
Same subfield
Colleague
Coauthor
PhD Advisor
PhD committee
Research overlap
(0.0002)
Table 5: Gender segregation across research networks and subfields
12345678
Italy
Spain
Notes: OLS estimates. The number of observations corresponds to the number of possible pairs between candidates and eligible evaluators with non-missing information in a given exam. In Italy, only evaluators who are based in an Italian university are considered. All regressions include evaluation fixed effects. Standard errors are clustered by field.
* p < 0.10, ** p < 0.05, *** p < 0.01.
40
Female candidate
Female candidate * Share of female evaluators
0.008 (0.008) -0.128*** (0.035)
0.006 (0.007) -0.124*** (0.034)
-0.008 (0.009) -0.061 (0.046)
-0.010 (0.009) -0.060 (0.046)
-0.011 (0.007) -0.016 (0.028)
-0.010 (0.007) -0.020 (0.028)
-0.011 (0.008) -0.017 (0.035)
-0.011 (0.008) -0.021 (0.035)
Connections in committee: Colleagues
0.181*** (0.036) 0.237*** (0.048)
0.180*** (0.044) 0.201*** (0.053)
0.319*** (0.031) 0.869*** (0.140) 0.633*** (0.107) 0.174*** (0.037)
0.319*** (0.031) 0.840*** (0.142) 0.575*** (0.115) 0.166*** (0.038)
Coauthors
PhD advisors
PhD thesis committee
Research similarity: Same subfield
0.046 (0.032)
Overlap in research interests
0.124*** (0.037)
Controls :
Expected connections
Expected same subfield
Expected overlap in research interests Number of observations
Yes
Yes Yes
Yes
Yes
Table 6: Connections and research similarity
12345678
69020
69020
35832
35832
31243
31243
27998
Yes 27998
Notes: IV estimates. All regressions include exam fixed-effects, an interaction between Female candidate and the Expected share of women in committee, and controls for all individual predetermined characteristics listed in Table C2. Connection variables are measured in shares. PhD thesis committee refers to candidates and evaluators who have been members of the same doctoral thesis committee. Same subfield is the share of evaluators who belong to the same subfield (settore scientifico disciplinario) as the candidate. Overlap in research interests is based on evaluators’ and candidates’ participation in doctoral thesis committees, which are classified in 2,000 different subfields (see more details in Data section). Expected connections is a vector including the expected share in the committee of colleagues, coauthors, advisors and PhD thesis committee. Standard errors are clustered by committee.
* p < 0.10, ** p < 0.05, *** p < 0.01.
Italy
Spain
Table 7: Heterogeneity analysis
1234
Italy
Spain
< median
-0.125*** (0.044)
STEMM
0.004 (0.041)
< median
-0.016 (0.037)
AP
-0.072** (0.032)
Research overlap
Discipline
Feminization of field
Level of promotion
≥ median 0.011
(0.046) SSH
-0.116** (0.054)
≥ median -0.149***
(0.042)
FP
-0.111* (0.059)
< median
-0.179*** (0.069)
STEMM
-0.128*** (0.034)
< median
-0.072 (0.057)
AP
-0.138*** (0.038)
≥ median 0.081*
(0.047) SSH
-0.026 (0.038)
≥ median -0.018
(0.040)
FP
0.120** (0.054)
Notes: IV estimates. The dependent variable is a dummy variable that takes value one if the candidate qualified. Each coefficient corresponds to an independent regression for the corresponding sample. Research overlap is a proportion of committee members with similar research interest as defined in section 3.4. SSH stands for social sciences and humanities, and STEMM for science, technology, engineering, mathematics, medicine, and psychology. Feminization of the field is measured by the proportion of women among full professors in the discipline. FP and AP stand, respectively, for full and associate professors. Standard errors are clustered by committee.
* p < 0.10, ** p < 0.05, *** p < 0.01.
41
42
Figure 1: The causal impact of committees’ gender composition, by disciplinary group
3
2
1
0
-1
-2
civil eng 15 7
arch
geo 28 4
soc 18 7
psych
vet 39 14
phys
chem
math
hist 110 13
med 93 26
bio 85 13
econ 64 15
law 83 16
lang 139 19
ind eng
126 20 319 1443 352 2729 367 1377 537 2510 764 1616 972 2439 1458 4778 1679 2841 1823 2999 2480 5194 2765 11363 2802 7169 3066 6005 3455 4211 3624 6976 4780 5370
24 5
30 4
36 6
42 8
Spain Italy
Note: The figure reports the effect of a higher proportion of women among evaluators on the relative success rate of female candidates in the corresponding disciplinary group and country. The confidence intervals are not adjusted for multiple comparisons. At the bottom of the figure, the number of committees and the number of candidates in a corresponding group are shown. The disciplinary groups are sorted according to the number of applicants in each group in Spain.
35 7
For Online Publication Appendix A. Summary of the literature
In this section we present a brief summary of the literature that analyzes how the gender composition of academic committees affects the relative success rate of female candidates. Table A1 summarizes the existent studies along a number of key dimensions: the type of analyzed evaluation, the field in which the analyzed evaluation took place, the empirical method used to identify the causal impact of the committee composition on the evaluation outcome, the size of the sample as measured by the number of applications evaluated by committees, and the main result of each study.
43
44
Paper
Type of evaluation
Field
Empirical method
Applications Results
Broder (1993)
Steinpreis, Anders and Ritzke (1999) Jayasinghe, Marsh and Bond (2003) Ellemers et al. (2004)
Milkman, Akinola and Chugh (2015) Moss-Racusin et al. (2012)
Abrevaya and Hamermesh (2012) Casadevall and Handelsman (2013) De Paola and Scoppa (2015) Williams and Ceci (2015)
Grant applications
Job applicants and
Grant applications
Work commitment
Prospective students request a meeting Laboratory manager position
Economics
Psychology
Several
Several
Several
Life Sciences
Economics
Microbiology
Economics and Chemistry Several
Application fixed effects Randomized field experiment Application fixed effects Identification based on observables Randomized field experiment Randomized field experiment Application fixed effects Identification based on observables Identification based on observables Randomized field experiment
1,479 Opposite-sex preference 238 No significant difference 2331 No significant difference 212 Opposite-sex preference 6,548 No significant difference 127 No significant difference 2,940 No significant difference 1,845 Same-sex preference 2,279 Same-sex preference 873 No significant difference
Paper submitted for publication Selection of conference speakers Job applicants
Job applicants
Table A1: Summary of the literature
tenure candidates of students
Note: (∗) We classify the empirical strategy followed by De Paola and Scoppa (2015) as identification based on observables and not as randomized natural experiment due to the nature of the empirical strategy implemented by the authors. This paper studies promotions in the Italian university system that was in place between 2008 and 2011. In this system, four out of five members of the evaluation committee were randomly selected from a pool of eligible evaluators. In their analysis, the authors study the relationship between the success rate of male and female candidates and the gender composition of committees, unconditional on the gender composition of the pool of eligible evaluators. The consistency of the estimation relies on the implicit assumption that the relative quality of male and female applicants is unrelated to the degree of feminization of the pool of eligible evaluators.
For Online Publication Appendix B. Institutional background
There are several important differences between the Spanish and the Italian systems of centralized national evaluations. To facilitate the comparison, Table B1 summarizes the main features of the two systems.
45
Table B1: Main features of the evaluation systems in Italy and Spain
Eligibility requirement for can- didates
Size of evaluation committees Assignment to committees Composition of committees
Constraints on randomization
Minimum research quality re- quirement for evaluators
Italy, Abilitazione Scientifica Nazionale, 2012-2014
None
5 evaluators
Based on a random draw
4 full professors based in Italian universi- ties, 1 professor based abroad
No university can have more than one evaluator within a single committee.
In STEMM disciplines, eligible professors should be above the median in their cat- egory and field in at least two of the fol- lowing dimensions: (i) the number of arti- cles published in scientific journals, (ii) the number of citations, (iii) and the H-index. In SSH disciplines, they should be above the median in at least one of the follow- ing dimensions: (i) the number of articles published in high impact scientific jour- nals (so-called A-journals), (ii) the overall number of articles published in any scien- tific journals and book chapters, and (iii) the number of published books. Voluntary
Based on a random draw
Qualified majority of 4 Unlimited
4 years (later extended to 6 years) 2 years application ban
Up until two weeks after the evaluation criteria are publicized
Evaluations are based solely on the ma- terial provided in candidates’ application packages, consisting of CVs and selected publications.
The lists of potential and actual evalua- tors and candidates, as well as the lists of qualified candidates, are published online. Furthermore, the CVs of all participants and individual evaluation reports are pub- lished online. The evaluation agency also collects and publicizes information on the bibliometric indicators of candidates.
Spain, Habilitacio ́n, 2002-2006
None
7 evaluators
Based on a random draw
In full professor exams, 7 full professors based in Spanish universities or public re- search centers. In associate professor ex- ams, 3 full professors and 4 associate pro- fessors.
Only one non-university researcher is al- lowed to be selected as a member of the committee for a given exam. Similarly, only one emeritus professor is allowed to be selected as a member of a given com- mittee.
Eligible associate professors should have one sexenio and eligible full professors should have two sexenios. Sexenios are granted by the Spanish education author- ity on the basis of applicants’ research output in any non-interrupted period of a maximum of six years.
Compulsory
Based on a random draw
Simple majority
Limited by the number of available posi- tions at the university level
Unlimited
None
Candidates can drop out from the process at any time
Oral exams to full professor positions have two qualifying stages. In the first stage, candidates present their CVs. In the sec- ond stage, candidates present a piece of their research work. Exams to associate professor, in addition to these two stages, have an intermediate stage where candi- dates give a lecture on a topic randomly chosen from a syllabus proposed by the candidate.
The lists of potential and actual evalua- tors and candidates, as well as the lists of qualified candidates, are published online.
Inclusion in the pool of eligible evaluators
Substitution of resigned evalu- ators
Voting rule
Number of qualifications granted by the committee Validity of a positive qualifica- tion
Penalization for a negative evaluation
Application withdrawal
Evaluation
Degree of transparency
46
For Online Publication
Appendix C. Data
The data on the participants in Italian evaluations, including the CV of all eligible evaluators and all candidates, was available at the website of the Italian Ministry of Higher Education and Research. We extracted all the individual characteristics that we use in the analysis from these CVs. Information on tenured researchers’ affiliation and the length of tenure was obtained from the Consortium of Italian universities (CINECA). Affiliation of non-tenured researchers is from the most recent publication of the CV.
We also downloaded from the website of the Italian Ministry approximately 295,000 individual evaluation reports, five per each candidate. Due to the data col- lection problem, we are missing information on individual evaluations for 202 can- didates. We are also missing 84 individual evaluation reports in three committees where evaluators abstained whenever there was a conflict of interest. We conducted a text analysis of the available individual evaluation reports and we identified ap- proximately 9,000 different sentences that indicate the evaluator’s decision to fail or to pass a given candidate.
The data on the participants in Spanish evaluations was collected from different sources, including the Spanish Ministry of Research and Science, Thomson Reuters (ISI) Web of Knowledge, the database of publications in Spanish language Dialnet, the European Patent Office and TESEO database on doctoral dissertations.29
Publications indexed in above sources are matched to the list of professors in Spain based on individuals’ names and field of research. This process suffers from an important problem with homonymity since there are lots of common surnames in Spain. In addition to this, bibliographic databases often incompletely record authors’ names (this especially concerns the data on publications before 2010 in the Web of Knowledge). Facing the choice between minimizing the number of false positives or the number of false negatives, we generally preferred the former. This means that, on the one hand, the individuals are authors of the outcomes assigned. On the other hand, we are unable to assign research outputs that have an incomplete record of authors’ names.
Below we describe in detail the process of data collection in the case of Spain.
29We would like to thank St ́ephane Maraut and Catalina Martinez for kindly sharing the data on academic inventors who have patented their inventions in the European Patent Office. For a description of how the patent data was collected and matched to professors, see Maraut and Mart ́ınez (2014), “Identifying Author-Inventors from Spain: Methods and a First Insight into Results,” Scientometrics, Vol. 101, pp. 445-476.
47
Spanish Ministry of Research and Science The Spanish system of centralized examinations known as ‘habilitacio ́n’ was in place between 2002 and 2006. In total, 1,016 exams took place, around five per discipline. We restrict the sample in several ways. We exclude exams where the number of available positions was larger or equal than the number of candidates (two exams, both in Basque Philology) and disciplines where the number of potential evaluators was not large enough to form a committee (55 exams).30 The final database includes 967 exams.
Information on candidates’ and evaluators’ first name, last name, tenure and ID number was retrieved from the website of the Ministry of Research and Science in July 2009 (http://micinn.es). Information on first names allows us to identify gender. In a few cases where it was not possible to assign gender based on first name, we searched online for a personal picture or document that would make it possible to assign gender.
The actual age of individuals is not observable. Instead, we exploit the fact that Spanish ID numbers contain information on their issue date to construct a proxy for the age of native individuals on the basis of his/her national ID number. In Spain, police stations are given a range of ID numbers which are assigned to individuals in a sequential manner. Since it is compulsory for all Spaniards to have an ID number by age 14, two Spaniards with similar ID numbers are likely to be of the same age (andgeographicalorigin).31 Inordertoperformtheassignment,wefirstuseregistry information on the date of birth and ID numbers of 1.8 million individuals in order to create a correspondence table which assigns year of birth to the first four digits of ID number (ranges of 10,000 numbers). To test the precision of this correspondence, we apply it to a publicly available list of 3,000 court clerks, which contains both the ID number and the date of birth. In 95% of the cases the assigned age is within a three-year interval of the actual age. In order to minimize potential errors, whenever our age proxy indicated that a candidates for an associate professorship is less than 27 years old and a candidate for full professorship is less than 35 years old, we assign age a missing value. This proxy is also not defined for non-Spaniards (less than 1% of the sample). We imputed the missing age with the average age of individuals at the same discipline and rank (around 5% of the sample).
In 2006 the system of habilitaci ́on was replaced by a system known as acred- itaci ́on, which is still in place. Under the acreditaci ́on system applicants aspiring
30In these cases, unfilled seats in the committee were filled with professors from related disci- plines.
31There are a number of exceptions. For instance, this methodology will fail to identify the age of individuals who obtained their nationality when they were older than 14. Nevertheless, immigration was a rare phenomenon in Spain until the late 1990s. Additionally, some parents may have their children obtain an ID number before they are 14. This may be the case particularly after Spain entered in the mid 90s the Schengen zone and IDs became a valid documentation to travel to a number of European countries.
48
for promotion are also required to be approved by a national review committee. These committees evaluate candidacies on a monthly basis and their decisions are published in the Official State Bulletin. We collected information on the identity of all candidates that qualified for a FP position before September 2013.
The Ministry provides information on affiliation and on tenure in the position for eligible evaluators. Given that most candidates to full professor positions are eligible evaluators themselves in exams to associate professor positions, it is possible to obtain their affiliation by matching the list of eligible evaluators with the list of candidates. Using this procedure, we were able to obtain the information on affilia- tion for 93% of candidates to full professor positions. We obtained the information on affiliation for the remaining 7% of candidates from the State Official Bulletin or directly from professors’ CVs that can be found online.
ISI Web of Knowledge We also collected information on the research output of eligible evaluators and candidates from the ISI Web of Knowledge.32
Information on scientific publications comes from the Thomson Reuters ISI Web of Knowledge (WoK). We consider publications published since 1972 by authors based in Spain, as well as the number of citations received by these publications before July 2009. The WoK database includes over 10,000 high-impact journals in the categories of Science, Engineering, Medicine and Social Sciences, as well as international proceedings coverage for over 110,000 conferences. For the purpose of this analysis, we considered all articles, reviews, notes and proceedings.
The assignment of articles to professors is non-trivial. For each publication and author, WoS provides information on his/her surname and on his/her initial. In Spain, some surnames are very common (e.g., Garcia, Fernandez, Gonzalez), and this may create problems with homonymity. Moreover, unlike most other countries, individuals are assigned two surnames (paternal and maternal) and sometimes also several first names. When Spanish authors sign a paper they may do it with only their paternal or with their maternal surname, or they may hyphenate the two surnames. Authors may also sign using their first name, their middle name, or both.
We use the following matching procedure in order to deal with the above prob- lems. First, we assign all publications and all professors in our sample to a broad disciplinary category. In order to attribute comparable disciplinary categories for publications and individuals, we aggregate disciplines defined by the Spanish Min- istry and ISI disciplinary areas into the following categories: Agriculture; Chem- istry; Biology; Geology; Physics; Mathematics and Computer Science; Engineering;
32We are grateful to the Fundaci ́on Espan ̃ola para la Ciencia y la Tecnolog ́ıa for providing us with access to the data.
49
Medicine, Veterinary and Pharmacology; Economics and Management; Psychology, Sociology and Political Science.33 Second, in each broad disciplinary category we match publications with individuals in our database using the information on their surnames and initials.
Specifically, the publication is assigned to a professor in the list of eligible eval- uators if it is in the same disciplinary category as the professor, and the author’s surname and initial, as reported by ISI, coincide (i) with the first surname and the first name’s initial of the professor, (ii) with the last surname and the first initial, (iii) with the first surname hyphenated with the second surname and the first initial. We also repeat stages (i) through (iii) substituting the first initial with the middle- name initial. If a given publication can be assigned to more than one possible match, the value of this publication is divided by the number of such possible matches.
Given that the propensity to publish differs substantially across the disciplines, we normalize the number of individual’s publications to have zero mean and unit standard deviation among applicants to the same exam and among eligible evalua- tors of a given category in a given exam. The number of citations of each publication depends on the time elapsed between the publication date and the date when the number of received citations is observed. Therefore, we first normalize the number of citations that each publication receives by subtracting the average number of citations received by Spanish-authored articles published in the corresponding ISI disciplinary area in the same year and then dividing by the corresponding standard deviation. Next, for each individual in our database we calculate the average num- ber of citations per publication. For individuals who have no ISI publications, this variable takes the minimum value in the corresponding discipline. Finally, similarly to the number of publications, we normalize the number of individual’s citations per publication to have zero mean and unit standard deviation among applicants to the same exam and among eligible evaluators of a given category in a given exam.
Dialnet Dialnet (http://dialnet.unirioja.es) is an open access bibliographic index created by the University of Rioja. It contains information on more than 8,000 journals and more than 3,5 million documents in Hispanic languages, including articles published in scientific journals, collective works and books. The database mainly covers publications in social sciences and humanities. Dialnet provides (in most cases) systematized information on individual authors’ first name, paternal surname, maternal surname and affiliation, thus limiting potential concerns about homonymity.
33In practice, apart from the case of journals Science and Nature, the ISI scientific categories are assigned to journals, not publications. In very rare cases a publication happened to be assigned to more than one broad disciplinary group.
50
We collected information on publications in Dialnet. Due to its lack of repre- sentativeness, we did not considered publications in Science and Engineering. We also excluded publications that appear in ISI Web of Science. We also restricted the set of journals considered to those which satisfy certain minimum research quality requirements (categories A, B or C) as established by the Integrated Scientific Jour- nals Classification (CIRC) (Torres-Salinas et al. 2010).34 Similarly, we considered only books and collective volumes that are published by publishers that satisfy a minimum quality requirement. In particular, we used the EPUC-CSIC publisher list, which summarizes the names of the main publishers in social sciences and humani- ties in Spain and abroad (Gim ́enez-Toledo, Tejada-Artigas and Man ̃ana-Rodr ́ıguez 2012).35 Publications that have been excluded from our study are mainly publica- tions in working paper series, non-refereed journals and volumes published by local universities (around 30%).
Teseo database on doctoral dissertations Since 1977, PhD candidates in Spanish universities have registered their dissertation in the database TESEO, which is run by the Ministry of Education. We retrieved all the information available in this database from the website https://www.educacion.gob.es/teseo in May 2011. While registration is compulsory, according to Fuentes and Arguimbau (2010) TESEO includes information on approximately 90% of all dissertations read in Spain duringthisperiod.36 Weobserveinformationon151,483dissertations.TESEOpro- vides the identity and affiliation of dissertations’ authors, advisors and committee members. Approximately 40% of dissertations are female authored. Female super- visors are scarce and represent only 18% of the total. While 58% of the students they supervise are female, in the case of male advisors, 61% of their students are male.
We match TESEO data with the list of candidates and evaluators. In exams to full professor positions we are able to find the dissertation of 71% of candidates and 41% of evaluators. In exams to associate professor positions we observe the dissertation of 83% of candidates and 70% of evaluators. Missing information may be due to the fact that individuals (i) did their PhD abroad, (ii) defended their dissertation before 1977, (iii) there are spelling mistakes, (iv) the dissertation was not
34Torres-Salinas, Daniel, Maria Bordons, Elea Gim ́enez-Toledo, Emilio Delgado-L ́opez-C ́ozar, Evaristo Jim ́enez-Contreras y El ́ıas Sanz-Casado (2010), “Clasificaci ́on integrada de revistas cient ́ıficas (CIRC): propuesta de categorizaci ́on de las revistas en ciencias sociales y humanas,” El profesional de la informaci ́on, v. 19, n. 6, pp. 675-683.
35Gim ́enez-Toledo, Elea, Carlos Tejada-Artigas and Jorge Man ̃ana-Rodr ́ıguez (2012), “Scholarly Publishers Indicators (SPI),” 1st edition. Available at: http://epuc.cchs.csic.es/SPI [Accessed Nov. 18 2013.]
36Fuentes, Eul`alia and Llorenc ̧ Arguimbau (2010), “Las Tesis Doctorales en Espan ̃a (1997- 2008): An ́alisis, Estad ́ısticas y Repositorios Cooperativos,” Revista Espan ̃ola de Documentacio ́n Cient ́ıfica, Vol. 33(1), pp. 63-89.
51
included in TESEO for unknown reasons (approximately 10% of all dissertations), or (v) there was a problem with homonymity (in our dataset 0.1% of individuals share the same name, middle name, paternal surname and maternal surname).
Each thesis has been classified by its author using the Unesco International Stan- dard Nomenclature for Fields of Science and Technology. This system developed by Unescoincludesmorethantwothousandsix-digitscategories.37 80%ofdissertations provide this information. Approximately half of the authors select one six-digit cat- egory, 35% select two categories, and 15% select three or more categories. There are on average around one hundred dissertations per category. We use this information to construct a measure of individuals’ research interests. In particular, we take into account every dissertation where an individual appears as an advisor, committee member or author. We were able to obtain information on the research interests of 98% candidates to full professor positions, 94% of candidates to associate professor positions, 98% of eligible full professors and 96% of eligible associate professors.
37Available at http://unesdoc.unesco.org/images/0008/000829/082946eb.pdf 52
53
Female 0.20 Tenure in position 13 Age - All Publications: 131 - Articles 73 - Books 8
0.07
-0.27
0.000
0.14
13 0.05 52 0.00 34 0.02 30 0.02
-0.33 -0.03 -0.14 -0.14 -0.06 -0.07
0.000 0.010 0.000 0.000 0.000 0.000
0.35
10 0.01 -0.03 45 0.01 -0.02 14 0.05 -0.09 12 0.05 -0.10
0.000 0.001 0.000 0.000 0.000 0.003
- Book chapters 22
- Conference proceedings 20
- Patents 0.42 - Other 7 Total Article Influence Score 133 A-journal articles 11 All Publications, previous 10 years 74 Total Article Influence Score, previous 10 years 72 A-journal articles, previous 10 years 6 PhD students advised - PhD committees - Based in the South 0.28 Observations 5,876
0.04 0.05 0.05 0.02 -0.00 0.00 -0.00 0.04 0.05 0.03 0.03 0.05
-0.17 -0.20 -0.19 -0.07 0.02 -0.01 0.00 -0.24 -0.14 -0.13 -0.18 -0.13
0.000 0.000 0.000 0.005 0.552 0.522 0.909 0.000 0.000 0.000 0.000 0.000
1 0.01 3 0.01 -
0.44 0.01 -0.02 1 0.01 -0.01 -
Table C1: Descriptive statistics – Eligible evaluators
1 2 3 4 5 6 7 8 9 10 11 12
All
Male
Female p-value
0.01
-0.03
0.169
0.36 0.05 -0.09 61,052
Italy
Spain
Full professors
Full professors
All Male Female p-value
Associate professors
All Male Female p-value
only evaluators each corresponding variable and sample. In columns 2, 3, 6, 7, 10 and 11 variables have been normalized to have zero mean and unit variance for individuals within each field and rank. Columns 4, 8 and 12 report the p-value of a t-test of the difference in means between male and female eligible evaluators in the corresponding variable. Article Influence Score (AIS) is only available for candidates in science, technology, engineering, mathematics, medicine. Information on publications in A-journal articles is only provided for candidates in social sciences and humanities. In Italy, southern regions refer to Abruzzo, Molise, Campania, Apulia, Basilicata, Calabria and islands. In Spain, southern regions
Notes: The table provides descriptive information for the pool of eligible evaluators in who are based in an Italian university. Columns 1, 5 and 9 report mean values for
qualification exams in Italy and in Spain. In Italy it includes
include Extremadura, Castille-La Mancha, Andalusia, Murcia, Valencia and islands.
0.10 0.00 -
-0.02
0.001
0.04 0.01 -0.02 -
0.000
33 0.01 4 0.03 18 0.02 19 0.01 2 0.02 5 0.03 25 0.05 0.34 0.02
-0.10 -0.15 -0.10 -0.07 -0.09 -0.20 -0.33 -0.12
0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
12 0.03 -0.07 2 0.04 -0.06 9 0.04 -0.08 8 0.04 -0.08 1 0.03 -0.04 1 0.08 -0.15 5 0.07 -0.13
0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
49,199
Table C2: Descriptive statistics – Applications
1 2 3 4 5 6 7 8 9 10
Applications to full professorships
Applications to associate professorships
Mean St.Dev.
Male
-0.01 0.72 0.76 -0.01 0.04 0.06 0.04 0.01 -0.01 0.01 -0.00 -0.01 -0.01 0.02 0.03 0.04 0.50 0.45 0.15 0.37 0.48 0.46
-0.01 0.03 0.04 0.01 0.01 0.00 0.00 0.00 0.01 -0.01 0.05 0.03 0.03 0.11
Female
0.01 0.77 0.80 0.03 -0.09 -0.14 -0.09 -0.03 0.02 -0.03 0.00 0.02 0.02 -0.04 -0.09 -0.07 0.51 0.37 0.20 0.34 0.46 0.44
0.03 -0.09 -0.09 -0.03 -0.02 0.00 0.01 0.00 -0.02 0.02 -0.10 -0.09 -0.08 0.09
p-value
Mean St.Dev.
0.41
43 7
0.47 0.50 0.74 0.44 14 60 53 54 30 41 2 4
6 10
8 17 0.19 1.39 8 20
6 18 0.22 0.2 0.11 0.15 1.30 0.99
3 5 0.50 0.29 0.36 0.48 0.13 0.34 0.37 0.48 0.50 0.50 0.44 0.47
Male Female p-value
Female 0.31 Age 49 Permanent position: 0.74 - same field 0.77 CV length (pages) 20 All Publications: 89 - Articles 53 - Books 3
- Book chapters 10
- Conference proceedings 14
- Patents 0.35 - Other 10 Number of coauthors per article 6 First-authored 0.22 Last-authored 0.15 Average Article Influence Score 1.31 A-journal articles 6 Application order 0.50 Above the median in 3 indicators 0.42 Withdrawal 0.16 Qualified 0.36 Failure 0.48 Proportion of positive votes 0.46 Number of applications 21,594
Female 0.27 Age 46 All Publications: 19
- Articles 17
- Books 0.64 - Book chapters 1.57 - Patents 0.04 Average number of coauthors 3 First-authored 0.25 Last-authored 0.24 Average Article Influence Score 0.75 A-journal articles 3 PhD students advised 2 PhD committees 7 Qualified 0.11 Number of applications 13,444
8 0.44 0.42 79 83 65 6 15 26 2.09 27 19 0.19 0.17 0.95 9 0.29 0.49 0.37 0.48 0.50 0.47
6 21 21 1.47 3.18 0.33 10 0.31 0.30 0.43 5 3 9 0.31
0.199 0.000 0.000 0.006 0.000 0.000 0.000 0.004 0.050 0.000 0.938 0.024 0.037 0.000 0.000 0.000 0.012 0.000 0.000 0.000 0.012 0.051
0.015 0.000 0.000 0.005 0.086 0.919 0.691 0.862 0.220 0.458 0.000 0.000 0.000 0.003
47,426
Spain
0.40 37
8
7 0.21 0.54 0.02 5 0.26 0.17 0.72 1 0.24 1 0.12 17,799
6 14 14 0.65 1.41 0.22 23 0.34 0.30 0.54 2 0.88 3 0.32
0.02 -0.03 0.46 0.48 0.72 0.76 -0.03 0.04 0.04 -0.06 0.07 -0.10 0.06 -0.08 0.01 -0.02 -0.01 0.01 0.03 -0.04 -0.00 0.00 -0.02 0.03 0.00 0.00 0.02 -0.03 0.03 -0.04 0.04 -0.05 0.50 0.50 0.37 0.33 0.12 0.16 0.38 0.35 0.5 0.5 0.45 0.43
0.03 -0.05 0.07 -0.10 0.07 -0.11 0.02 -0.02 0.01 -0.01 0.01 -0.01 0.00 0.00 0.01 -0.01 0.03 -0.05 0.03 -0.06 0.06 -0.06 0.03 -0.05 0.05 -0.08 0.12 0.11
0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.002 0.028 0.000 0.899 0.000 0.365 0.000 0.000 0.000 0.717 0.000 0.000 0.000 0.975 0.000
0.000 0.000 0.000 0.000 0.025 0.012 0.863 0.200 0.000 0.000 0.000 0.000 0.000 0.025
Italy
Notes: Columns 1 and 6 report mean values for each corresponding variable and sample. In columns 3, 4, 8,
variables have been normalized to have zero mean and unit variance for applications within each exam. Columns 5 and 10 report the p-value of a t-test of the difference in means between male and female candidates in the corresponding variable. Article Influence Score (AIS) is only available for candidates in science, technology, engineering, mathematics and medicine. Information on publications in A-journal articles is only provided for candidates in social sciences and humanities.
54
and 9 all productivity
Table C3: Descriptive statistics – Links and Research Overlap
12345
Italy
Colleagues Coauthors Same subfield
Spain
Colleagues
Coauthors
PhD advisor
PhD thesis committee Overlap in research interests
All
N Mean
Male Female
Mean Mean
0.027 0.030 0.015 0.013 0.597 0.599
0.047 0.043 0.005 0.004 0.002 0.002 0.014 0.011 0.183 0.218
p-value
0.000 0.000 0.020
0.000 0.000 0.322 0.000 0.000
2,555,839 2,555,839 1,373,790
5,445,067 5,445,067 5,445,067 5,445,067 4,711,621
0.028 0.014 0.598
0.046 0.004 0.002 0.013 0.196
Notes: The table provides information on links between candidates and eligible evaluators within each discipline. Information about research interests is only available for candidates with a permanent contract in an Italian university and for candidates who have defended their thesis in Spain or who have participated in a thesis committee in Spain. The vari- able Same subfield takes value one if a candidate and an eligible evaluator belong to the same subfield (settore scientifico-disciplinare). The variable Overlap in research interests measures the degree of overlap between the research interests of eligible evaluators and candidates, as measured by their participation in PhD thesis committees.
55
For Online Publication Appendix D. First-stage estimates
Below we report the first-stage estimates from the IV estimations of the effect of committee gender composition on the relative success of female candidates reported in Table 1.
Table D1: First-stage estimates
Femalefinal Female ∗Femalefinal Female ∗Femalefinal eieie
Dependent variable:
Second-stage estimates:
F emaleinitial 0.822*** -0.006
(0.048) (0.006) Female ∗Femaleinitial -0.042 0.788***
Column 5 of Table 1
0.810*** (0.055)
Yes Yes Yes Yes
218 218
0.966*** (0.020)
Yes Yes Yes Yes
2310 2310
Column 4 of Table 1
Italy
e
ie
(0.032) (0.066) F emaleexpected Yes Yes
Controls:
e
Female ∗Femaleexpected Yes Yes ie
Candidate characteristics Yes Yes Exam FE
F statistics: 188 74 Sanderson-Windmeijer F statistics: 323 380
F emaleinitial 0.954*** 0.000 e
(0.021) (0.002) Female ∗Femaleinitial 0.014 0.970***
ie
(0.012) (0.017) F emaleexpected Yes Yes
F emale ∗ F emaleexpected Yes Yes ie
Candidate characteristics Yes Yes Exam FE
F statistics: 2396 3135 Sanderson-Windmeijer F statistics: 2116 3503
Notes: Standard errors are clustered by committee. * p < 0.10, ** p < 0.05, *** p < 0.01.
Controls:
e
Spain
56
For Online Publication Appendix E. Nonlinearities
The effect of the gender composition of committee on the relative success rate of females may be non-linear for a number of reasons. First, the presence of a woman in the committee may affect the voting behavior of male evaluators (see section 4.5). If this is the case, the transition from zero to one female evaluator in the committee may have a different effect than the transition from one to two female evaluators, or from two to three female evaluators. Second, decisions in the committee are taken on a (qualified) majority basis. Therefore, having a committee where the (qualifying) majority of members are female might have a particularly strong effect.
In order to correctly identify the potential existence of nonlinear effects, it is necessary to control for the probability that a given number of women is assigned to the committee. We consider the following model:
yie =β0 +β1Femalei +γkFemaleiDke k
+δkFemaleiDexpected +Xiβ2 +Ziβ3 +μe +εie ke
k
where Dke is a dummy variable that takes value one if the number of female eval-
uators in committee e is equal to k and Dexpected is the probability that exactly ke
k female evaluators are assigned to a given committee. For Spanish evaluations, we directly compute these probabilities using information on the gender mix of the pool of eligible evaluators. For the Italian case, the direct computation is more complicated, since the assignment procedure required no more than one committee member from each university. Instead, we compute these probabilities using the outcomes of 1,000,000 simulated random draws, which account for the restrictions on the randomization.
Committees rarely included more then three women. Therefore, we only ana- lyze the effect of having one, two, and three or more female evaluators. In both countries, four positive votes are required for qualification. The estimation results are presented in Table E1. Overall, the linearity of the effect of committees’ gender composition cannot be rejected by the data.
57
Table E1: Nonlinearities
12
Female
Female* 1 female evaluator
Female* 2 female evaluators
Female* 3 or more female evaluators
Number of observations
Italy
0.000 (0.007) -0.017 (0.012) -0.036*** (0.012) -0.079*** (0.022) 69020
Spain
-0.012 (0.007) -0.002 (0.010) -0.005 (0.013) -0.005 (0.014) 31243
Notes: IV estimates. All regressions include as controls exams fixed-effects, the number of female evaluators in the committee, individual predetermined characteristics, and the expected probabilities to have 1, 2, and 3 or more female evaluators. Standard errors are clustered by committee.
* p < 0.10, ** p < 0.05, *** p < 0.01.
58
For Online Publication
Appendix F. Committee composition and evaluators’ resig- nations
In section 4.5 we estimate the effect of committee gender composition on the voting behavior of male evaluators. The consistency of these estimates relies on the assumption that evaluators’ resignation was not affected by the gender composition of the committee. To examine this possibility, we estimate the following equation on the sample of initially drawn evaluators:
Evaluatorfinal = β + β Female + β Femaleinitial + β Female ∗ Femaleinitial je 0 1 j 2 e 3 j e
+β4Femaleexpected +β5Femalej ∗Femaleexpected +εij, (10) je je
where Evaluatorfinal is an indicator for those initially drawn evaluators who served je
in the final evaluation committee, Femaleinitial is the share of women in the ini- e
tially drawn committee and Femaleexpected is the expected share of women in the je
committee conditional on the inclusion of evaluator j. Results from the estimation of equation (10) are reported in column 1 of Table F1. The presence of women in the committee does not affect the likelihood that a male or a female evaluator resigns. In column 2, we control for a number of evaluator characteristics includ- ing tenure, quality-adjusted productivity (total Article Influence Score in Sciences and the number of A-journal articles for Social Sciences and Humanities), and the location of their university. The estimates are unaffected by the inclusion of these controls.
59
Table F1: The effect of committee composition on evaluators’ resignations
Female evaluator
Share of women in the committee
Female evaluator * Share of women in the committee
Expected share of women
Female evaluator * Expected share of women
Controls:
Evaluator characteristics
Mean dependent variable Number of observations
0.117 (0.093) 0.118* (0.071) -0.085 (0.170) -0.187 (0.192) -0.118 (0.289)
0.922 920
Italy
0.106 (0.096) 0.115 (0.071) -0.079 (0.174) -0.173 (0.196) -0.127 (0.293)
Yes
0.922
920
Notes: OLS estimates. The dependent variable is an indicator that takes value one if the initially drawn evaluator serves in the final evaluation committee.
* p < 0.10, ** p < 0.05, *** p < 0.01.
60
For Online Publication
Appendix G. The effect of connections, by gender of evalua- tors and candidates
In Table G1 we explore whether the impact of connections depends on the gender of evaluators and candidates. We consider coauthors, colleagues and, in the case of Spain, also advisors. The gender of connections does not seem to play any role. Male and female candidates benefit equally from the presence of a female or a male connection in the committee.
Table G1: The effect of strong connections, by candidate and evaluator gender
12
Italy
Female candidate 0.006 (0.008)
Female candidate * Share of female evaluators -0.131*** (0.035)
Share of connections in committee 0.204*** (0.040)
Female candidate * Share of connections in committee 0.010 (0.065)
Share of female connections in committee -0.008 (0.085)
Female candidate * Share of female connections in committee 0.149 (0.125)
Number of observations 69020
Notes: IV estimates. All regressions include exam fixed-effects, individual characteristics, Female candidate* Expected share of women in committee, Expected con- nections in committee, Female candidate* Expected connections in committee, Expected female connections in committee and Female candidate* Expected female connections in committee. Standard errors are clustered by committee.
* p < 0.10, ** p < 0.05, *** p < 0.01.
Spain
-0.012* (0.007) -0.012 (0.028) 0.427*** (0.038) 0.020 (0.060) -0.036 (0.101) -0.084 (0.145) 31243
61
predetermined
For Online Publication
Appendix H. Heterogeneity analysis, alternative specifica- tions.
In section 4.6.1 we explore whether the effect of the committees’ gender compo- sition varies depending on whether evaluators and candidates share similar research interests and depending on the degree of feminization of the field. In Table 7 we re- port results from an analysis where we split the sample of candidates in each country in two groups based on the median value of each variable.
In this section we present an alternative specification. We estimate a model with triple interactions exploiting the full range of possible values of the running variables. First, we analyze the impact of research similarity. We estimate the following model:
Y = β + β Female + β Female ∗ Femalefinal + β Sfinal + β Female ∗ Sfinal+ ie01 i2 i e 3ie 4 iie
+ β Female ∗ Femalefinal ∗ Sfinal+ 5 i e ie
+ β F emale ∗ F emaleexpected + β Sexpected + β F emale ∗ Sexpected+ 6 i e 7ie 8 iie
+β Female ∗Femalefinal ∗Sfinal +Xβ +μ +ε (11) 9 i e ie i10eie
where Sfinal and Sexpected stand for the actual and the expected research similar- ie ie
ity between candidate i and committee e. We instrument the final composition of the committee (Femalefinal, Sfinal) using the outcome of the initial lottery draw
e ie
(Femaleinitial, Sinitial). In this model, coefficient β shows the effect of commit-
eie 2
tee gender composition when the committee members and candidates do not share
research interests (Sfinal = 0). Coefficient β5 shows how this effect changes when ie
candidates are evaluated by committees composed of evaluators who share research interests with the candidate.
As shown in Table H1, columns 1 and 2, in both countries the presence of women in the committee reduces the relative chances of success of female candidates when candidates and evaluators have different research interests. However, this effect disappears when candidates and evaluators share the same research interests.
Second, analyze the impact of the degree of feminization of the field. We estimate the following equation:
Y = β + β Female + β Female ∗ Femalefinal ie 0 1 i 2 i e
+ β Female ∗ Femalefinal ∗ Female + β Female ∗ Femaleexpected 3ied4ie
+β Female ∗Femalefinal ∗Female +Xβ +μ +ε , (12) 5ie di6eie
where F emaled is the proportion of women among full professors in the correspond- ing discipline. Equation (12) allows us to explore whether the effect of committee
62
gender composition varies depending on the feminization of the discipline. Results from this analysis are shown in columns 3 and 4 of Table H1. The effect of com- mittees’ gender composition on female candidates’ success rate does not depend significantly on the degree of feminization of the field.
Table H1: The effect of committee composition, by research interest overlap and the degree of feminization of the field
1234
Female
Female * Share of women in the committee
Research similarity
Female * Research similarity
Share of women in the committee * Research similarity
Female * Share of women in the committee * Research similarity
Female * Share of women in the committee * Feminization of the discipline
Average research similarity
Average feminization of the discipline
Candidate characteristics Exam FE
Number of observations
Italy
0.026 (0.031) -0.177** (0.077) 0.107** (0.049) -0.072 (0.063) -0.216 (0.138) 0.242** (0.106)
0.455
Yes
Yes 35832
Spain
0.012 (0.020) -0.096** (0.047) 0.231*** (0.046) -0.088 (0.069) -0.055 (0.092) 0.282*** (0.105)
0.262
Yes
Yes 27998
Italy
0.010 (0.009) -0.032 (0.109)
-0.350 (0.421)
0.218
Yes
Yes 69020
Spain
-0.011 (0.007) -0.008 (0.048)
-0.046 (0.269)
0.128
Yes
Yes 31243
Notes: Characteristics of final committees are instrumented by the characteristics of initial committees selected by random draw. Research similarity is measured in Italy as the propor- tion of committee members in the same subfield as the candidate and in Spain as the average overlap in research interests (see more details in Data section). Feminization of the discipline is measured by the proportion of women among all full professors in the discipline in 2012 in Italy and in 2002 in Spain. Columns 1 and 2 include also Female candidate* Expected share of women in committee, Expected research similarity, Female candidate* Expected research similarity, Expected share of women in committee * Expected research similarity and Female candidate* Expected share of women in committee * Expected research similarity. Columns 3 and 4 include Female candidate* Expected share of women in committee and Female can- didate* Expected share of women in committee * Feminization of the discipline. Standard errors are clustered by committee.
* p < 0.10, ** p < 0.05, *** p < 0.01.
63