Ibañez, Marcela; Rai, Ashok; Riener, Gerhard Working Paper Sorting through affirmative action: Three field experiments in Colombia DICE Discussion Paper, No. 183 Provided in Cooperation with: Düsseldorf Institute for Competition Economics (DICE), Heinrich Heine University Düsseldorf Suggested Citation: Ibañez, Marcela; Rai, Ashok; Riener, Gerhard (2015) : Sorting through affirmative action: Three field experiments in Colombia, DICE Discussion Paper, No. 183, ISBN 978-3-86304-182-3, Heinrich Heine University Düsseldorf, Düsseldorf Institute for Competition Economics (DICE), Düsseldorf This Version is available at: http://hdl.handle.net/10419/109211 Standard-Nutzungsbedingungen: Terms of use: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your Zwecken und zum Privatgebrauch gespeichert und kopiert werden. personal and scholarly purposes. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich purposes, to exhibit the documents publicly, to make them machen, vertreiben oder anderweitig nutzen. publicly available on the internet, or to distribute or otherwise use the documents in public. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, If the documents have been made available under an Open gelten abweichend von diesen Nutzungsbedingungen die in der dort Content Licence (especially Creative Commons Licences), you genannten Lizenz gewährten Nutzungsrechte. may exercise further usage rights as specified in the indicated licence. No 183 Sorting Through Affirmative Action: Three Field Experiments in Colombia Marcela Ibañez, Ashok Rai, Gerhard Riener April 2015 IMPRINT    DICE DISCUSSION PAPER    Published by    düsseldorf university press (dup) on behalf of  Heinrich‐Heine‐Universität Düsseldorf, Faculty of Economics,  Düsseldorf Institute for Competition Economics (DICE), Universitätsstraße 1,  40225 Düsseldorf, Germany  www.dice.hhu.de      Editor:    Prof. Dr. Hans‐Theo Normann  Düsseldorf Institute for Competition Economics (DICE)  Phone: +49(0) 211‐81‐15125, e‐mail: [email protected]       DICE DISCUSSION PAPER    All rights reserved. Düsseldorf, Germany, 2015    ISSN 2190‐9938 (online) – ISBN 978‐3‐86304‐182‐3      The working papers published in the Series constitute work in progress circulated to  stimulate discussion and critical comments. Views expressed represent exclusively the  authors’ own opinions and do not necessarily reflect those of the editor. Sorting Through A(cid:30)rmative Action: Three Field Experiments in Colombia∗ Marcela Ibaæez†, Ashok Rai‡, Gerhard Riener§ April 2015 Abstract A(cid:30)rmative action to promote women’s employment is a intensely debated policy. Do a(cid:30)rmative action policies attract women and does it come at a cost of deterring high quali(cid:28)ed men? In three (cid:28)eld experiments in Colombia we compare characteristics of job-seekers who are told of the a(cid:30)rmative action selection criterion before they apply with those who are only told after applying. We (cid:28)nd that the gains in attracting female applicantsfaroutweighthelossesinmaleapplicants. A(cid:30)rmativeactionismoree(cid:27)ective in areas with larger female discrimination and deters male job-seekers from areas with low discrimination. JEL code: J21, J24, J48, C93 Keywords: Field experiment; A(cid:30)rmative Action; Labor market; Gender participation gap ∗WearegratefultoPatriciaCastroforhergenerosityandherinvaluablehelpconductingtheexperiments as well as to the Economics department at the Universidad de los Andes, BogotÆ for their hospitality. This studywas(cid:28)nancedbytheCourantResearchCenterforPoverty,EquityandGrowth,UniversityofG(cid:246)ttingen. Gerhard Riener gratefully acknowledges the (cid:28)nancial support from the Deutsche Forschungsgemeinschaft under the grant RTG 1411. †Courant Research Center Poverty, Equity and Growth, Georg August University G(cid:246)ttingen, 37073 G(cid:246)t- tingen, Germany. email: [email protected] ‡Department of Economics, Williams College, Williamstown, MA. email: [email protected] §Department of Economics, University of Mannheim. email: [email protected] 1 1 Introduction A(cid:30)rmative action policies are the subject of intense and polarized debate [Cohen and Sterba, 2003, Fullinwider, 2011]. Supporters point to the opportunities to address historical and statistical discrimination and to the advantages of diversity in both the workplace and in the classroom [Weber and Zulehner, 2014, Clayton and Crosby, 1992]. Critics contend that a(cid:30)rmative action is reverse discrimination [Newton, 1973], violates the principle of merit [Walzer, 1983, pp. 143(cid:21)154] and can lead to economic ine(cid:30)ciencies [Coate and Loury, 1993]. In this paper we study the sorting of job applicants in response to a(cid:30)rmative action. We analyze some of the key questions in the debate in a naturally-occurring labor market: Do a(cid:30)rmative action policies for women actually encourage women to apply for jobs? Does this come at the cost of fewer applications from men? What kind of women are attracted to an a(cid:30)rmative-action job? What kind of men are deterred? We conducted this study in Colombia, a country with substantial degree of female seg- regation in the labor market [see the review in Peæa et al., 2013]. Although the proportion of women who complete a university degree in Colombia is larger than that of men (57.6 percent for undergraduates and 50.9 percent for graduates), women are only 70% as likely as men to enter the labor force. Besides employment rate and average earnings are higher for men than women.1 In response to these inequalities, the government adopted a(cid:30)rmative action rules for higher political o(cid:30)ce. Yet, a(cid:30)rmative action is not commonly used in the private sector and hence Colombia provides a controlled environment to test for the e(cid:27)ect of the voluntary introduction of this policy. To investigate the e(cid:27)ect of a(cid:30)rmative action on the labor market, we conducted three large-scale (cid:28)eld experiments. In two of the experiments the announced positions concerned research assistants and the third experiment concerned the hiring of a consultant to work for a consultancy. In all of the experiments we apply a two stage design. In the (cid:28)rst stage we recruited a large pool of job-seekers posting job advertisements. In the second stage we randomly varied the information that interested job-seekers received. Half of the job-seekers were told that a(cid:30)rmative action would play a role in selection before they completed the application form; the other half were informed of the a(cid:30)rmative action policy only after the application process was completed. This procedure has two main advantages: First it allows us to measure the impact of a(cid:30)rmative action on application rates, or the proportion of candidates of each gender to apply to a position. Second it allows us to observe personal characteristics of job-seekers before the policy is announced. Hence, we can attribute di(cid:27)er- ences in the resulting distribution of characteristics of applicants to the a(cid:30)rmative action 1 Cepeda and Bar(cid:243)n (2012) estimate that female average salary is about 6.8% lower for women once that di(cid:27)erences in (cid:28)eld of study are accounted for. 2 policy. We use two di(cid:27)erent a(cid:30)rmative action rules. The research-assistant experiments used a quota rule for selection: (cid:28)fty percent of the positions were reserved were women. The consultant experiment used a preferential treatment rule: women with equal quali(cid:28)cations would be preferred. A(cid:30)rmativeactionpoliciesmightreduceperceivedcompetitivenessandinducemorewomen to apply but this policy might eventually worsen the problem. For instance, if women an- ticipate a patronization equilibrium, in the spirit of Coate and Loury [1993] in which they are receiving the job as (cid:16)token(cid:17) females, they might choose not to apply as their self-image would su(cid:27)er [Heilman et al., 1992, Unzueta et al., 2010, Bracha et al., 2013]. Our results establish that a(cid:30)rmative action has no such perverse selection e(cid:27)ects. Our main (cid:28)nding is that a(cid:30)rmative action encourages women to apply (cid:21) and this does not come at the expense of reducing male applications. In the three experiments, women are 5 to 20 percent more likely to apply with a(cid:30)rmative action than without. In two of the experiments there is no signi(cid:28)cant deterrence of males; in one research-assistant experiment males are 9 percent less likely to apply with a(cid:30)rmative action than without. That loss is made up by an equal gain in female applicants, however. Furthermore, we compared the applicant pool with and without a(cid:30)rmative action on a variety of dimensions. For all three experiments we collected basic information on quali(cid:28)- cations (experience, performance in cognitive tests and performance in similar job) before manipulation. Our results indicate that there are non-linear e(cid:27)ects of quali(cid:28)cation on the likelihood to apply. Under a(cid:30)rmative action, the best quali(cid:28)ed women and men are more likely to apply to the job o(cid:27)ers than under the control treatment. For the most recent experiment we also collected information on cognitive abilities, per- sonality tests and attitudes towards risk and time before manipulation. We found evidence of di(cid:27)erences in personality types with female applicants being more impulsive in the a(cid:30)r- mative action treatment compared with the control. No di(cid:27)erences are found with respect to risk and time preferences for the applicants under a(cid:30)rmative action. Inordertofurtherinvestigatetheselectione(cid:27)ectsofa(cid:30)rmativeaction,inthetwoassistant experiments we allow applications from job-seekers from any area of studies. This allows us to compare, how female discrimination a(cid:27)ects applications by male and female candidates. Interestingly, our results suggest that in the absence of a(cid:30)rmative action, discrimination in the labor market, measured as gender di(cid:27)erences in average income, discourages female applicats.A(cid:30)rmative action is e(cid:27)ective at closing this gap by attracting female candidates from areas with high discrimination without creating discouraging e(cid:27)ects on male applicants. Thisstudycontributestotheliteratureonfemalesegregationinlabormarkets. Itis,tothe 3 best of our knowledge, the (cid:28)rst paper that provides (cid:28)eld evidence on the impact of a(cid:30)rmative action policies in favor of women on sorting in the labor market. Few papers use natural experiments to test if a(cid:30)rmative action policies encourage minority students to apply for college and the results are rather mixed [Long, 2004, Card and Krueger, 2005, Dickson, 2006, Hinrichs, 2012]. Bertrand et al. [2010] consider the impacts of a(cid:30)rmative action policies that favor university admission from low cast students in India. They (cid:28)nd that the marginal low cast entrant comes from a less advantaged background than the marginal high-cast displaced indicating that the policy favors the target population. Our paper is complementary to these papers by considering the sorting e(cid:27)ects of a(cid:30)rmative action policies that favor women in the work place. Most of the papers on sorting e(cid:27)ects of a(cid:30)rmative action in a labor market settings refer to laboratory experiments. Niederle et al. [2013], Balafoutas and Sutter [2012] and Calsamiglia et al. [2013] consider self-section into a tournament and (cid:28)nd that a(cid:30)rmative action rules can incentivize women to enter into competitive environments. The (cid:28)eld context in which we conduct the experiment allows to capture dimensions di(cid:27)erent from aversion to competition that constraint women from participating in the labor market. For instance, culturalnormstowardswomenworking,availabilityofchildcareandsupportathomecanplay and important role explaining female participation in the labor market [Fogli and Veldkamp, 2011, FernÆndez, 2013, Bauernschuster and Schlotter, Forthcoming, Barone and Mocetti, 2011, Coen-Pirani et al., 2010] [Bauernschuster and Schlotter, Forthcoming] [Barone and Mocetti, 2011, Coen-Pirani et al., 2010] . Besides, unlike lab experiments, participants in the our (cid:28)eld experiment are unaware that they are participating in an experiment, and hence, our results are not subject to experimental demand e(cid:27)ects that could confound the (cid:28)ndings from previous lab experiments. Previous work using (cid:28)eld evidence examined the sorting of workers into di(cid:27)erent jobs considering the e(cid:27)ect of job characteristics and incentive schemes. For instance Bellemare and Shearer [2010] or Bonin et al. [2007] consider the e(cid:27)ect of wage volatility, Guiteras and Jack [2014] and Dal B(cid:243) et al. [2013] consider the e(cid:27)ect of value of the compensation, Ashraf et al. [2014] consider the salience of career incentives Eriksson et al. [2009], Dohmen and Falk [2011] and Flory et al. [2014] consider the e(cid:27)ect of competitive versus individual remuneration schemes, Fernandez and Friedrich [2011] and Barbulescu and Bidwell [2013] consider the type of job (male vs female stereotypical jobs); Lefebvre and Merrigan [2008] and Havnes and Mogstad [2011] consider the the impact of provision of child care. Our paper is complementary to this research as it considers the impact of a(cid:30)rmative action rules on job-seekers sorting, a topic not explored in this literature yet. Theoretical models on a(cid:30)rmative action mainly attempted to explain i) the long e(cid:27)ects 4 of a(cid:30)rmative action on incentives to exert e(cid:27)ort and ii) the impact on performance on admission tests. However, the predictions are quite mixed.2 For instance Coate and Loury [1993] (cid:28)nds that a(cid:30)rmative action policies that fosters minorities by decreasing the standard of performance can decrease the incentives to invest in education, while Moro and Norman [2003](cid:28)ndsoppositeresultsusingageneralequilibriummodelwithendogenoushumancapital formation. Regarding the e(cid:27)ect on e(cid:27)ort during admission test, various models have shown that a(cid:30)rmative action can induce higher e(cid:27)ort if underlying initial heterogeneity among preferred and non preferred groups is not too large and if competition is increased [Fu, 2006, Fain, 2009, Franke, 2012]. Yet, models than consider di(cid:27)erent forms of heterogeneity among potential candidates can lead to opposite results [Hickman, 2010, Balart, 2011]. The sorting e(cid:27)ects of voluntary a(cid:30)rmative action policies have not been discussed extensively in the literature. We therefore adapt Borjas model (1987) to conceptualize the sorting e(cid:27)ects of a(cid:30)mative action and derive predictions concerning our setup. Previous empirical research has focused on the impacts of a(cid:30)rmative action policy on political and labor market outcomes (see Holzer and Neumark, 2000 for a review on early nonexperimentalevidenceonthee(cid:27)ectsofa(cid:30)rmativeactionintheworkplaceandDahlerup, 2012). In this paper, we focus only on the selection e(cid:27)ects of a(cid:30)rmative action and do not consider the impacts in performance in the work place. Recently, Howard and Prakash [2012] considered the e(cid:27)ect of a female quota rules on public employment in India and (cid:28)nd that this policy increases representation of women from scheduled castes in high-skilled occupations. While they consider the e(cid:27)ect of the policy on (cid:28)nal employment outcome (the combination of supply and demand e(cid:27)ects), we consider the sorting e(cid:27)ect of the policy and focus on the supply side e(cid:27)ects. Moreover, our experimental approach allows investigating the e(cid:27)ects of a(cid:30)rmative action on a large set of characteristics of the applicant pool (quali(cid:28)cations, risk and time preferences, personality, socioeconomics) a topic not addressed previously. The remainder of the paper is organized as follows: Section 2 presents the conceptual framework. Section 3 describes the local context of the labor market in Colombia while Section 4 presents the experimental design and procedures. The results are presented in Section 5. We conclude with a discussion in Section 6. 2 Conceptual framework The e(cid:27)ect of a(cid:30)rmative action on self-selection in the labor market can be conceptualized using Roy’s (1951) selection model. For that, it is useful to follow the conceptualization of Borjas [1987]. In the context of a(cid:30)rmative action policies, agents decide whether to apply 2 For a recent review article on models of discrimination and a(cid:30)rmative action see Fang and Moro [2010]. 5 to a (cid:28)rm that uses a(cid:30)rmative action which we will refer to as the AA (cid:28)rm. Log expected earnings in the AA (cid:28)rm are given by: w = µ −C +ε 1 1 1 where, ε ∼ N(0,σ2). The term µ is the expected earnings in the AA (cid:28)rm that depend 1 1 1 on the probability to be hired, π and the wage level for a given quali(cid:28)cation level, ω. C is the application cost which we assume to be the same for applications to all (cid:28)rms. The term ε captures unobserved factors that a(cid:27)ect earnings. We assume that job-seekers otherwise 1 confront a discriminatory labor market in which the log earnings are di(cid:27)erent across gender. We refer to the conditions in the discriminatory labor market as the D market. In the D market, log earnings for men, M, and women, F, are given by: w = µ +ε M0 M0 M0 w = µ +ε F0 F0 F0 where µ < µ ; ε ∼ N(0,σ2 ) for k ∈ {M,F}. We assume that workers know the F0 M0 kD k0 components of the earning functions in AA (cid:28)rm and in the D market. Further, we assume that there are no di(cid:27)erences in application cost by men and women. The self-selection decision rule implies that job-seekers apply to the (cid:28)rm that uses a(cid:30)rmative action if: I = µ −µ −C +ε −ε > 0 1 k0 1 k0 If women are discriminated in the labor market and receive lower mean expected earnings thanmen,wagediscriminationimpliesthatfemaleapplicantshaveahighermarginalincentive to apply to the (cid:28)rm using a(cid:30)rmative action than male applicants. The larger the gender wage gap, the larger the incentive for female job-seekers to apply and the lower the incentive for male job-seekers to apply. We summarize this in our (cid:28)rst proposition: Proposition 1. A (cid:28)rm that uses an a(cid:30)rmative action policy attracts more female than male applicants in a discriminatory labor market characterized by a gender wage gap where women receive lower wages. A(cid:30)rmative action policies, can however generate selection biases by endogenous selection of workers into the AA (cid:28)rm. The conditional earnings in the D market for job-seekers who apply to the AA (cid:28)rm are given by: E(w I > 0) = µ +E(ε I > 0) 0 0 0 6 while, the expected earnings for applying to the AA (cid:28)rm are: E(w I > 0) = µ −C +E(ε I > 0) 1 1 1 under normality assumptions, the conditional means of earnings for workers who sort in the AA (cid:28)rm are given by: σ σ σ k0 1 k0 E(w I > 0) = µ + (ρ− ) k0 k0 σ σ v 1 σ σ σ k0 1 1 E(w I > 0) = µ −C + ( −ρ) 1 1 σ σ v k0 where k ∈ {M,F}, σ is the correlation between of the error terms (ε − ε ), and v 1 k0 σ = cov(ε ,ε ) and ρ is the correlation of ε and ε . This expression shows that the k01 k0 1 0 1 average job-seeker who is willing to self-select in the AA (cid:28)rm is better than the average (cid:18) (cid:19) σ k0 job-seeker in the D market if ρ− > 0. This worker would also out-perform other σ 1 (cid:18) (cid:19) σ 1 workers in the AA (cid:28)rm if −ρ > 0. Hence, positive sorting, in which relatively better σ k0 performing candidates in the D (cid:28)rm sort into the AA (cid:28)rm and out-perform other job-seekers (cid:16) (cid:17) (cid:16) (cid:17) σ 1 in this (cid:28)rm (Q = E ε I > 0 > 0 and Q = E ε I > 0 > 0) occurs when > 1 0 1 1 σ v0 (cid:18) (cid:19) σ σ 1 0 and ρ > min , . The (cid:28)rst expression implies that the distribution of earnings in the σ σ 0 1 AA (cid:28)rm is more spread than in the D (cid:28)rm. The second expression implies that there is signi(cid:28)cant correlation in earnings in the AA (cid:28)rm and D (cid:28)rm. On the other hand, negative hierarchical sorting in which the worst job-seekers in the D (cid:28)rm self-select to apply to the AA (cid:28)rm and are also worst that the average applicants in the (cid:18) (cid:19) σ σ σ 0 1 0 second (cid:28)rm (Q < 0 and Q < 0) occurs when > 1 and ρ > min , . This means 0 1 σ σ σ 1 0 1 that the distribution of earnings in the AA (cid:28)rm are more concentrated than in the D (cid:28)rm. Besides, as before, there is substantial correlation in earnings in both (cid:28)rms. In a discriminatory labor market, it is reasonable to assume that earnings of the discrim- inated gender (female) would be more concentrated than earning of the privileged gender (male) so σ2 < σ2 < σ2 . This leads to our second hypothesis: F0 1 M0 Proposition 2. If a(cid:30)rmative action policy eliminates di(cid:27)erences in the spread of the earn- ings distributions such that σ2 < σ2 < σ2 , there will be positive hierarchical sorting for F0 1 M0 women and negative hierarchical sorting for men provided that the correlation of skills values is su(cid:30)ciently high. 7