Measuring racial bias in employment services in Colombia

Duryea, Suzanne; Millán-Quijano, Jaime; Morrison, Judith; Oviedo Gil, Yanira Marcela Working Paper Measuring racial bias in employment services in Colombia IDB Working Paper Series, No. IDB-WP-01594 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Duryea, Suzanne; Millán-Quijano, Jaime; Morrison, Judith; Oviedo Gil, Yanira Marcela (2024) : Measuring racial bias in employment services in Colombia, IDB Working Paper Series, No. IDB-WP-01594, Inter-American Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0012870 This Version is available at: https://hdl.handle.net/10419/299410 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/3.0/igo/ Measuring Racial Bias in Employment Services in Colombia Suzanne Duryea Jaime Millan-Quijano Judith Morrison Yanira Oviedo WORKING PAPER No IDB-WP-01594 Inter- A merican Development Bank Gender and Diversity Division March 2024 Measuring Racial Bias in Employment Services in Colombia Suzanne Duryea (IDB) Jaime Millan-Quijano (NCID, Universidad de Navarra and CEMR) Judith Morrison (IDB) Yanira Oviedo (Econometria S.A.) Inter- A merican Development Bank Gender and Diversity Division March 2024 Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library Measuring racial bias in employment services in Colombia / Suzanne Duryea, Jaime Millan-Quijano, Judith Morrison, Yanira Oviedo Gil. p. cm. — (IDB Working Paper Series ; 1594) Includes bibliographical references. 1. Minorities-Economic aspects-Colombia. 2. Gender mainstreaming- Colombia. 3. Equality-Colombia. 4. Discrimination in employment-Colombia. 5. Race discrimination-Colombia. I. Duryea, Suzanne. II. Millan-Quijano, Jaime. III. Morrison, Judith A. IV. Oviedo, Yanira. V. Inter-American Development Bank. Gender and Diversity Division. VI. Series. IDB-WP-1594 JEL codes: J15, J21, J71. Keywords: Implicit stereotypes, labor market discrimination, developing countries. http://www.iadb.org Copyright © 2024 Inter-American Development Bank ("IDB"). This work is subject to a Creative Commons license CC BY 3.0 IGO (https://creativecommons.org/licenses/by/3.0/igo/legalcode). The terms and conditions indicated in the URL link must be met and the respective recognition must be granted to the IDB. Further to section 8 of the above license, any mediation relating to disputes arising under such license shall be conducted in accordance with the WIPO Mediation Rules. Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration pursuant to the United Nations Commission on International Trade Law (UNCITRAL) rules. The use of the IDB's name for any purpose other than for attribution, and the use of IDB's logo shall be subject to a separate written license agreement between the IDB and the user and is not authorized as part of this license. Note that the URL link includes terms and conditions that are an integral part of this license. The opinions expressed in this work are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. Measuring Racial Bias in Employment Services in Colombia.*† Suzanne Duryea IDB Jaime Millán-Quijano NCID, Universidad de Navarra and CEMR Judith Morrison IDB Yanira Oviedo Econometría S.A. March 22, 2024 Abstract In this paper, we document de facto, implicit, and explicit racial biases within the public employment service in Colombia. By combining administrative data about job seekers and job openings with direct surveys to job counselors, including a Race Implicit Association Test, we compute different types of racial bias. We find that while job counselors do not self-report biased attitudes against Afro-descendant individuals, the majority exhibit high levels of implicit bias, which also correlates strongly with observed lower referral rates of Afro-descendants to job openings. In addition, we randomly provide information to job counselors about their implicit bias and test if this information changes their referral behavior. While we demonstrate that the implicit bias of counselors is a major contributor to racial gaps in labor outcomes, we do not find that providing feedback on this unconscious bias changes their referral behavior. JEL classification: J15, J21, J71 Keywords: Implicit stereotypes, labor market discrimination, developing countries. *Contact authors: Suzanne Duryea. [email protected]. Inter-American Development Bank 1300 New York Avenue, N.W. Washington, DC 20577. Jaime Millán-Quijano. [email protected]. ICS, Universidad de Navarra. Edificio de Bibliotecas - Entrada Este, 2ª Planta. Pamplona, Spain 31009. AEA RCT Registry AEARCTR-0008376 †We are grateful to ASOCAJAS, particularly to Renata Samaca, who helped us with the implementation of our study, and the Unidad del Servicio de Empleo for providing the administrative data. We also thank the data collection team in Econometría S.A., and Laura Goyeneche, Laura Gómez Cely, María Camila Árias, Juan José Rincón, and Sofia Vaca for their assistance and data cleaning. The authors acknowledge financial support from the Spanish Ministry of Economy and Competitiveness Grant PID2020-120589RA-I00, and IDB grant RG-E1724. The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. 1 1 Introduction Extensive literature exists on labor market gaps by race in high-income countries, including the profound impact of discrimination in the hiring process (Charles and Guryan, 2008, 2011; Kline et al., 2022). A global review of audit studies finds that discrimination is ubiquitous against non-majority race-ethnic groups. Across 97 field experiments, whites receive about 50% more callbacks than non-Whites (Quillian et al., 2019). In the case of Latin America (LAC), approximately 24% of the total population identifies as Afro-descendant. While employment rates do not differ substantially across race, job quality is lower for Afro-descendants. A gap by race in formal employment of over 10 percentage points is observed for Brazil, Colombia, and Uruguay, with gaps persisting over the last 15 years (Araujo et al., 2022). However, in contrast to high-income countries, few studies exist that attempt to carefully measure labor market discrimination by race or ethnicity, mainly due to the lack of data (Ñopo et al., 2010; Ñopo, 2012). This paper presents different estimates of racial bias in the publicly financed employment system in Colombia. To achieve this, we collected extensive data from various sources to measure de facto, implicit, and explicit racial bias of job counselors within the system.1We gathered administrative data of over 300,000 job applicants and 70,000 job posts from across the country over one year. This data allowed us to observe who was referred to which job, enabling us to compare referral rates by race. We combined the administrative data with detailed interviews with 349 job counselors in 81 job centers. During the interviews, we conducted a Race Implicit Association Test (IAT), following Greenwald et al. (2009). Additionally, we asked participants about their explicit preferences toward Afro-descendant workers. Finally, following Alesina et al. (2018), we randomized feedback on the IAT results across job centers to test whether information about implicit, sometimes unconscious, bias against/towards Afro-descendants could influence job referral behavior within the employment services. We find that job counselors are 15% less likely to refer Afro-descendants for an opening than non-Afro-descendants. With the use of the IAT, we also show that two-thirds of job counselors revealed strong preferences for whites over blacks. Nevertheless, when asked about perceptions about white and black individuals, job counselors do not directly reveal preferences towards white individuals. By combining the administrative data with the survey to job counselors, we find that the difference in the forwarding rates between Afro- and non- Afro-descendants correlates with the implicit bias of job counselors. In fact, Afro-descendants are less likely to be referred to an opening only if the median job counselor of a given job center has a high level of bias against Afro-descendants. Finally, by using the random feedback of the IAT score, we find that information about their own implicit bias does not change job 1De facto bias refers to preferential treatment as measured by realized behaviors. Implicit bias, as measured by the IAT is understood to be of an unconscious or unknown nature. Explicit bias is measured by preferences individuals are willing to report. 2 counselors’ behavior. We contribute to several strands of the literature. Firstly, we show the distribution and correlation of de facto, implicit, and explicit racial bias among job counselors of the Public Employment Services in Colombia, a middle-income country. We are not the first aiming to understand the role of race in Colombia’s labor markets. Previous works, typically focus on large cities, include studies in Cali, the city with the largest number of Afro-descendants (Diosa Ramírez, 2015; Marulanda and Rodríguez, 2014; Paz Moreno and Delgado Cortez, 2017; Heredia et al., 2010), and Cartagena, the city with the highest proportion of Afro-descendants (Romero-Prieto, 2007). There are also some studies for the two biggest Colombian cities, Bogotá (Garavito et al., 2013), and Medellín (Álvarez Ossa et al., 2014; García Sánchez, 2010). Our research is the first to analyze racial differences at the national level. Furthermore, we joined recent efforts to comprehend racial discrimination in developing countries, particularly in LAC. For instance, a recent study in Brazil found that employer preferences for white workers explain approximately 6% to 7% of the racial wage gap (Gerard et al., 2021). In another study, using data from Brazilian firms in the formal sector, (Miller and Schmutte, 2023) finds strong patterns of co-racial hiring. New firms that are disproportionately comprised of white employees will initially tend to hire white workers, although with persistent growth the hiring becomes more diverse. Additionally, in Mexico, Arceo-Gomez and Campos-Vazquez (2014) shows that indigenous-looking-women received fewer interview requests than mestizo- or caucasian-looking women.2 Secondly, our study adds to recent works measuring implicit bias against different minority groups and its correlation with discriminatory behavior. Glover et al. (2017) examines the dynamics between managers with high levels of unconscious bias against immigrants and immigrant workers in France. Carlana (2019) measures the implicit gender bias of teachers in Italy and shows that the gender gap in math performance increases when students are assigned to math teachers with strong gender stereotypes. Additionally, studies have also used the information on unconscious bias as a low-cost intervention to provoke behavior change. Alesina et al. (2018) found that Italian teachers informed about their implicit bias against immigrants changed their relative grades with respect to immigrant and non-immigrant students. Teachers who had reported explicit bias against immigrants did not change how they graded the students. This suggests that being informed about unconscious bias can help change behavior related to differential treatment, for example, by job counselors. Other studies include Alan et al. (2021) on teachers’ ethnic prejudice in Turkey, and Corno et al. (2022) on racial stereotypes in student dorms in South Africa. Finally, our research contributes to the literature on understanding the role of intermediation services in reducing labor market distortions and increasing welfare for minority groups. 2Other studies include Salardi (2016); Cornwell et al. (2017); Hirata and Soares (2020) in Brazil, Canelas and Gisselquist (2018) in Guatemala, and Garavito et al. (2013) in Colombia, and cross-country analyses such as Chong et al. (2008); Woo-Mora (2022) 3 Recent literature has found positive effects of labor market intermediation services on the probability of employment of job seekers (Crépon and Van Den Berg, 2016), with job search assistance programs more successful for individuals with less access to contributory benefits (Card et al., 2018). Furthermore, the role of public employment services in combating discrimination has been underscored by the International Labor Organization (ILO). Indeed, ILO Convention C111, considered a core labor standard and ratified by all countries in Latin America and the Caribbean, addresses discrimination in employment and establishes a series of actions for Governments, to promote equality and eliminate discrimination in vocational training and placement services (International Labor Organization (ILO), 1958). Public employment services are crucial to follow these recommendations. However, the role of racial bias in labor market intermediation services has received little attention. A notable exception is a recent study in Switzerland that found that rates of contact by recruiters in the Swiss public employment service are 4-19% lower for individuals in minority ethnic groups (Hangartner et al., 2021). Another study that examined the role of ethnicity in labor market intermediation services in Peru, inferred the ethnicity of the job applicant and did not find biased treatment on the part of job counselors (Moreno et al., 2012). Despite finding null effects of IAT information on counselors’ behavior, our experimental framework provides valuable information on the limits of revealing unknown biases in changing attitudes and actions. This paper is organized as follows. After this introduction the next section describes the public employment system in Colombia, followed by of the data in Section 3. Section 4 presents our different estimates of racial bias and their correlations. Section 5 summarizes the results of our experimental intervention and Section 6 concludes. 2 Institutional background – Afro-descendants in the Colombian Labor Market and Public Employment Services (PES). According to the latest report by the Colombian National Statistics Department (DANE), differences between whites and Afro-descendants in the labor market are more related to the quality of jobs than to participation itself (DANE, 2023).3Using data from a national household survey the report found that the occupation rates of whites and Afro-descendants are similar, 89% and 87%, respectively. However, 70.8% of Afro-descendant workers are in the informality, while 55% of white workers have informal jobs. Also, Afro-descendants are less prevalent among professional and directorial jobs. These differences help to explain why the average household monthly income of an Afro-descendant is only 65% of the average income of the household of a white individual (USD 220 and USD 334, for Afro-descendants and whites, respectively). 3In the report whites are described as non-Ethical, which refers to non-Afro-descendant and non-Indigenous individuals. 4 Under this framework, the Colombia Public Employment Service (PES) was established in 2013 to reduce the high unemployment rates in the country and enhance the quality of employment.4The PES provides labor market intermediation services through employment agencies or job centers. The service is offered by local governments and non-profit private institutions called Cajas de Compensación Familiar (CCF), which are funded by contributions from firms. Job counselors in employment agencies are responsible for registering and guiding job seekers and employers. They also undertake pre-selection and referral of candidates to job postings. Interactions among applicants, firms, and job counselors are mostly online, and all CVs and job postings are uploaded to a common online platform. However, job seekers and firms can also attend their local offices to access direct advice from a counselor (Núñez Méndez and Osorio, 2015). Once job seekers and potential employers register on the platform, job counselors evaluate each profile and posting to identify strengths and weaknesses of CVs for different job listings. They then match job seekers with job listings to minimize information gaps. Nevertheless, applicants and firms can also search and apply for jobs or invite job seekers for interviews. Additionally, job counselors can link job seekers to workshops (e.g., motivation, soft skills, support for self-employment), job fairs, and other events that could enhance their employability. Employers also receive advice to facilitate and optimize their recruitment process, providing information on best practices. 3 Data To document different estimates of bias against Afro-descendants in the context of labor market intermediation within PES we use three different data sources.5 3.1 PES administrative records Our first source of data comes from each CV and job vacancy posted at PES by seekers/firms affiliated to a CCF, which are private providers of social services to firms and workers, funded by public resources collected through the payroll of workers (Ministerio de Trabajo and ILO, 2014). We use all CVs from adults 20 to 65 years old, and listings uploaded from March 1st, 2021 to February 28th, 2022. For each CV we observe demographic information such as gender, age, education, and self-identified race. From each vacancy, we observe wage offer, contract type, sector, and each CV that was referred and hired. In total, we observe 348,919 4From 2001 to 2018 the average unemployment rate in Colombia was 11%, 3 pp higher than the average of LAC (Ramos and Álvarez García, 2020). 5Figure E.1 in Appendix E summarizes the study timeline including all data collections and interventions. 5 Figure 2: Differential estimated probability to refer and hire an Afro-descendant CV vs a non-Afro-descendant CV by the median job center’s IAT score, and other job counselors’ characteristics. (A) By job center’s Race IAT Median (B) Probability of referring a CV (Afro - non-Afro) by job center’s counselors’ IAT and quality (C) Probability of hiring a referred CV (Afro - non-Afro) by job center’s counselors’ IAT and quality Notes: Panel A shows the estimated differential probabilities following estimates of the equation Re fiJt =α0+α1A f roi+ α2g(IATJ) + α3A f roi×g(IATJ) + α5A f roi×Xi+Mi+CCFJ+RJt +µiJt. Where Re fiJt =1 if the CV i, managed by job center J, in month twas referred to a job post. A f roi=1 if the individual iidentifies himself as Afro-descendant, and g(IATJ)is a function of the IAT scores of the job counselors at job center J, for example, the median IAT of the counselors of a given job center. Xiis a vector of characteristics of the job-seeker, Miare municipality fixed effects, CCFiare CCF fixed effects, RJt are region-month fixed effects, and µiJt is a random unobservable error term. The area reflects a 95% confidence interval. The vertical dashed lines indicate the critical thresholds suggested by Greenwald et al. (2009). Panels B and C report the differential probabilities following estimates of the equation YiJt =α0+α1A f roi+α2DJ+α3A f roi×DJ+α5A f roi×Xi+Mi+CCFJ+RJt +µiJt, where Yis referring (Panel B) and hiring conditional on being referred (Panel C), Dhas three definitions, and takes value of 1 if (i) job center’s IAT is above 0.65, (ii) job center has at least one counselor with 5 years of more of experience, and (iii) job center has at least one counselor with postgraduate education. Dashed lines represent a 95% confidence interval. 12 to non-Afro-descendant CVs by job counselor’s IAT and quality (at the job center level). The figure shows that job centers with high bias are the ones where Afro-descendants have a lower probability to be referred. When we look at job counselor quality, job centers with less educated and less experienced counselors are the ones that show a negative bias toward Afrodescendant CVs. Additionally, Panel C shows the differential probability of hiring for CVs that were referred by job counselors. In the figure, we can see that the level of implicit bias or counselor’s quality does not create variation in the hiring bias against Afro-descendants. Therefore, our evidence does not support the idea that the discrimination of job counselors against Afro-descendants comes from an effort to internalize the bias of the firms that are hiring. What is more, once a CV has been referred, job counselors’ characteristics seem to play no role in affecting firms discrimination against Afro-descendant workers. 5 Effect of information about implicit bias on race bias Finally, we use the random allocation of information about IAT results to estimate the impact of such information on Afro-descendant referral and hiring rates. We combine data from UAESPE and our vignette study. Unfortunately, all our evidence suggests that knowing about their implicit association bias (IAB), does not change the referring behavior of counselors. Before discussing our results, it is worth noting Table C.1 in Appendix C summarizes the balanced test with respect to job counselors and CV characteristics. The first piece of evidence comes from Figure 3, which shows that for both Afro- and non-Afro-descendant job seekers the referral rates did not change after the IAT feedback to the treatment group finished. We also estimate formally the effect by using data from UAESPE we can use the timing of our intervention to estimate the effect of information on referral rates using a difference-in- difference strategy taking advantage of the random allocation of the IAT feedback. The details of these estimations are in Appendix C. Summarizing, in line with Figure 3, we do not find any changes in referral behavior induced by the bias information. 6 Conclusion Labor market discrimination by race is a topic of great interest with a large knowledge gap in low- and middle-income economies, due mostly to a lack of data. In this paper we documented strong de facto and implicit biases against Afro-descendant job seekers in the Colombian public employment system. While the IAT feedback did not have a strong impact, different from the case of Alesina et al. (2018) intervention in Italy, our study highlights a key part of the labor market where racial bias plays a role in contributing to gaps in employment outcomes for Afro-descendants in Colombia. Better understanding this bias, at the level of intermediation counselors or at the level of employers, can improve the design of future in- 13 Figure 3: Referral rate by month, race, and experimental arm (A) Afro-descendant applicants (B) non-Afro-descendant applicants Notes: Dashed lines represent a 95% confidence interval. Standard errors clustered at the job center level. The period of study implementation between gray vertical lines (from first contact to CCFs to IAT feedback to the treatment job counselors). 14 terventions that aim to prevent or reduce this bias. Moreover, there is a potential for a brief intervention to have sustainable impacts as shown by Miller (2017) who finds that short-run programs can have long-run effects on the racial composition of firms, such as interventions that induce persistent changes in recruitment policies. 15 References Alan, S., E. Duysak, E. Kubilay,and I. Mumcu (2021): “Social Exclusion and Ethnic Segregation in Schools: The Role of Teacher’s Ethnic Prejudice,” The Review of Economics and Statistics, 1–45. Alesina, A., M. Carlana, E. LaFerrara,and P. Pinotti (2018): “Revealing stereotypes: Evidence from immigrants in schools,” Tech. rep., National Bureau of Economic Research. Álvarez Ossa, L. et al. (2014): “Mujeres, pobres y negras, triple discriminación: una mirada a las acciones afirmativas para el acceso al mercado laboral en condiciones de trabajo decente en Medellín (2001-2011),” . Araujo, M. C., S. Duryea,and L. Etcheverry (2022): “Gender and Diversity Sector Framework Document,” Tech. rep., Inter-American Development Bank (IDB), https://www.iadb. org/document.cfm?id=EZSHARE-1011213690-92. Arceo-Gomez, E. O. and R. M. Campos-Vazquez (2014): “Race and marriage in the labor market: A discrimination correspondence study in a developing country,” American Economic Review, 104, 376–80. Blundell, R. W. and J. L. Powell (2004): “Endogeneity in semiparametric binary response models,” The Review of Economic Studies, 71, 655–679. Canelas, C. and R. M. Gisselquist (2018): “Human capital, labour market outcomes, and horizontal inequality in Guatemala,” Oxford Development Studies, 46, 378–397. Card, D., J. Kluve,and A. Weber (2018): “What works? A meta analysis of recent active labor market program evaluations,” Journal of the European Economic Association, 16, 894–931. Carlana, M. (2019): “Implicit stereotypes: Evidence from teachers’ gender bias,” The Quarterly Journal of Economics, 134, 1163–1224. Charles, K. K. and J. Guryan (2008): “Prejudice and wages: an empirical assessment of Becker’s The Economics of Discrimination,” Journal of political economy, 116, 773–809. ——— (2011): “Studying Discrimination: Fundamental Challenges and Recent Progress,” Annual Review of Economics, 3, 479–511. Chong, A., H. Ñopo, L. Ronconi,and M. Urquiola (2008): “The mystery of discrimination in latin america [with comments],” Economía, 8, 79–115. Corno, L., E. LaFerrara,and J. Burns (2022): “Interaction, stereotypes, and performance: Evidence from South Africa,” American Economic Review, 112, 3848–75. 16 Cornwell, C., J. Rivera,and I. M. Schmutte (2017): “Wage discrimination when identity is subjective evidence from changes in employer-reported race,” Journal of Human Resources, 52, 719–755. Coutts, A. (2020): “Racial bias around the world,” . Crépon, B. and G. J. Van Den Berg (2016): “Active labor market policies,” Annual Review of Economics, 8, 521–546. DANE (2023): “Mercado Laboral de los Grupos Étnico-Raciales En Colombia antes y después de los confinamientos por el Covid-19,” Tech. Rep. 2, Departamento Administrativo Nacional de Estadística, https://www.dane.gov.co/files/investigaciones/ notas-estadisticas-casen/abr-2023-Mercado-Laboral-Etnico-Raciales.pdf. Diosa Ramírez, J. (2015): “¿ Causan diferenciación salarial la característica étnica, el género y la ubicación espacial en el mercado laboral de Cali?” Tech. rep. Garavito, C. A. R., J. C. Cárdenas, J. D. O. M., and S. Villamizar (2013): La discriminación racial en el trabajo: Un estudio experimental en Bogotá, Dejusticia. García Sánchez, A. (2010): “Espacialidades del destierro y la re-existencia: Afrodescendientes desterrados en Medellín, Colombia,” . Gerard, F., L. Lagos, E. Severnini,and D. Card (2021): “Assortative matching or exclusionary hiring? the impact of employment and pay policies on racial wage differences in Brazil,” American Economic Review, 111, 3418–57. Glover, D., A. Pallais,and W. Pariente (2017): “Discrimination as a Self-Fulfilling Prophecy: Evidence From French Grocery Stores,” The Quarterly Journal of Economics, 1219– 1260. Greenwald, A. G., T. A. Poehlman, E. L. Uhlmann,and M. R. Banaji (2009): “Understanding and using the Implicit Association Test: III. Meta-analysis of predictive validity.” Journal of Personality and Social Psychology, 97, 17–41. Hangartner, D., D. Kopp,and M. Siegenthaler (2021): “Monitoring hiring discrimination through online recruitment platforms,” Nature, 589, 572–576. Heredia, J. A. C., C. H. O. Quevedo, F. U. Giraldo, C. A. V. López, N. J. S. Alvarado, H. Bonilla, D. M. Cortazar, D. I. Osorio,and N. Páez (2010): “Informe final del proyecto: Desigualdad de oportunidades educativas y segmentación laboral en la población de 15 a 29 años de brasil y colombia, según autoclasificación racial,” . Hirata, G. and R. R. Soares (2020): “Competition and the racial wage gap: Evidence from Brazil,” Journal of Development Economics, 146, 102519. International Labor Organization (ILO) (1958): “C111 Discrimination (Employment and Occupation) Convention, Geneva,” https://www.ilo.org/dyn/normlex/en/f?p= NORMLEXPUB:12100:0::NO::P12100_Ilo_Code:C111. 17 Kline, P., E. K. Rose,and C. R. Walters (2022): “Systemic discrimination among large US employers,” The Quarterly Journal of Economics, 137, 1963–2036. Maccini, S. and D. Yang (2009): “Under the weather: Health, schooling, and economic consequences of early-life rainfall,” American Economic Review, 99, 1006–1026. Marteleto, L. J. (2012): “Educational inequality by race in Brazil, 1982–2007: structural changes and shifts in racial classification,” Demography, 49, 337–358. Marulanda, L. P. and J. J. M. Rodríguez (2014): “La calidad del empleo en la población afrodescendiente colombiana: una aproximación desde la ubicación geográfica de las comunas,” Revista de Economía del Rosario, 17, 315–347. Miller, C. (2017): “The Persistent Effect of Temporary Affirmative Action,” American Economic Journal: Applied Economics, 152–190. Miller, C. and I. Schmutte (2023): “The Dynamic Effects of Co-Racial Hiring,” Tech. rep. Ministerio de Trabajo and ILO (2014): “Libro Blanco del Sistema de Subsidio Familiar,” Tech. rep., International Labor Organization. Moreno, M., H. Ñopo, J. Saavedra,and M. Torero (2012): “Detecting gender and racial discrimination in hiring through monitoring intermediation services: the case of selected occupations in Metropolitan Lima, Peru,” World Development, 40, 315–328. Ñopo, H. (2012): New century, old disparities: Gender and ethnic earnings gaps in Latin America and the Caribbean, World Bank Publications. Ñopo, H., J. Atal,and N. Winder (2010): “New century, old disparities: Gender and ethnic wage gaps in Latin America,” IZA discussion paper. Nosek, B. A., A. G. Greenwald,and M. R. Banaji (2007): “The Implicit Association Test at age 7: A methodological and conceptual review.” . Núñez Méndez, J. and A. F. Osorio (2015): “Evaluación Institucional y de Gestión del Servicio Público de Empleo,” Tech. rep., Fedesarrollo. Paz Moreno, D. P. and S. J. Delgado Cortez (2017): “Inclusión laboral y diversidad de la población afrodescendiente: los desafíos del sector empresarial en Santiago de Cali,” . Quillian, L., A. Heath, D. Pager, A. H. Midtbøen, F. Fleischmann,and O. Hexel (2019): “Do some countries discriminate more than others? Evidence from 97 field of racial discrimination in hiring,” Sociological Science, 6, 467–496. Ramos, C. I. and R. D. Álvarez García (2020): “La tasa natural de desempleo en Colombia 2001-2018: evolución y estimaciones,” Entramado, 16, 76–93. 18 Romero-Prieto, J. E. (2007): “¿ Discriminación laboral o capital humano?: determinantes del ingreso laboral de los afrocartageneros,” Documentos de Trabajo Sobre Economía Regional y Urbana; No. 98. Salardi, P. (2016): “The Evolution of Gender and Racial Occupational Segregation Across Formal and Non-Formal Labor Markets in Brazil, 1987 to 2006,” Review of Income and Wealth, 62, S68–S89. Woo-Mora, G. (2022): “Unveiling the Cosmic Race: Skin tone and ethnoracial inequalities in Latin America,” Tech. rep., Working paper, Paris School of Economics. 19 A IAT resources • Content: Implicit Race Association Test (IAT) – International Standard Version • Source: https://implicit.harvard.edu/implicit/Study?tid=-1 • Words: Category items A. In English: Good Joy, Love, Peace, Wonderful, Pleasure, Glorious, Laughter, Happy Bad Agony, Terrible, Horrible, Nasty, Evil, Terrible, Failure, Pain B. In Spanish Good Alegría, Amor, Paz, Maravilloso, Placer, Glorioso, Risa, Feliz Bad Agonía, Terrible, Horrible, Desagradable, Malvado, Malísimo, Fallo, Dolor • Pictures: –Black faces: –White faces: 20 B IAT results feedback The feedback to job counselors happened in two instances. First we sent an e-mail (following Alesina et al., 2018), including the exact numerical value of the test, the category the are placed according to Greenwald et al. (2009). We complement the e-mail with a phone call in order to clarify doubts. In this section we show the e-mail template and the script followed by our call center. Given that both, e-mail and calls, were personalized, in this section you will find XXX where the score was placed. In addition, in the email, we place an "X" in the respective box in the scale figure, and give one of the following PERSONALIZED MESSAGE, depending on the job counselor’s result: • High preference towards white people: You have a high unconscious tendency to associate positive things with white people and negative things with Afro-descendant people. This could lead to a strong preference for white people over Afro-descendant people in different areas of life (for example, work, personal, among others). • Intermediate preference towards white people: You have an intermediate and unconscious tendency to associate positive things with white people and negative things with Afro-descendant people. This could lead to a preference for white people over Afrodescendant people in different areas of life (for example, work, personal, among others). • Low preference towards white people: You have a low unconscious tendency to associate positive things with white people and negative things with Afro-descendant people. This preference could lead to a slight preference for white people over Afro-descendant people in different areas of life (for example, work, personal, among others). • Neutral: You are neutral or do not have unconscious tendencies to associate positive or negative things with white or Afro-descendant people. This could lead to not having preferences between Afro-descendant or white people in different areas of life (for example in work, personal, among others). • Low preference towards Afro-descent people: You have a low unconscious tendency to associate positive things with Afro-descendant people and negative things with white people. This could lead to a slight preference for Afro-descendant people over white people in different areas of life (for example, work, personal, among others). • Intermediate preference towards Afro-descent people: You have an intermediate and unconscious tendency to associate positive things with Afro-descendant people and negative things with white people. This could lead to a preference for Afro-descendant people over white people in different areas of life (for example, work, personal, among others). • High preference towards Afro-descent people: You have a high unconscious tendency to associate positive things with Afro-descendant people and negative things with white 21 Table D.3: First stage estimates for different definitions of the instrument Outcome: Probability of self-identify Afro-descendant Reference population: All adults Economic active Employed adults adults Age cell size (years): 5 10 5 10 5 10 (1) (2) (3) (4) (5) (6) Panel A. All applicants Afro-descendant proportion by cell 0.090∗∗ 0.091∗∗ 0.091∗∗ 0.091∗∗ 0.092∗∗ 0.092∗∗ (0.022) (0.027) (0.022) (0.027) (0.022) (0.027) F-Test 16.610 11.124 17.003 11.319 16.857 11.209 N 156053 156053 156053 156053 156053 156053 Panel B. Only forwarded applicants Afro-descendant proportion by cell 0.104∗∗ 0.106∗∗ 0.104∗∗ 0.104∗∗ 0.108∗∗ 0.108∗∗ (0.036) (0.040) (0.035) (0.039) (0.036) (0.041) F-Test 8.585 6.947 8.649 6.915 8.853 7.067 N 59905 59905 59905 59905 59905 59905 Notes: Standard errors clustered at municipality-gender-age cell level in parentheses. +p<0.1, ∗p<0.05, ∗ ∗ p<0.01. Only includes CVs that entered the system between March 1st and August 5th, 2021. All estimations control for age, experience, and educational level, and include municipality and CCF-month of registry fixed effects. The F-Test represents the F statistic from the null hypothesis that ϕ1=. 28 Table D.4: Effect of Afro-descendant CV on the probability of being forwarded by Sector and job-type. Pre-study By sector All Services Commerce Manufacture Others (1) (2) (3) (4) (5) Reduced form estimates -0.220∗∗ -0.237∗∗ -0.017∗∗ -0.027∗∗ 0.033∗∗ (0.035) (0.034) (0.005) (0.005) (0.010) IV estimates -2.449∗∗ -2.637∗∗ -0.194∗∗ -0.297∗∗ 0.368∗ (0.756) (0.780) (0.071) (0.109) (0.152) Mean non-afro 0.280 0.197 0.036 0.040 0.087 FS-F First Stage F-Test 16.610 16.610 16.610 16.610 16.610 R2-0.459 -0.697 -0.001 -0.024 -0.013 Observations 156053 156053 156053 156053 156053 By job type All By wage By contract type ≤1 MMLW >1 MMLW Short term Long term (1) (6) (7) (8) (9) Reduced form estimates -0.220∗∗ -0.032+-0.215∗∗ -0.090∗∗ -0.158∗∗ (0.035) (0.018) (0.035) (0.019) (0.036) IV estimates -2.449∗∗ -0.360 -2.386∗∗ -0.996∗∗ -1.759∗∗ (0.756) (0.220) (0.745) (0.343) (0.613) Mean non-afro 0.280 0.176 0.172 0.184 0.171 FS-F First Stage F-Test 16.610 16.610 16.610 16.610 16.610 R2-0.459 0.032 -0.634 -0.084 -0.321 Observations 156053 156053 156053 156053 156053 Notes: Standard errors clustered at municipality-gender-age cell level in parentheses. +p<0.1, ∗p<0.05, ∗ ∗ p<0.01. Only includes CVs that entered the system between March 1st and August 5th, 2021. All estimations control for age, experience, and educational level, and include municipality and CCF-month of registry fixed effects. Afro-descendant identification instrumented with the proportion of afro-descendant population at the municipality-gender-age cell (5 years) at the 2018 census – As in Table D.3 Panel A column 1. The F-Test represents the F statistic from the null hypothesis that ϕ1=. 29 Table D.5: IV-Probit estimated effect of Afro-descendant CV on the probability of being forwarded by Sector and job-type. Estimated difference P(Afro-descendant) - P(Non-Afro) Coefficient s.e. Non-Afro mean Observations (1) (2) (3) (4) All -0.054 (0.008)∗∗ 0.283 154441 By Wage Less than 1 MMLW -0.023 (0.007)∗∗ 0.179 153337 More than 1 MMLW -0.000 (0.008) 0.174 153039 By Contract type Short term -0.014 (0.008)+0.187 153297 Long term -0.009 (0.007) 0.174 152888 By Sector Services -0.006 (0.008) 0.200 153751 Commerce -0.022 (0.003)∗∗ 0.040 141501 Manufacture 0.016 (0.006)∗∗ 0.043 144929 Others 0.023 (0.006)∗∗ 0.090 150569 Notes: Bootstrap standard errors after 1000 repetitions in parentheses. +p<0.1, ∗p<0.05, ∗ ∗ p<0.01. Only includes CVs that entered the system between March 1st and August 5th, 2021. All estimations control for age, experience, and educational level, and include municipality and CCF-month of registry fixed effects. Afro-descendant identification instrumented with the proportion of afro-descendant population at the municipality-gender-age cell (5 years) at the 2018 census. To account for endogenous reporting we use a control function approach by introducing the error term from the first stage estimations into the Probit estimation. 30 E Additional tables and figures Figure E.1: Study timeline 2021 · · · · · ·• February (1st) UAESPE data first entry. · · · · · ·• August (5th) Invitation to CCFs. (27th) Baseline survey starts. · · · · · ·• September (13th) Baseline survey ends. (20th) Random allocation of job centers. (23th to 29th) IAT score feedback to treated job counselors. · · · · · ·• October (12th) Vignette study starts. · · · · · ·• November (9th) Vignette study ends. 2022 · · · · · ·• February (1st) UAESPE data last entry. (21st to 25th) IAT score feedback to control job counselors. Figure E.2: IAT score distribution by job counselors’ gender and race (A) By gender (B) By race Notes: Race IAT scores. A positive value indicates a stronger association between "white"-"good" and "black"-"bad". The vertical dashed-lines indicate the critical thresholds suggested by Greenwald et al. (2009). 31 Figure E.3: Proportion of Afro-descendant job seekers by Department Notes: Proportion of Afro-descendants among all seekers and hired seekers by Department of residence of the job seeker. We use data from UAESPE of all job seekers 20 to 65 years old registered between 01/March/2021 and 28/February/2022. We excluded the CCF COMFAGUAJIRA as its behavior distances from all other CCFs with respect the Afro-descendant CVs registration and referrals. Table E.6: Effect of race on the probability of CV-forwarding or hired by job-offer type. Pre-study Any job By wage By contract type ≤1 MMLW >1 MMLW Short term Long term (1) (2) (3) (4) (5) A. Probability of being forwarded Afro-descendant -0.041+-0.029 -0.018 -0.016 -0.026 (0.021) (0.018) (0.015) (0.015) (0.016) Mean non-afro 0.280 0.176 0.172 0.184 0.171 R20.172 0.163 0.127 0.155 0.112 Observations 156,053 156,053 156,053 156,053 156,053 B. Probability of being hired | being referred Afro-descendant -0.055∗∗ -0.074∗∗ -0.008 -0.065∗∗ -0.011 (0.020) (0.026) (0.005) (0.022) (0.008) Mean non-afro 0.239 0.260 0.040 0.214 0.047 R20.181 0.248 0.045 0.230 0.067 Observations 59,905 35,915 133,456 39,215 131,624 Notes: Standard errors clustered at job center level in parentheses. +p<0.1, ∗p<0.05, p<0.01. Only includes CVs that entered the system August 5th, 2021. All estimations control for age, zone, experience and educational level, and include region-month of registry and city of residence fixed effects. 32 Figure E.4: Explicit bias. Evaluation of whites and Afro-descendants with respect to intelligence and laziness (A) Intelligence (B) Laziness Notes: Information from direct interviews to job counselors. Figure E.5: Proportion of job counselors that would not like to have specific populations as neighbors Notes: Proportion of individuals who reply yes to the question – among the following groups, who would you not like to have as neighbors?. 33 Figure E.6: Differential estimated probability to forward an Afro-descendant CV vs a non-Afro-descendant CV by the minimum, maximum, and mean IAT score in a given job center. (A) Minimum (B) Mean (C) Maximum Notes: Base on the estimated probabilities following estimates of the equation Re fiJt =α0+α1A f roi+α2g(IATJ) + α3A f roi× g(IATJ) + α5A f roi×Xi+Mi+CCFJ+RJt +µiJt, where gis the minimum, mean, and maximum respectively. Area reflects a 95% confidence interval. Only includes CVs that entered the system August 5th, 2021. All estimations control for age, zone, experience and educational level, and include region-month of registry and city of residence fixed effects. The vertical dashedlines indicate the critical thresholds suggested by Greenwald et al. (2009). 34 Figure E.7: Differential estimated probability to forward a CV with respect to race, gender and age, by job center IAT score. Pre-study Notes: Report the differential probabilities following estimates of the equation YiJt =α0+α1CATi+α2DJ+α3CATi×DJ+ α5CATi×Xi+Mi+CCFJ+RJt +µiJt, where Yis referring a CV, and CAT is A f roifor Race, Womanifor gender, and [Age <32] for Age. Dashed lines represent 95% confidence interval using the Delta method. Figure E.8: Referring and hiring probabilities by job center characteristics (A) Probability of referring a CV by job center characteristics (B) Probability of being hired by job center characteristics and referring status Notes: Dashed lines represent 95% confidence interval. Standard errors clustered at job center level in parentheses. Predicted conditional probabilities controlling by race, gender, age, zone, experience, education, and municipalities, CCF and month fixed effects. 35