Teachers' preferences for proximity and the implications for staffing schools: Evidence from Peru

Bertoni, Eleonora; Elacqua, Gregory; Hincapié, Diana; Méndez, Carolina; Paredes, Diana Working Paper Teachers' preferences for proximity and the implications for staffing schools: Evidence from Peru IDB Working Paper Series, No. IDB-WP-1073 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Bertoni, Eleonora; Elacqua, Gregory; Hincapié, Diana; Méndez, Carolina; Paredes, Diana (2019) : Teachers' preferences for proximity and the implications for staffing schools: Evidence from Peru, IDB Working Paper Series, No. IDB-WP-1073, Inter-American Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0001977 This Version is available at: https://hdl.handle.net/10419/208209 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-nc-nd/3.0/igo/legalcode Teachers’ Preferences for Proximity and the Implications for Staffing Schools: Evidence from Peru Eleonora Bertoni Gregory Elacqua Diana Hincapié Carolina Méndez Diana Paredes IDB WORKING PAPER SERIES No IDB-WP-01073 Inter-American Development Bank Education Division October 2019 Teachers’ Preferences for Proximity and the Implications for Staffing Schools: Evidence from Peru Eleonora Bertoni Gregory Elacqua Diana Hincapié Carolina Méndez Diana Paredes Inter-American Development Bank Education Division October 2019 Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library Teachers’ preferences for proximity and the implications for staffing schools: evidence from Peru / Eleonora Bertoni, Gregory Elacqua, Diana Hincapié, Carolina Méndez, Diana Paredes. p. cm. — (IDB Working Paper; 1073) Includes bibliographic references. 1. Teachers-Selection and appointment-Peru. 2. Teachers-Supply and demand- Peru. 3. Public schools-Peru. I. Bertoni, Eleonora. II. Elacqua, Gregory M., 1972- III. Hincapié, Diana. IV. Méndez, Carolina. V. Paredes, Diana. VI. Inter-American Development Bank. Education Division. VII. Series. IDB-WP-1073 http://www.iadb.org Copyright © 2019 Inter-American Development Bank. This work is licensed under a Creative Commons IGO 3.0 Attribution- NonCommercial-NoDerivatives (CC-IGO BY-NC-ND 3.0 IGO) license (http://creativecommons.org/licenses/by-nc- nd/3.0/igo/legalcode) and may be reproduced with attribution to the IDB and for any non-commercial purpose, as provided below. No derivative work is allowed. Any dispute related to the use of the works of the IDB th at cannot be settled amicably shall be submitted to arbitration pursuant to the 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 CC-IGO license. Following a peer review process, and with previous written consent by the Inter-American Development Bank (IDB), a revised version of this work may also be reproduced in any academic journal, including those indexed by the American Economic Association’s EconLit, provided that the IDB is credited and that the author(s) receive no income from the publication. There fore, the restriction to receive income from such publication shall only extend to the publication’s author(s). With regard to such restriction, in case of any inconsistency between the Creative Commons IGO 3.0 Attribution-NonCommercial- NoDerivatives license and these statements, the latter shall prevail. Note that link provided above includes additional terms and conditions of the license. 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. [email protected] [email protected] 1 Abstract* This paper explores rank-ordered teacher candidates’ preferences for public schools in Peru by analyzing the 2015 teacher hiring process. Our analysis shows that, in seeking permanent positions in public schools, candidates appear to search closer to where they attended their Teacher Education Program (TEP) and prefer to work in urban areas. Moreover, candidates seem to prefer schools with higher enrollment, basic services and located in wealthier areas. These preferences vary by candidates’ attributes. Proximity from their TEP seems to be particularly important for females, while urbanicity is more relevant for candidates with high scores in the national teacher test and older than 35 years old. When controlling for previous workplace location, TEP locations and urbanicity play a less important role in teacher preferences. Understanding which school characteristics teachers value the most can help us design new policies and modify the existing ones to attract teachers to hard-to-staff schools. JEL Classification: H75, I24, J38, N36 Keywords: Teacher hiring, Teacher preferences, Teacher labor markets, Peru * We are thankful to Brenda Teruya for providing outstanding research assistance. 2 1. Introduction There is considerable evidence that teachers are important for improving educational quality and narrowing racial or socioeconomic achievement gaps (Araujo et al., 2016; Chetty et al., 2014; Hanushek y Rivkin, 2012)1. Yet, hiring qualified and effective teachers remains one of the biggest challenges in education. This challenge is even more pressing in schools that serve disadvantaged students, given that evidence shows that effective teachers have a greater academic impact on the lowest performing students (Rivkin, Hanushek & Kain, 2005). Peru is an example of a school system that struggles to attract high quality teachers to vulnerable schools. In 2015, from a total of 19,630 vacancies advertised for permanent teaching positions, 40% of the openings did not receive any applications. The vacancies that did not have candidates vary by geographical and socioeconomic level and are concentrated in the most disadvantaged areas of the country. More than 50% of unselected school vacancies were located in the two highest quintiles of district poverty2, and 95% were concentrated in schools in rural areas. The Loreto region (located in the Amazon Rainforest) alone3 accounts for almost 20% of the unselected vacancies. These unselected vacancies usually end up being assigned to temporary teachers. This might be worrying given that there is some evidence suggesting that teachers with temporary contracts can have a negative influence on student learning (Ayala & Sánchez, 2016), especially on disadvantaged students (Marotta, 2019). Moreover, in Peru, most of the temporary teachers that end up occupying “undesired” vacancies are low performing teachers who did not pass the national teacher test (Prueba Unica Nacional - PUN).4 In 2016, 69% of temporary teachers hired in unselected vacancies did not achieve the minimum score on the PUN and 27% of them did not even take the test. Researchers can gain insight into this topic by examining Peru’s national teacher assignment process. This paper analyzes teacher candidates’ preferences5 for public schools in Peru in the centralized assignment system by answering the following research questions: i) Which school characteristics drive candidates’ preference ranking? and ii) How do these preferences vary according to different candidates’ characteristics? To answer these questions, 1 See also Rivkin et al., 2005 and Rockoff, 2004. 2 While in the two lowest quintiles of district poverty the unselected vacancies amounted to 26%. 3 Peru is divided into 26 regions (24 departments and 2 provinces with special regimes, the province of Lima and the constitutional province of Callao). 4 As per the teacher hiring process defined by the 2012 Law of Magisterial Reform, to be able to apply for a position, teacher candidates need to achieve the minimum score on each of the three sub-tests of the national teacher test (PUN): (1) Logical reasoning (25%); (2) Reading comprehension (25%); and (3) Pedagogical knowledge of the specialization (50%). 5 More precisely, the teacher candidates that we are examining here are only those who passed the PUN and, thus, could list their preferences for a set of vacancies. 3 we take advantage of the unique information on stated teacher candidates’ preferences provided by the Peruvian Ministry of Education in the 2015 Teacher Hiring Process (Concurso de Nombramiento). In Peru, when applying for a position as a permanent teacher, candidates rank school vacancies according to their order of preference. We take advantage of the detailed information on teacher candidates’ ordered preferences and estimate a rank-ordered logit model. This model allows us to analyze how teacher candidates evaluate different vacancies’ characteristics when constructing their ranking. Our results reveal that, in seeking permanent positions in public schools, candidates appear to search closer to where they attended their Teacher Education Program (TEP) and prefer to work in urban areas (or closer to their province’s capital). These preferences vary by candidates’ attributes. Proximity from TEP seem to be particularly important for females, while the urban location is more relevant for candidates with PUN scores in the highest quintile and who are older than 35 years old. In addition, consistent with the literature, candidates prefer larger schools, located in low-poverty districts and with access to basic services. The literature on teacher labor markets and teacher preferences shows that teachers sort according to specific school characteristics. More precisely, studies in the United States have shown that teachers prefer schools that are closer to their hometown or to where they concluded their teacher education program (Boyd et al., 2005; Engel et al., 2014).6 Additionally, teachers sort according to student socioeconomic level (Krieg et al., 2016; Boyd et al., 2010; Lankford et al., 2002), student achievement (Boyd et al., 2010; Krieg et al., 2016, Lankford et al., 2002) and tend to prefer schools with better working conditions (Ronfeldt, 2012; for the Netherlands: Bonhomme et al., 2016). At the same time, these preferences often vary according to the teachers’ characteristics. For example, female candidates have been found to prefer schools that are closer to their hometown or to where they concluded their teacher education program (Boyd et al., 2005; Krieg at al., 2016), and more qualified teachers are more willing to move further away from their hometown (Boyd et al., 2005), while more experienced teachers tend to stay closer to their teacher education program (Krieg at al., 2016). In Latin America, the literature on teachers’ labor markets is limited to the work of Jaramillo (2013) for Peru and Rosa (2017) for the City of Sao Paulo, Brazil. The former, by collecting survey data in two regions of the country (i.e. Loreto and Lambayeque), suggests the presence of highlyregionalized and low-mobility teacher labor markets in Peru, where almost 80% of the sampled teachers worked in their region of birth or in the region where they graduated from college. 6 This is in line with what Reininger (2012) found for a set of US States. The author finds that teachers are more likely to be local (live close to their high school hometown) than college graduates in other occupations. 4 Moreover, teachers in these two regions rarely moved to a different school over their 12-year career (on average). The study by Rosa (2017) analyzes one-sided matching7 in teacher labor markets in Sao Paulo by estimating a conditional logit model to examine school attributes that are associated with teacher choices. The study provides evidence that teacher choices are largely related to school location, students’ socioeconomic characteristics and school quality. This paper contributes to the literature on teacher labor markets in several ways. First, to the best of our knowledge, this is the first large-scale empirical study on teacher candidates’ stated preferences in Latin America. Indeed, the only two other studies focusing on teacher preferences in Latin America restrict their analysis to a single city (Rosa, 2017) and to a sample of regions (Jaramillo, 2013). On the contrary, our analysis is based on a large census of teacher applicants moving through a national centralized admission system. Second, this is the first analysis of the detailed rank-ordered preferences of the 2015 teacher national contest in Peru. The literature on teacher preferences mainly focuses on empirical studies that examine the attributes that determine a candidate’s final job allocation (Boyd et al., 2005; Engel et al., 2014; Rosa, 2017), or studies that rely on interview data and teachers’ self-reported preferences (Burns et al., 2008; Ronfeldt et al., 2014; Rots et al., 2007). In contrast to these studies, our work relies on teachers’ stated preferences for schools. Understanding which school characteristics teachers value the most when they apply to a teaching position can help policymakers effectively tackle the staffing challenges of the most disadvantaged schools in the country. Education systems in Latin America have implemented different policies to face teacher shortages in hard-to-staff schools. Several countries, including Peru, provide monetary incentives to teachers that work in rural and remote schools. In Chile and Mexico these incentives are higher as the teacher advances in the career path. Countries have also introduced non-monetary incentives for teachers who work in hard-to-staff schools, such as shorter time requirements to apply for a promotion, more flexibility in their teaching schedule, and more training opportunities. To address teacher shortages in rural schools, some countries have established cooperation systems between schools (Chile and Colombia), hybrid classrooms (Pará, Brazil), and programs to strengthen TEP in rural areas (Colombia and Peru). The rest of the paper is organized as follows. Section 2 describes the institutional context of the Peruvian public-school system. Section 3 introduces the data used in this study and presents descriptive statistics. Section 4 presents the empirical strategies employed in the analysis. Results are presented in section 5. Finally, section 6 concludes and discusses some policy implications. 7 A one-side matching process is a process in which teachers are completely free to choose the school in which they work. It differs from a two-sided matching where school administrators have the power to refuse teachers. 5 2. Institutional Context 2.1. Teacher Hiring Process in the Peruvian Public-School System In 2015, the Peruvian government implemented a new teacher evaluation that was required to obtain a tenured teaching position in the public-school system. To be eligible to apply for a teaching position, candidates had to hold a bachelor’s degree in education. The evaluation consisted of two stages: a national stage and a decentralized stage. The national stage is carried out by the Ministry of Education (MINEDU) and includes a standardized written test (PUN) divided into three sub-tests: logical reasoning (25%), reading comprehension (25%), and pedagogical knowledge of the specialization (50%). To pass the national stage, candidates need to obtain at least 60% of the questions correct on each sub-test. Applicants are evaluated within a specific area of specialization by the education level (preprimary/primary/secondary) and subject (e.g. Secondary-Sciences) they plan to teach. Only those candidates who score above the threshold required on the national stage can establish their school preferences within their area of specialization and within one of the 26 regions of Peru. There are two school selection rounds. In the first round, candidates can rank up to 5 school preferences. Then, the Ministry of Education assigns each candidate a maximum 2 out of their 5 preferred schools, based on their PUN score and their preferences ranking. Candidates that missed the first round or were not assigned to any of their school preferences during the first round can participate in the second round. Each vacancy can have up to 20 candidates.8 A candidate with a relatively lower score in the national stage may be less competitive and not be assigned to any of their 5 preferences in the first round. In the second round, there are no limitations with respect to the number of preferences they can list. Once candidates have been assigned to up to 2 of their preferred schools, they enter the decentralized stage, which is carried out by each school or by the local education administrative units (Unidad de Gestión Educativa Local - UGEL) in the case of single-teacher institutions. The decentralized stage includes an evaluation of their resume (25%), a personal interview (25%), and a classroom observation (50%). To pass the decentralized stage, candidates need a score of 30 points (out of 50) in the classroom observation component. Finally, the Ministry of Education used the weighted sum of the scores obtained in the national and decentralized stages (the national stage has a weight of 67% on the final score) to 8 One school can have more than one vacancy in the same area of specialty. 12 Because the utility function is partly stochastic, the probability of candidate 𝑖 to rank first alternative 𝑗∗ from 𝐶 may be written as: 𝑃∗= 𝑃𝑈∗≥ 𝑈,𝑗 = 1,2,…,𝐽 =𝑃(𝜀 −𝜀∗≤ 𝑥󰆒∗− 𝑥󰆒, 𝑗 =1,2,…,𝐽) If the stochastic error terms are assumed to be identically and independently distributed (IID) according to the double exponential distribution, one can show that the choice probabilities have the following form (McFadden,1974): 𝑃∗=exp (𝑥∗ 󰆒 𝛽) ∑exp (𝑥 󰆒 𝛽   ) If one applies the Ranking Choice Theorem (Luce and Suppes, 1965) to the stochastic utility model, assuming that the alternative index j is a serial preference index, it follows that: 𝑃𝑈 ≥ 𝑈 ≥⋯ ≥ 𝑈= 𝑃(𝑈∗≥ 𝑈,𝑗 =𝑗∗,…,𝐽)  ∗ Where 𝑃𝑈 ≥ 𝑈 ≥⋯ ≥ 𝑈 is the joint probability that alternative 1 is preferred to alternative 2 which is preferred to alternative 3, and so on to alternative 𝐽−1 which is preferred to alternative 𝐽 for candidate 𝑖, and 𝛽 representing the relative importance of the vacancies’ characteristics to the sample of candidates. The probability that a candidate 𝑖 submits a particular ranking on schools within a region and specialization area is a product of standard logit formulas (Train, 2009; Hastings et al., 2006). 4.1. Potential for Strategic Choice Candidates that pass the national stage, which involves a standardized written test (PUN), can select up to 5 preferred schools. They are assigned up to 2 schools, based on their PUN score and school preferences. Each vacancy can have up to 20 candidates. Almost 15% of the 10,392 selected vacancies had more than 20 interested candidates (i.e. the candidate ranked the vacancy). Within each region and area of specialization, candidates are ranked based on their PUN scores. The candidate with the highest PUN score will be assigned to its top two choices. The candidate with the lowest PUN score would not be assigned to any school, if each of the 5 schools that she selected already had 20 candidates. Out of the 23,319 candidates that 13 established their school preferences in the first round of selection, 86% were assigned to their first choice, and only 4% were not assigned to any of their preferred schools.22 This mechanism may create incentives for candidates with low PUN scores to misstate their preferences, not listing their most preferred schools if they have a low probability of competing for the vacancy or obtaining a permanent position. The PUN score has a weight of 67% in the final score, and the candidate with the highest final score (PUN score plus decentralized stage score) are granted a permanent position. A candidate with a low PUN score, relative to other candidates in the same region and area of specialization, may strategically apply to schools that are less attractive and therefore will likely have fewer candidates, in order to increase the chances of obtaining a permanent position. Candidates could also attempt to size up their competition before submitting their rankorder list of schools. The list of candidates who passed the PUN, including their disaggregated PUN scores, the region where they took the test, and their area of specialization are publicly available. Candidates compete for permanent positions within a region and an area of specialization. The region where candidates took the test may be different from the region where they apply for a position, and candidates cannot see the regions of application of other candidates.23 However, if they assume that most candidates remain in the same region, they could size up their competition by adding up the disaggregated PUN scores and calculating a ranking by themselves. There are some factors that might hinder strategic behavior. First, the novelty of the contest reduces the chances of strategic hedging. Since 2015 was the first year in which the contest was implemented24, candidates might not have known exactly how the slots for the decentralized stage and permanent positions were assigned or how to calculate their rank position within their region and area of specialization. Even though candidates may not have listed their most preferred schools, the ranking might still reflect their preferences among the selected schools. In other words, the first ranked school should be the most preferred alternative among the selected schools, the second ranked school the most preferred alternative among the rest, and so on. In this paper, we analyze teacher preferences among ranked schools, giving us 22 Nevertheless, candidates who are not assigned to any school for the decentralized stage after the first selection round can participate in a second selection round, in which they can select among schools with remaining vacancies within their area of specialization. 23 Out of the 23,319 candidates who passed the PUN and selected vacancies, 90% chose the same region where they took the PUN. 24 The previous teacher law that was in place between 2007 and 2012 (Ley de Carrera Magisterial) held hiring contests in 2009 and 2011. The general structure of the contests was similar to the 2015 contest (with a national and decentralized stage), but there were differences in some instruments and their weights. For instance, in the 2009 contest, the PUN score had a weight of 50% (vs. 67% in the 2015 contest). 14 information on which school characteristics are associated with a higher rank in their preference set.25 5. Results 5.1. Preference Parameter Estimates Table 7 shows the point estimates from the rank-order logit model. The estimates represent the estimated changes in a candidate’s utility for a unit change in the exogeneous variables (Punj and Staelin, 1978). We investigate the importance candidates give to each school characteristic by assessing their sign, relative magnitude, statistical significance, and stability across specifications (Beuermann et al., 2018). Table 7 presents four specifications: the first two only include school characteristics, and the last two add interactions between school characteristics and candidate attributes. We find that teacher candidates prefer schools (i.e. are better ranked) with higher enrollment and basic services (Column 1 of Table 7). In addition, candidates prefer schools that are closer to their teacher education program (TEP) and that are located in less poor districts. Out of the 23,046 candidates in our sample, 65% selected vacancies in the same region where they studied. The significance of the distance from TEP in shaping the decision is consistent with previous literature on the determinants of teacher’s initial job placements (Boyd et al. 2005; Krieg, Theobald and Goldhaber, 2016). Regarding the urban and rural categories, the rurality base category is composed by the most rural schools (rural 1), which are schools located in areas with less inhabitants and furthest away from the province capital. The coefficients for the least rural (rural 3) and urban schools have significantly positive effects on candidate’s utility, with a bigger effect for urban schools, signaling their preference for more urban locations. The coefficient for moderate rural schools (rural 2) is not statistically significant, suggesting that candidates are indifferent between the most rural (rural 1) and moderate rural (rural 2) schools. This last result could be due to that fact that most urban and moderate rural schools share similar constraints in living conditions. By assessing the sign and magnitudes of the significant point estimates, we can describe scenarios in which candidates are indifferent between different schools’ types (i.e. their utility would be the same). We analyze what would it take for a candidate to choose a least rural (rural 3) school in place of an urban school, conditional on other school characteristics being the same. We find that candidates would be willing to work in a rural 3 (least rural) school instead of an urban school conditional on lower poverty rates, higher enrollment or shorter distance from 15 TEP. Comparing the coefficients on poverty, urban and least rural, the point estimates suggest that candidates would be indifferent between working in a rural 3 (least rural) school and an urban school when the poverty rate of the least rural school is 10 percentage points lower than the urban one. A comparison with enrollment implies that candidates would be indifferent between working in a rural 3 (least rural) school and an urban school when the enrollment of the least rural school is approximately 490 students higher than the urban one. A comparison with the distance from TEP shows that candidates would be indifferent between working in a rural 3 (least rural) school and an urban school when the distance between the least rural school and TEP is 9km shorter than the distance between the urban school and TEP. Previous research from the United States suggests that teachers prefer jobs that are closer to their residential location (Engel, Jacob and Curran, 2014; Killeen, Loeb and Williams, 2015; Hanson and Johnston, 1985; Hanson and Pratt, 1988). To reduce the potential that home residence might be endogenous to employment opportunities, other studies include proxies of residency, such as high-school location (Boyd et al., 2005; Reininger, 2012). We neither have information on candidates’ hometown location nor on their residential location26, but we do have information on the workplace location for the subsample of candidates that were working as temporary teachers in public schools in 2015.27 The estimates in Column 2 show that candidates prefer schools that are closer to their previous workplace location.28 The inclusion of distance from the previous workplace eliminates the urban/rural effects and reduces the importance of distance from TEP location. Under this specification, urbanicity is no longer significant, which implies that candidates’ previous workplace location is a better predictor of school ranking. In addition, by analyzing the magnitude of the coefficients, we find that the distance from TEP coefficient is 6 times smaller than in previous specifications. This finding highlights the importance of candidate’s previous workplace location relative to their TEP location, particularly for this subsample of candidates that has, on average, 8 years of experience. Out of the 14,220 candidates with previous workplace information, 93% selected vacancies in the same region where they worked in 2015, 67% selected vacancies in the same region where they studied, and 65% in the same region where they studied and worked.29 Most 26 Jaramillo (2013) analyses two regions of Peru: Lambayeque and Loreto and found that most teachers work in the same region where they were born (77.0% and 85.9%, respectively). 27 In 2015, 30% of the teachers in the public sector were temporary teachers (Nexus, 2015). Moreover, out of the 23,046 candidates that selected vacancies for a permanent position, 61% were temporary teachers in public schools. 28 Assuming that, on average, temporary teachers choose the closest school to their residence. We cannot distinguish between the preferences of the teachers and those of the hiring schools in determining their workplace as temporary teachers. However, given the temporality of the contract (1 school year), teachers might have fewer incentives to move to another region or too far away within a region to work as temporary teachers. 29 Even though we do not have previous workplace information for 39% of the candidates, we would expect similar preferences for proximity to their previous workplace over proximity to TEP. These candidates have, on average, 7 years of experience, and 70% of them select schools in the same region where they studied. 16 candidates do not consider moving to another region different from where they studied, which implies that teacher labor markets are generally geographically segmented in Peru. Since most teachers decide to work in the same region where they studied, it is important to develop policies to increase the local supply and quality of teachers in high-deficit regions. Candidates seem to be willing to work further away from their previous workplace to teach in schools with basic services and located in wealthier areas. A comparison of the coefficients on basic services and distance from their previous workplace implies that candidates are willing to work almost 4 km away from their previous workplace to teach in a school with basic services. Comparing the coefficients on poverty and distance from their previous workplace, the point estimates suggest that candidates are willing to work 6km away from their previous workplace to teach in a school in the 25th percentile of poverty than in one in the 50th percentile. Next, in order to better understand the coefficients, we compare two hypothetical schools (A and B) which share all characteristics except one. The probability of preferring a school with basic services to one without them is 51%. For continuous variables, Figure 3 shows the probability of preferring school A to school B, as the analyzed characteristic change values for school B. Figures 3.1, 3.3 and 3.4 show that, as the school B is farther away from TEP, farther away from the previous workplace and has higher poverty relative to school A, respectively, the probability of preferring school A increases. Figure 3.2 shows that, as the school B has a higher enrollment than school A, the probability of preferring school A decreases. In Table A230 we present additional specifications for the subsample of candidates that applied to primary schools and whose selected schools have 2014 National Student Evaluation (ECE) results. The ECE was taken on November 2014 by 2nd grade students of primary schools and 4th grade for bilingual schools. The 2014 ECE results were publicly available in February 2015 at the regional level and not at the school level. Even though ECE results at the school level were not public in 201531, it represents a proxy for student performance, which could have been shared through teacher networks. In Column 1 we present the results for the sample of 6,502 candidates whose ranking includes schools with ECE results, and in Column 2 we narrowed the sample to the 3,938 candidates that have previous workplace information. In both specifications, the ECE Reading performance is not significant, most likely because the candidates did not have access to ECE results at the school level.32 30 Table in Appendix. 31 Since 2018, the public has had access to ECE school level results through the school identification website of the Ministry of Education http://identicole.minedu.gob.pe/. 32 As a robustness check, we include ECE Math scores instead of Reading scores, and the results remain insignificant. These results are available upon request. 17 5.2. Heterogeneous Effects by Candidates’ Attributes Following Boyd et al. (2005) and Krieg, Theobald and Goldhaber (2016), teachers’ preferences for distance from school can vary with their own attributes, such as their gender, age and academic performance. Literature in the United States shows that, for most women, the job search proceeds from a given residential location, that women travel shorter distances to work than men, and that they are more likely to work within the local community (Hanson and Pratt, 1988). One explanation could be that women are constrained to work close to home because of family responsibilities. In Latin America, traditional gender roles persist, in which women are expected to take most of the household and family responsibilities (OIT, 2019; Ñopo, 2012). Other studies suggest that women’s stronger preferences for short distance and commuting time are mainly explained by women’s lower income and their greater reliance on public transportation (Hanson and Johnston, 1985). Studies analyzing teacher job markets suggest that female teachers are more likely to work closer to their TEP (Boyd et al., 2005) and to their student teaching location (Krieg, Theobald and Goldhaber, 2016). Regarding age, the literature shows that, as adults grow older, they become less willing to take risks (Schildberg-Hörisch, 2018). Schurer (2015) documents that risk tolerance declines strongly for all socioeconomic groups from late adolescence up to age 45. From age 45 onwards, risk tolerance continues to decline for the most disadvantaged and stabilizes for all other groups. In addition, younger candidates might have fewer household responsibilities, and therefore, more flexibility to choose their work location.33 Studies in United States suggest that individuals who begin their teaching career when they are younger are more likely to take jobs farther away from their TEP and their student teaching location, but closer to their hometown (Boyd et al., 2005; Krieg, Theobald and Goldhaber, 2016). The literature shows mixed results on the impact of teacher academic performance and knowledge (as measured by standardized tests) on distance preferences. On the one hand, Boyd et al. (2005) find that more qualified teachers (measured by SAT scores) are slightly more willing to expand their job search away from their hometown. On the other hand, Krieg, Theobald and Goldhaber (2016) found some evidence that more qualified teachers (measured by college GPA scores) work in schools closer to their student teaching location. To analyze the heterogenous effects by teacher candidates attributes in Peru, we create three dummy variables for: (i) female candidates; (ii) candidates who are less than 35 years old, which represents the 50th percentile of the age variable in our sample; and (iii) candidates who have a national teacher test (PUN) score 33 While we do not have candidates’ marital status information, we assume younger candidates to be more likely to be single, thus might be more flexible to move further away from their residential location while looking for a job. 18 in the highest quintile. We interact these dummy variables with the distance variables and urban location of schools. In our sample, a higher percentage of females remain in the same region where they studied and previously worked: 69% of females applied to the same region where they studied (vs. 65% for males), and 94% applied to the same region where they previously worked (vs. 90% for males). In sum, females appear to have stronger preferences for proximity from their TEP and their previous workplace than males. Regarding candidates’ age, 70% of candidates less than 35 years old applied to the same region where they studied (vs. 65% for older candidates), and 93% applied to the same region where they previously worked (similar for older candidates). Similarly, a higher percentage of high PUN scores candidates remain in the same region where they studied (73% vs. 67% for low PUN scores candidates). Columns 3-4 in Table 7 are analogous to Columns 1-2 with the addition of interactions between school characteristics and candidates’ attributes. In Column 3 we include the interactions of distance from TEP and urban with candidate’s attributes. Females and candidates with high PUN scores have stronger preferences for proximity to their TEP and urban areas, while younger candidates have weaker preferences for more urban areas. Figure 4 compares school A and B, and shows that, as the school B is further away from TEP relative to school A, the probability that school A is preferred to (ranked in a higher position than) school B increases more rapidly for females and high PUN candidates, and more slowly for younger candidates. Regarding urban location, Figure 5 show the probability that an urban school is preferred to a most rural (rural 1) school for candidates with different attributes, holding other school characteristics constant. Female, older teachers and candidates with high PUN scores have stronger preferences for urban schools. For instance, the probability of preferring an urban school to a most rural (rural 1) school is 56% for candidates with high PUN score, higher than the 53% probability for candidates with lower PUN scores. The last specification (column 4 of Table 7) adds the interactions between distance from previous workplace and candidate’s attributes, for the subsample of candidates with previous workplace information. The results show that candidates with high PUN scores have stronger preferences for schools that are closer to their previous workplace; while females and younger teachers seem to have similar preferences as males and older teachers, respectively. Figure 6 shows the probability of preferring school A over school B, as the distance between the previous workplace and school B change values. As school B is further away from the previous workplace, the probability of preferring school A over school B is higher for high PUN score candidates. Females show stronger preferences for proximity from TEP and for urban schools. As we control for distance from previous workplace (column 4 of Table 7), the effect of distance from 19 TEP and urban location vanishes for the base category (for males that are more than 35 years old and have low PUN scores). However, for females, the interactions with distance from TEP and urban continue to be statistically significant. For younger candidates, only the interaction with urban schools is significant; and for high PUN candidates, the interactions with urban schools and with distance from previous workplace are significant. These results are consistent with the literature and suggest that females value proximity to TEP more than other candidates.34 5.3. Robustness Checks In this section we present a series of robustness checks: (i) substitute the urban/rural variables with the distance from the province capital; (ii) include an analysis by educational level; and (iii) estimate a conditional logit model on the first ranked school for each candidate. 5.3.1. Distance from province capital Table 8 presents the rank-ordered logit results including the school distance from the closest province capital, instead of the urban/rural variables. We do not include distance from the closest province capital and the urban/rural variables in the same regression because the definition of rurality in Peru is based, in part, on travel time to the closest province capital (as explained in section 1) and therefore, these variables are highly correlated. The results in Table 8 suggest that our preferences estimates are not particularly sensitive to this alternative specification. Considering the sample of all teacher candidates and the specifications without interactions (Column 1 of Table 8), we find that candidates prefer schools that are closer to the province capital, which is consistent with previous research on teacher preferences in developing countries (Rosa, 2017) and preferences for urban schools observed in the baseline results (Table 7). Candidates are willing to work in schools further away from the province capital conditional on the school having basic services or being in wealthier areas. Comparing the coefficients on basic services and distance from capital, the point estimates suggest that candidates are willing to work 15 km away from the capital to teach in a school with basic services. A comparison with the coefficient on poverty implies that candidates are more willing to work 21 km away from the capital to teach in a school in the 25th percentile of poverty than in one in the 50th percentile. When controlling for previous workplace (Column 2 of Table 8), we find that distance from previous workplace plays a more important role in teacher preferences than distance from province capital. 34 We also analyze the interactions between the candidate’s attributes and other school characteristics (enrollment and basic services). These interactions were not significant, and their inclusion did not affect the results of the interactions with distances and urban location. 20 In the specifications with interactions (Column 3 and 4 of Table 8), we find that females, older teachers and high PUN score candidates have stronger preferences for proximity to the province capital. Figure 7 compares schools A and B, and illustrates that, as school B is further away from the province capital relative to school A, the probability that school A is preferred to (ranked in a higher position than) school B increases more rapidly for females and high PUN candidates, and more slowly for younger candidates. Regarding the interactions with distance from TEP, only the interaction with female remains significant. Moreover, the distance from previous workplace continues to be particularly relevant for high PUN score candidates (Column 4 of Table 8). 5.3.2. Analysis by educational level As documented in Table 1, the shortage of qualified teacher candidates varies by educational level. For instance, in pre-primary, the offered vacancies surpass the number of candidates that pass the PUN and participate in the decentralized stage. In this context, understanding candidates’ preferences by educational level could help policymakers tackle the staffing challenges in the educational levels with highest teacher shortages. Table 9 presents the rankordered logit results separately by educational level, for pre-primary, primary and secondary teacher candidates. In the specification with the full sample of candidates and without interactions (Columns 1 in Table 9), we find significant parameter estimates for distance from TEP, poverty, enrollment and distance from previous workplace in all levels, in line with the pooled results (Table 7). Regarding candidates’ attributes, the main difference across education levels is gender: the percentage of female candidates is 99%, 77% and 56% in pre-primary, primary and secondary, respectively. In each level, between 50%-60% of candidates are younger than 35 years old, and between 16%-21% have high PUN scores. When analyzing the interaction terms (in Columns 3-4 in Table 9), we find some differences in candidates’ preferences by educational level. Female candidates in pre-primary and primary do not have significant preferences for proximity to TEP, while the opposite is true for secondary female candidates. Regarding age, younger pre-primary and secondary candidates have weaker preferences for urban areas, while the result does not hold for primary. High PUN score candidates in pre-primary and primary do no show significant preferences for proximity to TEP and urban areas, and have weaker preferences to proximity to previous workplace, in contrast to secondary candidates (Columns 3-4 in Table 9). This suggest that high performing candidates in pre-primary and primary might be willing to move further away from their prev ious workplace, which is a proxy for residential location. 21 Overall, we do not find meaningful differences in the preferences structure among candidates by educational levels. In order to explain the teacher shortage in pre-primary, a deeper analysis of the supply of qualified teachers in this level would be needed. 5.3.3. Conditional logit Table 10 presents a simple conditional logit specification to examine school characteristics that are associated with candidates’ first school choices. This specification estimates the probability of choosing a school as a first choice given the set of characteristics of the other schools included in the candidates’ preference set. The conditional logit estimates are consistent with the baseline regressions.35 The characteristics that explain candidates’ first choices are similar to the ones that explain candidates’ school ranking. In the first two specifications (Columns 1 and 2 in Table 10), we find significant parameter estimates for distance from TEP, urban, distance from previous workplace, poverty and basic services. The rural categories (Least Rural and Moderate Rural) are not significant under the conditional logit model, given the strong preferences for urban schools as the first choice. Around 62% of candidates rank urban schools first in their preference set. The specifications with interactions (Columns 3 and 4 of Table 10) yielded similar preference results. Females prefer schools that are closer to their TEP, while high PUN score candidates prefer schools closer to their previous workplace. Both females and high PUN score candidates have stronger preferences for urban schools. 6. Conclusions and Policy Implications There is a significant shortage of permanent teachers in Peru, especially in disadvantaged areas. In the 2015 teachers’ national contest, almost 40% of vacancies for permanent teaching positions received no applications. More than 50% of the unselected vacancies were located in the poorest areas of the country and 95% were located in rural areas. These undesired vacancies generally ended up being filled by temporary teachers which, research suggests, can have a negative influence on student learning (Ayala & Sánchez, 2016), especially for disadvantaged students (Marotta, 2019). In this paper we explore the school preferences of 23,046 permanent teacher candidates in order to identify which public school characteristics drive candidates’ school preferences. A central motivation to analyze teacher preferences is to determine ways to attract teachers to hard- to-staff schools. Understanding which school characteristics teachers value the most can help 35 In the baseline regressions (Table 7) we estimate a rank-ordered logit model using the rankings of candidates’ top five school choices. 28 Marotta, L. (2019). Teachers’ Contractual Ties and Student Achievement: The Effect of Temporary and Multiple-School Teachers in Brazil. Comparative Education Review 63(3), 356-376. Matthews, P., Klaver, E., Lannert, J., Conluain, G., & Ventura, A. (2009), Policy Measures Implemented in the First Cycle of Compulsory Education in Portugal. International Evaluation for the Ministry of Education, Office for Education Statistics and Planning, Ministry of Education, Portugal. Ñopo, H. (2012). New Century, Old Disparities. The World Bank. https://doi.org/10.1596/978-0- 8213-8686-6 OECD. (2017). Organization of the school offer. OECD Review of Policies to Improve the Effectiveness of Resource Use in Schools OIT. (2019). Panorama Laboral Temático 5: Mujeres en el mundo del trabajo. Retos pendientes hacia una efectiva equidad en América Latina y el Caribe. Lima: OIT / Oficina Regional para América Latina y el Caribe. Punj, G., & Staelin, R. (1978). The Choice Process for Graduate Business Schools. Journal of Marketing Research, XV, 588-98. Reininger, M. (2012). Hometown disadvantage? It depends on where you’re from: teachers’ location preferences and the implications for staffing schools. Educational Evaluation and Policy Analysis, 34(2), 127-145. Rivkin, S. G., Hanushek, E. A., & Kain, J. F. (2005). Teachers, schools and academic achievement. Econometrica, 73(2), 417–458. Rockoff, J.E. (2004). The impact of individual teachers on student achievement: Evidence from panel data. American Economic Review, 94 (2), 247–252. Ronfeldt, M. (2012). Where Should Student Teachers Learn to Teach? Effects of Field Placement School Characteristics on Teacher Retention and Effectiveness. Educational Evaluation and Policy Analysis, 34(1), 3-26. Ronfeldt, M., Kwok, A., & Reininger, M. (2014). Teachers’ Preferences to Teach Undeserved Students. Urban Education, 51(9), 995-1030. Rosa, L. (2017). Teacher Preferences in Developing Countries Evidences from the City of Sao Paulo, Brazil. Stanford University. Draft. Rots, I., Aelterman, A., Vlerick, P., & Vermeulen, K. (2007). Teacher education, graduates’ teaching commitment and entrance into the teaching profession. Teaching and Teacher Education, 23, 545-556. 29 Ryan, R.M. & Deci, E. (2000). Intrinsic and extrinsic motivations: Classic definitions and new directions. Contemporary Educational Psychology 25, 54-67. Schildberg-Hörisch, H. (2018). Are Risk Preferences Stable? Journal of Economic Perspectives, 32(2), 135-154. Schurer, S. (2015). Lifecycle patterns in the socioeconomic gradient of risk preferences. Journal of Economic Behavior & Organization, 119(C), 482-495. 30 Table 1: Summary statistics of 2015 teacher hiring process in Peru Offered vacancies Selected vacancies Candidates in national stage Candidates in decentralized stage Candidates that won a vacancy (1) (2) (3) (4) (5) (2)/(1) (3)/(1) (4)/(1) (5)/(1) Pre-primary 8,896 4,356 28,775 5,654 2,432 49% 3.2 0.6 27% Primary 6,460 4,496 77,594 6,597 2,949 70% 12.0 1.0 46% Primary - Physical Ed. 55 52 2624 144 38 95% 47.7 2.6 69% Secondary (total) 4,219 3,754 83,404 11,306 2,718 89% 19.8 2.7 64% Secondary - Arts 428 378 4,807 404 207 88% 11.2 0.9 48% Secondary - Sciences 286 284 9,292 817 225 99% 32.5 2.9 79% Secondary - Communication 564 563 17,317 3277 530 100% 30.7 5.8 94% Secondary - Physical Ed. 229 228 7,846 769 205 100% 34.3 3.4 90% Secondary - Religion 703 420 2,433 212 138 60% 3.5 0.3 20% Secondary - Vocational Ed. 641 544 11,649 1244 355 85% 18.2 1.9 55% Secondary - Civic Ed. 137 133 1491 387 116 97% 10.9 2.8 85% Secondary - History, Geography, Econ. 172 172 9629 913 149 100% 56.0 5.3 87% Secondary - English 494 469 5,178 1,040 316 95% 10.5 2.1 64% Secondary - Math 453 453 11,826 1,838 387 100% 26.1 4.1 85% Secondary - Humanities 112 110 1936 405 90 98% 17.3 3.6 80% Total 19,630 12,658 192,397 23,701 8,137 64% 9.8 1.2 41% 31 Table 2: Vacancies’ characteristics according to whether they were selected by at least one candidate All vacancies Unselected vacancies Selected vacancies t-test N. Most Rural (Rural 1) 36% 59% 23% *** 16,743 Moderate Rural (Rural 2) 28% 30% 28% *** 16,743 Least Rural (Rural 3) 13% 7% 17% *** 16,743 Urban 22% 5% 32% *** 16,743 Poverty (%) 47% 56% 42% *** 16,588 Enrollment (100s) 1.63 0.56 2.28 *** 16,740 Basic services 53% 32% 66% *** 16,576 Distance from prov. capital (km) 29.45 42.55 21.58 *** 16,743 Student test scores in Reading (standardized) -0.06 -0.64 0.17 *** 3,518 Student test scores in Math (standardized) -0.07 -0.75 0.10 *** 2,871 N. 16,743 6,283 10,460 32 Table 3: Candidates’ characteristics according to their performance in the National Teacher Test All Candidates that do not pass the test Candidates that pass the test t-test Age 37.1 37.5 34.5 *** Female 66% 65% 72% *** Teaching experience in public schools (years) 4.51 4.54 4.24 *** Teaching experience in private schools (years) 2.17 2.01 3.31 *** Studied in an Institute 62% 64% 47% *** Studied in a Public Institute 43% 44% 35% *** Studied in a Private Institute 20% 22% 12% *** Studied in a University 38% 36% 53% *** Studied in a Public University 27% 25% 43% *** Studied in a Private University 11% 11% 10% *** Studied in a rural Institute or University 5% 5% 3% *** Studied in a University ranked in the top 15 9% 8% 16% *** Teacher test score 99.1 92.3 147.5 *** N. 192,397 168,696 23,701 33 Table 4: Summary statistics for the Rank-ordered logit analysis Note: Teacher test score Q5 is a dummy variable for candidates who have a national teacher test (PUN) score in the highest quintile. Basic services include electricity, water and sanitation. N. Mean Std. Dev. Min Max School Ranking 105,061 3.09 1.41 1 5 Candidate-level characteristics Female 23,046 72% 45% 0 1 Age < 35 23,046 55% 50% 0 1 Teacher test score Q5 23,046 19% 39% 0 1 Temporary teacher in 2015 23,046 61% 49% 0 1 Distance from Teacher Education Program (km) 23,046 149 252 0.1 1,844 Distance from previous workplace (km) 13,901 51 89 0 1,510 Vacancy-level characteristics Most Rural (Rural 1) 10,174 23% 42% 0 1 Moderate Rural (Rural 2) 10,174 28% 45% 0 1 Least Rural (Rural 3) 10,174 17% 38% 0 1 Urban 10,174 32% 47% 0 1 Poverty (%) 10,174 42% 22% 0.0 97.4 Enrollment (100s) 10,174 2.28 3.51 0.0 28.7 Basic services 10,174 67% 47% 0 1 Distance from prov. capital (km) 10,174 21.6 17.7 0.0 158.0 Student test scores in Reading (standardized) 2,465 -0.4 1.2 -5.1 3.9 Student test scores in Math (standardized) 2,275 -0.3 1.2 -4.2 3.7 34 Table 5: Vacancies’ characteristics by candidates’ school ranking Note: This table shows vacancies’ average characteristics according to candidates’ ranking. In the school ranking, column 1 indicates teacher candidate’s most preferred option while column 5 indicates the least preferred one. The rural and urban categories sum 100%. 12345 Most Rural (Rural 1) 8% 8% 9% 9% 10% *** 105,061 Moderate Rural (Rural 2) 14% 16% 16% 17% 18% *** 105,061 Least Rural (Rural 3) 16% 17% 17% 17% 16% 105,061 Urban 62% 59% 58% 57% 56% *** 105,061 100% 100% 100% 100% 100% Poverty (%) 30% 31% 31% 31% 32% *** 105,061 Enrollment (100s) 4.06 3.72 3.71 3.57 3.47 *** 105,061 Basic services 85% 83% 83% 82% 81% *** 105,061 Distance from prov. capital (km) 15.23 16.28 16.46 17.02 17.80 *** 105,061 Student test scores in Reading (standardized) 0.03 0.00 -0.01 -0.03 -0.05 *** 26,505 Student test scores in Math (standardized) 0.03 0.01 -0.01 -0.03 -0.06 *** 25,770 School ranking t-test 1 vs 5 N. 35 Table 6: Choice set composition according to candidate’s characteristics Note: This table shows the distribution of vacancies’ characteristics among candidates’ preference sets. The categories within a variable sum 100%. For continuous variables we present the distribution of candidates’ preferences over the variables’ quintiles (Q1-Q5). ***p<0.01, **p<0.05, *p<0.1 Candidates' characteristics All Male Female t-test Age > 35 Age < 35 t-test Teacher test score Q5 Teacher test score Q1 t-test Province Just 1 province 59% 46% 63% *** 61% 57% *** 64% 55% *** Multiple provinces 41% 54% 37% *** 39% 43% *** 36% 45% *** Degree of rurality Just urban 38% 33% 40% *** 44% 33% *** 49% 31% *** Just rural 22% 21% 22% ** 18% 24% *** 14% 30% *** Just Most Rural (Rural 1) 2% 3% 1% *** 1% 2% *** 1% 4% *** Just Moderate Rural (Rural 2) 1% 1% 1% *** 1% 1% *** 1% 2% *** Just Least Rural (Rural 3) 1% 1% 1% *** 1% 1% *** 1% 1% Mix rural 18% 17% 18% *** 15% 19% *** 11% 23% *** Mix urban/rural 40% 46% 38% *** 37% 43% *** 38% 39% *** Distance from Teacher Education Program (100km) Just Q1 7% 4% 9% *** 8% 7% *** 11% 4% *** Just Q2 6% 4% 6% *** 6% 6% 7% 5% *** Just Q3 6% 5% 6% *** 6% 6% 6% 6% Just Q4 9% 12% 8% *** 9% 9% * 7% 10% *** Just Q5 18% 18% 18% ** 20% 16% *** 15% 21% *** Mix Q1-Q5 54% 57% 53% *** 51% 56% *** 55% 53% ** Poverty (%) Just Q1 40% 30% 44% *** 46% 35% *** 50% 34% *** Just Q2 3% 3% 3% *** 3% 3% 3% 4% *** Just Q3 2% 2% 1% *** 1% 2% *** 1% 2% *** Just Q4 1% 1% 1% * 1% 1% *** 1% 1% *** Just Q5 (poorest) 2% 3% 1% *** 1% 2% *** 1% 2% *** Mix Q1-Q5 52% 61% 49% *** 47% 57% *** 45% 56% *** 36 Table 7: Rank-ordered logit results Note: Columns 2 and 4 consider the subsample of candidates with previous workplace information. Regarding urban/rural variables, the omitted category is Most Rural (Rural 1). ***p<0.01, **p<0.05, *p<0.1 All Candidates w/previous workplace information All Candidates w/previous workplace information (1) (2) (3) (4) Distance from Teacher Education Program (km) -0.0055*** -0.0009*** -0.0051*** 0.0000 (0.0002) (0.0003) (0.0003) (0.0005) Urban 0.1689*** 0.0991*** 0.1440*** 0.0648 (0.0258) (0.0325) (0.0371) (0.0472) Least Rural (Rural 3) 0.0895*** 0.0375 0.0913*** 0.0384 (0.0237) (0.0294) (0.0237) (0.0294) Moderate Rural (Rural 2) 0.0096 -0.0004 0.011 -0.0006 (0.0215) (0.0264) (0.0215) (0.0264) Poverty (%) -0.4800*** -0.3614*** -0.4759*** -0.3536*** (0.0446) (0.0578) (0.0447) (0.0580) Basic services 0.0536*** 0.0486*** 0.0518*** 0.0472*** (0.0137) (0.0171) (0.0137) (0.0171) Enrollment (100s) 0.0101*** 0.0078*** 0.0104*** 0.0082*** (0.0016) (0.0022) (0.0016) (0.0022) Distance from previous workplace (km) -0.0099*** -0.0099*** (0.0003) (0.0005) Distance from Teacher Education Program (km) *female -0.0009*** -0.0012** (0.0003) (0.0005) *age < 35 0.0005 -0.0003 (0.0003) (0.0005) *teacher test score Q5 -0.0011** 0.0002 (0.0005) (0.0008) Urban *female 0.0878*** 0.0877** (0.0297) (0.0383) *age < 35 -0.0943*** -0.0883** (0.0279) (0.0363) *teacher test score Q5 0.1039*** 0.1321*** (0.0357) (0.0452) Distance from previous workplace (km) *female -0.0001 (0.0005) *age < 35 0.0006 (0.0005) *teacher test score Q5 -0.0024*** (0.0008) N. 105,061 63,398 105,061 63,398 Candidates 23,046 13,901 23,046 13,901 Without interactions With interactions Dependent variable: School ranking 37 Table 8: Rank-ordered logit results - Distance from the province capital Note: Columns 2 and 4 consider the subsample of candidates with previous workplace information. ***p<0.01, **p<0.05, *p<0.1 All Candidates w/previous workplace information All Candidates w/previous workplace information (1) (2) (3) (4) Distance from Teacher Education Program (km) -0.0052*** -0.0009*** -0.0048*** 0.0000 (0.0002) (0.0003) (0.0003) (0.0005) Distance from prov. capital (km) -0.0038*** -0.0004 -0.0037*** -0.0000 (0.0004) (0.0005) (0.0008) (0.0011) Poverty (%) -0.4884*** -0.4134*** -0.4801*** -0.4037*** (0.0442) (0.0572) (0.0442) (0.0573) Basic services 0.0678*** 0.0624*** 0.0664*** 0.0616*** (0.0134) (0.0167) (0.0134) (0.0167) Enrollment (100s) 0.0124*** 0.0107*** 0.0126*** 0.0109*** (0.0015) (0.0021) (0.0015) (0.0021) Distance from previous workplace (km) -0.0099*** -0.0099*** (0.0003) (0.0005) Distance from Teacher Education Program (km) *female -0.0007* -0.0011** (0.0004) (0.0005) *age < 35 0.0001 -0.0005 (0.0004) (0.0005) *teacher test score Q5 -0.0008 0.0004 (0.0005) (0.0008) Distance from prov. capital (km) *female -0.0025*** -0.0022** (0.0008) (0.0011) *age < 35 0.0035*** 0.0025** (0.0008) (0.0011) *teacher test score Q5 -0.0027** -0.0025* (0.0011) (0.0014) Distance from previous workplace (km) *female -0.0000 (0.0005) *age < 35 0.0005 (0.0005) *teacher test score Q5 -0.0022*** (0.0008) N. 105,061 63,398 105,061 63,398 Candidates 23,046 13,901 23,046 13,901 Without interactions With interactions Dependent variable: School ranking 44 Figure 5: Probability of preferring an Urban school to a Most Rural (Rural 1) school Note: The probabilities were calculated using the estimated coefficients in Model 3 of Table 7. National Teacher Test (Prueba Única Nacional - PUN). 0.53 0.55 0.53 0.51 0.53 0.56 0.48 0.49 0.50 0.51 0.52 0.53 0.54 0.55 0.56 male female 35 or more Less than 35 Score Q1-4 Score Q5 Gender Age PUN score Probability of prefering an Urban School to a Rural 1 school 45 Figure 6: Probability of preferring a school closer from previous workplace National Teacher Test (PUN) score Note: School A and school B are two hypothetical schools which share all characteristics except the distance from the previous workplace. These graphs show the probability of preferring school A to school B, as school B is farther away from the previous workplace. The probabilities were calculated using the estimated coefficients in Model 4 of Table 7. The X- axis shows the differences until it reaches the 99th percentile for the distance from previous workplace. 46 Figure 7: Probability of preferring a school closer from province capital 7.1 Gender 7.2. Age 7.3 National Teacher Test (PUN) score Note: School A and school B are two hypothetical schools which share all characteristics except the distance from the province capital. These graphs show the probability of preferring school A to school B, as school B is farther away from the province capital. The probabilities were calculated using the estimated coefficients in Model 3 of Table 8. The X-axis shows the differences until it reaches the 99th percentile for the distance from province capital. 47 Appendix Table A1: Candidates’ characteristics according to the selection round they participated in All Candidates Candidates in first round Candidates only in second round t-test Age 34.5 34.5 35.3 ** Female 72% 72% 69% Public experience (years) 4.24 4.24 4.11 Private experience (years) 3.31 3.30 3.85 ** Studien in an Institute 47% 47% 41% ** Studied in a Public Institute 35% 35% 32% Studied in a Private Institute 12% 12% 9% ** Studied in a University 53% 53% 59% ** Studied in a Public University 43% 43% 46% Studied in a Private University 10% 10% 13% * Studied in a rural Institute or University 3% 3% 3% Studied in a University ranked in the top 15 16% 16% 17% Teacher test score 147.5 147.6 145.0 *** N. 23,701 23,319 382 48 Table A2: Rank-order logit results in the subsample with student test results Note: The subsample with student test (ECE) results considers candidates for primary schools that have selected schools with 2014 ECE results. Columns 2 and 4 restricts the subsample to candidates with previous workplace information. Regarding urban/rural variables, the omitted category is Most Rural (Rural 1). As a robustness check, we include ECE Math scores instead of Reading scores, and ECE results remain insignificant. ***p<0.01, **p<0.05, *p<0.1. All Candidates w/previous workplace information All Candidates w/previous workplace information (1) (2) (3) (4) Distance from Teacher Education Program (km) -0.0038*** -0.0005 -0.0038*** -0.0009 (0.0004) (0.0006) (0.0009) (0.0016) Urban 0.1107** 0.0472 0.1179 0.0472 (0.0530) (0.0678) (0.0874) (0.1148) Least Rural (Rural 3) 0.0694 -0.0028 0.0715 -0.0034 (0.0491) (0.0620) (0.0492) (0.0623) Moderate Rural (Rural 2) 0.0516 0.0314 0.0526 0.0284 (0.0427) (0.0524) (0.0427) (0.0524) Poverty (%) -0.5127*** -0.3080** -0.5120*** -0.3058** (0.0965) (0.1301) (0.0967) (0.1305) Basic services 0.0909*** 0.0812** 0.0908*** 0.0795** (0.0297) (0.0370) (0.0297) (0.0370) Enrollment (100s) 0.0135*** 0.0147*** 0.0135*** 0.0151*** (0.0039) (0.0053) (0.0039) (0.0053) Student test scores in Reading (standardized) -0.0035 0.0056 -0.0034 0.0052 (0.0113) (0.0144) (0.0113) (0.0144) Distance from previous workplace (km) -0.0084*** -0.0110*** (0.0006) (0.0015) Distance from Teacher Education Program (km) *female -0.0003 -0.0005 (0.0009) (0.0014) *age < 35 0.0002 0.0002 (0.0008) (0.0013) *teacher test score Q5 0.0006 0.0047** (0.0011) (0.0018) Urban *female 0.0632 0.0767 (0.0719) (0.0941) *age < 35 -0.0943 -0.1212 (0.0581) (0.0774) *teacher test score Q5 0.0168 0.0772 (0.0746) (0.0955) Distance from previous workplace (km) *female 0.0018 (0.0014) *age < 35 0.0020 (0.0012) *teacher test score Q5 -0.0017 (0.0018) N. 26,505 15,791 26,505 15,791 Candidates 6,502 3,937 6,502 3,937 Dependent variable: School ranking Without interactions With interactions