Navigating centralized admissions: The role of parental preferences in school segregation in Chile

Elacqua, Gregory; Kutscher, Macarena Working Paper Navigating centralized admissions: The role of parental preferences in school segregation in Chile IDB Working Paper Series, No. IDB-WP-01564 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Elacqua, Gregory; Kutscher, Macarena (2023) : Navigating centralized admissions: The role of parental preferences in school segregation in Chile, IDB Working Paper Series, No. IDB-WP-01564, Inter-American Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0005484 This Version is available at: https://hdl.handle.net/10419/299456 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/ Navigating Centralized Admissions: The Role of Parental Preferences in School Segregation in Chile Gregory Elacqua Macarena Kutscher WORKING PAPER No IDB-WP-01564 Inter-American Development Bank Division of Education December 2023 Navigating Centralized Admissions: The Role of Parental Preferences in School Segregation in Chile Gregory Elacqua Macarena Kutscher Inter-American Development Bank Division of Education December 2023 Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library Elacqua, Gregory M., 1972 - Navigating centralized admissions: the role of parental preferences in school segregation in Chile / Gregory Elacqua, Macarena Kutscher. p. cm. — (IDB Working Paper Series ; 1564) 1. School choice -Decision making-Chile. 2. Private schools-Chile. 3. Public schools -Chile. I. Kutscher, Macarena. II. Inter-American Development Bank. Education Division. III. Title. IV. Series. IDB -WP-1564 Key words: School Choice, Centralized Assignment Systems, Segregation JEL Codes: A20, D12 I24 http://www.iadb.org Copyright © 2023 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. Navigating Centralized Admissions: The Role of Parental Preferences in School Segregation in Chile∗ Gregory Elacqua† Macarena Kutscher† Abstract In this paper, we aim to understand some of the mechanisms behind the low impact of a Chilean educational reform on socioeconomic integration within the school system. We focus on pre-kindergarden (pre-K) admissions, which account for the highest volume of applications since all students (except those applying to private schools) must seek admission through the centralized system. We employ a discrete choice model to analyze parents’ school preferences. Our analysis reveals that the school choices of low-SES families are more strongly influenced by a school’s non-academic attributes – which are often omitted from analyses of parental preferences due to data availability constraints – rather than academic quality. For instance, low-SES parents tend to prefer schools with fewer reported violent incidents, schools where students report facing less discrimination and exclusion, and schools where students demonstrate higher levels of self-efficacy. Disadvantaged families also tend to favor schools that have a religious affiliation, offer more ”classical” sports (e.g. soccer), or have a foreign name. These results have significant implications for understanding the preferences of disadvantaged families and the impact of centralized admission systems on reducing segregation. By recognizing the non-academic factors driving school choices, policymakers can better design admission systems that truly foster school diversity and equality. JEL Classification: A20, D12, I24 Keywords: School Choice, Centralized Assignment Systems, Segregation. ∗We thank Mauricio Aburto for his excellent assistance with the data analysis. We gratefully acknowledge the Inter-American Development Bank for funding this research. 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. The authors have no conflicts of interest or financial and material interests in the results. All errors are our own. †Inter-American Development Bank 1 1 Introduction Effective and inclusive education systems play an important role in promoting equal opportunities for all children. As part of a global shift in policy towards promoting diversity and equal access in education, various governments have introduced centralized school choice systems in an effort to make the admission process more transparent, efficient, and equitable (Elacqua et al.,2021). In these systems, families apply to schools through an online platform, ranking schools in order of preference. A mathematical algorithm, typically designed to be ”strategyproof,” then uses this information to assign students to available slots, ensuring compliance with government-established priority criteria. Centralized choice guarantees equal access to schools and has the potential to promote diversity within the education system. At the end of the enrollment period, each student is assigned to their top preference if there is an available seat. This eliminates supply-side selection, as schools are unable to screen students. Since the application process is centralized through a web platform, the system also decreases search costs for disadvantaged families and the time costs of applying to multiple schools. Moreover, the allocation algorithm can give higher priority or reserve some seats in the school for minority students, weakening the link between place of residence and school allocation by eliminating distance-based restrictions.1 Yet, existing evidence shows little or no significant impact of centralized choice reform on school segregation (Lauen,2007;Denice and Gross,2016;Kutscher et al.,2023;Honey and Carrasco,2022). This appears to be related to the fact that the actual implementation of these systems often deviates from the theoretical assumptions underpinning them. Factors such as information frictions and behavioral biases can undermine the effectiveness of centralized assignment mechanisms in promoting integration. Evidence suggests that higher-income individuals are more adept at navigating centralized systems and making well-informed choices (Luflade, 2017;Ajayi et al.,2020).2Additionally, lowand high-SES families may have different preferences in terms of school attributes. Indeed, studies from centralized market designs consistently show that disadvantaged families place more weight on proximity than school quality (Hastings et al.,2009, among others). Residential segregation may also act as a barrier to equitable access to schools as there can be considerable variation in the local supply of schools across socioeconomic groups. Therefore, even a policy aimed at leveling the playing field for all families may not necessarily lead to increased socioeconomic integration across schools. In 2016, Chile initiated a national reform known as the School Admission System, or Sistema de Admisi´on Escolar (SAE), which replaced a decentralized school choice scheme with a 1Examples of centralized admission systems in the region that establish priorities based on students’ socioeconomic or disadvantaged status include Chile, Recife (Brazil), and Palmira (Colombia). In Chile, 15% of each school’s vacancies are prioritized for students belonging to the lowest 40% socioeconomic level. In Recife, families that are recipients of the federal cash transfer program (Bolsa Familia) are prioritized. In Palmira, priority is given to families that were victims of the internal conflict, many of whom have been displaced. 2Families frequently rely on friends, family, and past experiences when judging a school’s quality (Elacqua et al.,2006). 2 student-school matching process employing a Deferred Acceptance (DA) algorithm. Prior to this reform, schools were allowed to implement their own admission criteria and procedures. The application process required direct applications to schools and frequently involved screening or cream-skimming practices, including interviews, entrance exams, income verification, and sometimes even additional documentation such as religious marriage certificates. These practices were criticized for their perceived role in fostering high levels of school segregation (Santos and Elacqua,2016;Valenzuela et al.,2014). Though the reform sought to improve equity and reduce segregation in the education system, initial studies have found little impact on access for low-income students (Kutscher et al.,2023;Honey and Carrasco,2022). In this paper, we aim to understand some of the mechanisms behind the low impact of the Chilean reform on socioeconomic integration in the education system. We focus on prekindergarden (pre-K) admissions, since all students entering this level–except those applying to private non-voucher schools–must seek admission through the centralized system. First, we review students’ enrollment and application patterns. Then, we explore differences in the patterns of parental school preferences between disadvantaged and non-disadvantaged families. We focus on application data covering all pre-K students in 2019, when the reform was already fully implemented nationwide. Importantly, our investigation extends beyond academic considerations and includes a comprehensive range of non-academic attributes of the schools, including sports infrastructure, religious affiliations, extracurricular activities, and others. The analysis reveals contrasts in the application patterns of disadvantaged and non-disadvantaged families. Consistent with other studies, non-disadvantaged students are more willing than their disadvantaged peers to travel longer distances to schools and apply to high-performing schools. Low-income families, on average, apply to fewer schools and are less likely to apply to schools in high demand (measured as the ratio of applications to vacancies), schools with higher-SES students, and previously selective schools. When we examine the impact of the SAE policy on school segregation, our results indicate that the reform has had no discernible effect on the socioeconomic composition of pre-K students across schools. Additionally, we do not find encouraging trends in the representation of disadvantaged families in schools considered more ”desirable.” These results suggest that the removal of school admission barriers alone may only have a limited impact on the actual distribution of students. We accordingly explore whether disadvantaged and non-disadvantaged families differ in how they prioritize school attributes. We analyze parental school preferences by fitting a discrete choice model to their rank-ordered preference lists. In line with previous work, we find that parents assign higher rankings to closer schools, higher-performing schools, and those with a higher socioeconomic composition (Hastings et al., 2007;Burgess et al.,2015;Abdulkadiro˘glu et al.,2020;Beuermann et al.,2023). However, we find substantial differences depending on family socioeconomic background. Low-SES families have lower odds of selecting a more distant school, one that charges fees, or one that had a selec3 tive admission process before the reform. In addition, they are less likely than non-disadvantaged parents to list high-quality schools or schools with a higher SES level. Furthermore, we observe that the choices made by low-SES families are more strongly influenced by the school’s non-academic attributes, which are often omitted from parental preference analyses due to data availability constraints. Low-SES parents tend to favor schools that offer more ”classical” sports, have a foreign name, and have a religious affiliation. They are also more likely to rank schools with a relatively more favorable climate: those with fewer reported violent incidents, where students report facing less discrimination and exclusion, and where students demonstrate higher levels of self-efficacy. We assess the robustness of our results to various specifications. For instance, we acknowledge that there may be disparities in the attributes of accessible schools and this may, to some extent, conflate parental preferences with their local constraints. Disadvantaged and non-disadvantaged families may have different sets of school choices due to residential or spatial inequalities. We take this into consideration by standardizing the school attributes at the education market level. Thus, instead of comparing preferences for these attributes in absolute terms, we do so at a relative level. Our results remain consistent to this and other robustness checks. Our paper contributes to the literature in several ways. We add to the small body of studies examining the effects of centralized school admission systems on socioeconomic integration across schools. Two recent papers explore the Chilean reform by taking advantage of the discontinuities in the introduction of the policy. Kutscher et al. (2023) assess segregation in the first year of secondary school (ninth grade) and find that it increased following the reform in school districts with high levels of pre-existing residential segregation and in districts with a significant presence of private schools. Honey and Carrasco (2022) also study the Chilean reform and find little shortterm effect on the enrollment of low-income students in desirable schools (i.e., high-performing or previously selective schools). Our findings are also relevant to recent empirical research leveraging preference data from centralized school assignment mechanisms to investigate parental preferences (Beuermann et al.,2023;Abdulkadiro˘glu et al.,2020;Glazerman and Dotter,2017;Burgess et al.,2015). We make two important contributions to this literature. First, while most of these studies analyze revealed preferences in coordinated school admission systems, our investigation has the advantage of exploiting a national reform in a developing country. Second, much of this prior work has been limited in terms of the characteristics that could be measured and studied. The richness of our data allows us to include a comprehensive range of non-academic school attributes, including sports infrastructure, religious affiliation, extracurricular activities, and school safety and climate, among others. The findings herein have important implications for our understanding of parental school preferences, especially among vulnerable populations. They indicate that parental preferences, in addition to well-documented constraints related to information and residential segregation, may 4 hinder school integration. This suggests that addressing disparities in school access and ensuring equitable educational opportunities for all students requires not only addressing structural and informational barriers but also working to challenge and reshape entrenched preferences that perpetuate educational inequalities. Such efforts should take into account the diverse needs and aspirations of vulnerable populations and promote inclusion and diversity within the education system. The rest of the paper is organized as follows. The next section describes the education system in Chile and the new centralized admission system reform. Section 3 discusses the data, focusing in particular on the classification of disadvantaged and non-disadvantaged students. It also reports descriptive statistics on the reform and examines school enrollment patterns of preK students. Section 4 presents our empirical strategy for estimating parental preferences, and Section 5 sets forth the results of the rank order logit model and robustness checks. In Section 6, we offer some conclusions for policy related to parental choice and educational inequality in centralized student assignment systems. 2 Contextual Background The education system in Chile consists of eight years of primary education and four years of secondary education. Schools are divided into public schools, financed by government vouchers (subsidies); private-voucher schools, financed by vouchers and additional fees to parents; and private non-voucher schools, which do not receive government funding. As of 2022, public schools represent approximately 37 percent of total enrollment, while private voucher schools account for approximately 53 percent of students, and the private non-voucher sector enrolls 10 percent of students. Chile’s educational system is known for being highly segregated. In an effort to combat this, the Chilean government introduced a centralized school admission system in 2016, called Sistema de Admisi´on Escolar (SAE), as the central pillar of a major education reform aimed at promoting social inclusion and reducing the high levels of school segregation. The previous (decentralized) student admission process was highly unregulated. Most private non-voucher and private-voucher schools selected students based on elements such as interviews with parents, entrance exams, proof of income, and religious marriage certificates. In fact, a significant number of schools continued to employ selective admission procedures even after having been legally restricted in their ability to do so in 2011 (Carrasco,2014).3 The new school admission system was rolled out between 2016 and 2019, replacing the country’s widely studied decentralized school choice system (Epple et al.,2017;Hsieh and Urquiola, 3There is no evidence that banning selective admission impacted school screening processes. According to the Ministry of Education, in 2005, 21% of private-voucher schools conducted interviews with parents and 28% used assessments to screen students. In 2012, after the 2008 SEP law made selective admissions technically illegal, and schools could be fined for screening students, 31% of private-voucher schools were still conducting interviews and testing new students. 5 a metropolitan area. For the purposes of this study, we define school districts as municipalities. This definition is based on the observation that, in our sample, approximately 90% of prekindergarten students attend a school within their municipality of residence.10 We employ the Duncan and Exposure Indices, two standard segregation measures commonly used in the literature (Valenzuela et al.,2014;Santos and Elacqua,2016). The Duncan index measures the percentage of students designated as low socioeconomic status who would have to be reallocated across schools for equal representation of students from all socioeconomic backgrounds within the district. The index ranges from 0 to 1, with higher scores indicating greater segregation. The Exposure index reports, for the average disadvantaged student in a given district, the proportion of students in her school who are non-disadvantaged. A low exposure index indicates that students from different socioeconomic strata attend separate schools. Therefore, we perform a difference-in-differences regression on school segregation at the municipality level to estimate the average impact of the SAE. Further details and a discussion on the identification assumptions can be found in Appendix A3. The results, displayed in Table A1, show that the introduction of the centralized admission system in the Chilean context did not significantly reduce school segregation. Even if the policy did not significantly reduce segregation, it may have had a positive impact on the representation of low-income students in ”desirable” schools. We accordingly investigate changes in the proportion of disadvantaged students in schools before and after the implementation of the centralized admission system. To this end, we employ a similar difference-indifferences strategy, but at the school level (see Appendix A3 for further details). Our analysis groups schools into categories based on their administration type (public, private voucher, or private non-voucher), academic achievement, average socioeconomic composition, level of selectivity before the reform (i.e., whether the school selected students through parental interviews, exams, or required proof of income), and their religious affiliation.11 The results are shown in Table A3. When examining schools of various administration types (public, voucher, and private) and different levels of selectivity before the reform, we find that the proportion of disadvantaged students increased in voucher schools but decreased in those that were more selective before the reform. Additionally, the representation of low-income students increased in schools with belowaverage performance on the SIMCE exam. While we do observe an increase in the presence of disadvantaged students in schools classified as having a medium to medium-high socioeconomic composition in their student body, we also note an increase in their presence in low-SES schools. Consequently, we do not identify consistent patterns that would suggest that low-income students are taking advantage of the system. 10We also replicate our analysis using an alternative, data-driven definition of education market. Results can be found in Appendix A3. 11The exact definition of these variables is provided in the next section. 12 3.3 Final sample and description of variables The previous results suggest that the elimination of school admissions barriers alone may have only a limited effect on the actual distribution of students. We therefore explore whether disadvantaged and non-disadvantaged families have different preferences regarding the prioritization of school attributes. For the analysis of parental preferences, we focus on application data from all pre-K students in 2019, when the reform was fully implemented nationwide. Table 2 reports the main statistics on students and school characteristics in our final sample. Disadvantaged students represent 44% of the total sample, and on average they listed three schools with a mean Euclidean distance of about 4 kilometers from their home. Panel B of Table 2 displays the statistics of the main school variables included in our analysis. We have incorporated a comprehensive set of characteristics to predict school demand. Specifically, these include the school’s enrollment size, the number of teachers per student, the number of educational assistants (educational psychologists, psychologists, speech therapists, social workers, special education assistants, and hall monitors), a binary variable indicating whether the school is public or voucher, a binary variable indicating whether the school is part of an integration program for children with special needs (PIE), an indicator of whether the school charges fees to parents, a binary variable taking a value of one if the school had a selective admission process before the reform, the average math and reading test scores on the national fourth-grade standardized exams, and the average socioeconomic classification of the school. Additionally, we include indicators of whether the school has a religious affiliation, whether it has a foreign name, and information on its infrastructure, extracurricular activities, and sports offerings. Notably, we also introduce a set of non-academic variables obtained from student and parental questionnaires conducted alongside the standardized exams. These variables encompass measures of students’ average self-efficacy, parental perceptions of the frequency of violent events in the school (including acts of student vandalism, fights, threats, and harassment among students and towards teachers), a measure of exclusion (the percentage of students reporting feeling discriminated against or left out, whether due to sexual orientation, immigrant status, gender, or other reasons), parents’ expectations that their children will attend higher education, and participation in extracurricular activities. Detailed information on the construction of these variables can be found in Appendix A4. 13 Figure 5: Distribution of applications, by disadvantaged status, 2017-2022 (a) Average distance between applicants’ homes and their preferred schools 0 .1 .2 .3 .4 .5 Density 0246810 Distance (km) Disadvantaged Not Disadvantaged (b) Ratio of applications over school vacancies 0 .05 .1 .15 .2 .25 Density 0 10 20 30 40 Average Applicants-to-Seat application by students Disadvantaged Not Disadvantaged (c) School quality (test scores) 0 .2 .4 .6 .8 Density -4 -2 0 2 4 Standarized test scores Disadvantaged Not Disadvantaged (d) Number of ranked schools 0 .1 .2 .3 .4 .5 Density 0 2 4 6 8 10 Number of ranked schools Disadvantaged Not Disadvantaged (e) School SES 0 .1 .2 .3 .4 Density Low Med-Low Medium Med-High High School SES Disadvantaged Not Disadvantaged (f) School selectiveness 0 .2 .4 .6 .8 Density Not selective Selective (prior to SAE implementation) Disadvantaged Not Disadvantaged Notes: This figure shows the distribution of applications by disadvantaged status. Panel (a) focuses on average home-school distance; Panel (b) on school demand (ratio of applications to vacancies); Panel (c) on the average SIMCE scores in math and reading; Panel (d) on the number of schools in families’ ranked lists; Panel (e) on the average SES of the student bodies of the schools listed; and Panel (f) on whether the school had a selective admissions process prior to the SAE coming into force. For more details on the variables, see Appendix A4. 14 Table 2: Summary statistics N Mean SD p10 p50 p90 A - Students Disadvantaged 133,017 0.44 0.50 0.00 0.00 1.00 Distance to listed schools (km) 133,017 3.56 30.05 0.54 1.56 5.60 Distance to 1st listed school (km) 133,017 3.16 30.33 0.27 1.21 5.50 Distance to 2nd listed school (km) 126,507 3.31 28.85 0.39 1.41 5.39 Distance to 3rd listed school (km) 73,606 3.70 34.34 0.47 1.61 5.49 Number of listed schools 133,017 3.09 1.77 2.00 3.00 5.00 Choice set size 133,017 34.49 22.21 9.00 29.00 67.00 B - Schools Enrollment 4,377 482.88 432.90 98.00 360.00 1,033.00 N of teachers per students 4,377 0.08 0.04 0.04 0.07 0.14 N of ed. assistants per students 4,377 0.02 0.02 0.01 0.02 0.04 N of sport infrastructure 4,377 1.28 0.57 1.00 1.00 2.00 N of art extracurriculars offered 4,377 3.06 1.73 1.00 3.00 5.00 N of classic sports offered 4,377 3.56 1.35 2.00 4.00 5.00 N of niche sports offered 4,377 1.04 0.99 0.00 1.00 2.00 Public 4,377 0.53 0.50 0.00 1.00 1.00 PIE 4,377 0.84 0.37 0.00 1.00 1.00 Foreign name 4,377 0.06 0.23 0.00 0.00 0.00 Religious 4,377 0.45 0.50 0.00 0.00 1.00 Any monthly fee 4,377 0.12 0.33 0.00 0.00 1.00 Had a selective admission 4,377 0.10 0.30 0.00 0.00 1.00 Math test scores 4,377 -0.06 0.83 -1.15 -0.07 1.00 Reading test scores 4,377 -0.00 0.83 -1.05 -0.03 1.10 Low SES school 4,377 0.23 0.42 0.00 0.00 1.00 Med-low SES school 4,377 0.44 0.50 0.00 0.00 1.00 Medium SES school 4,377 0.25 0.44 0.00 0.00 1.00 Med-high SES school 4,377 0.08 0.27 0.00 0.00 0.00 High SES school 4,377 0.00 0.07 0.00 0.00 0.00 Parental percep. of sch. violence 4,377 0.16 0.96 -0.85 -0.03 1.41 Exclusion 4,377 0.43 0.15 0.25 0.42 0.62 Parental college expectations 4,377 0.73 0.16 0.52 0.74 0.94 Self-efficacy 4,377 0.01 0.82 -0.92 0.02 0.98 Particip. in extracurricular activities 4,377 0.05 0.83 -0.94 0.03 1.10 Notes: This table displays summary statistics for the sample used in the parental preferences estimation. 15 4 Methodology In this section, we describe the methodology used to gain a better understanding of family preferences. Which school attributes do families value? Do families from different socioeconomic levels value the same attributes? The literature on parental school preferences suggests that parents highly value proximity and academic performance. Moreover, high-income families generally tend to prioritize academic quality, while low-income families prioritize proximity to home (Hastings et al.,2007;Burgess et al.,2015). However, a growing body of literature shows that parents may also value schools that improve outcomes that are not highly correlated with test scores (Beuermann et al.,2023). We analyze parental school preferences by fitting a discrete choice model to students’ rankordered preference lists. We follow a random utility framework assuming the standard model of a utility-maximizing individual (McFadden,1974). Let Uij denote family i’s utility from enrolling in school j, and let J={1,· · · , J}represent their set of available schools. Following Abdulkadiro˘glu et al. (2020)’s notation, the school ranked in k-order on a student’s choice list is: Rik = argmax j∈J\{Rim:m<k} Uij We define the utility of student iin school jas follows: Uij =γ1Zj+δ1Wij +Di×(γ2Zj+δ2Wij) + ϵij where Zjrepresents features of the school, Wij represents variables that depend on the applicant-school pair, such as the distance from student i’s home address to school j, and Di indicates whether the student is classified as disadvantaged. We assume ϵifollows an extreme value distribution of type I. Hence, it is a rank-ordered multinomial logit model, also known in the literature as exploded logit. The logit model implies the conditional likelihood of the rank list Ri= (Ri1,· · · , Ril(i)), with l(i) being the length of the list submitted by the student, is: L(Ri|Xi, Zj, Wi) = l(i) Y k=1 exp(γ1Zj+δ1Wij +Di×(γ2Zj+δ2Wij)) Pj∈J\{Rim:m<k}exp(γ1Zj+δ1Wij +Di×(γ2Zj+δ2Wij)) Hence, the probability of observing a specific ranking can be written as the product of these terms, representing a sequential decision in which the student first chooses the most preferred school, then the next most preferred school among the remaining options, and so on. Note that this model imposes the assumption of the independence of irrelevant alternatives (Long and Freese,2006). This means that the model assumes that the relative preferences for two alternatives do not depend on the other alternatives available. Thus, in this setting, the ranking of school A versus school B remains the same whether or not school C is available as an 16 alternative. In addition, even though we include a very rich set of school characteristics, there still could be unobserved school factors that influence parents’ choices and are correlated with attributes in our model. In the Chilean context, parents select all the schools they prefer (a minimum of two if they live in an urban area) in order of preference and without any residential proximity restrictions (i.e., parents do not have to choose schools within a particular district). This is an important advantage compared to other contexts where parents can nominate a finite number of schools or there are constraints on their possible choices.12 In addition, since our focus is on pre-K, all schools have most of their vacancies available, such that students have a high chance of being accepted at their listed schools. One important decision is how to define the set of schools that families choose from, as there are no legal or geographical restrictions in Chile. We define school districts as municipalities. This definition is based on the observation that, in our sample, approximately 90% of pre-K students attend a school within their municipality of residence. As a robustness check, we also use a different, data-driven definition for education markets. We acknowledge that this model may, to a degree, conflate families’ preferences with the constraints they face. In other words, the differences in preferences between disadvantaged and non-disadvantaged families could, in part, reflect disparities in the attributes of accessible schools due to residential or spatial inequalities. We attempt to capture this by standardizing school attributes at the education market level. Thus, instead of comparing preferences for these attributes in absolute terms, we do it at a relative level. This means that we can ascertain whether families are choosing better quality schools, for example, from among the options they have available. In addition, we run a conditional logit model on families’ first preference. We do this as a way of acknowledging that some parents might have only one school in mind for their children, and they only choose the other school(s) on their list to meet the government’s minimum requirement. In other words, the first preference could more reliably capture families’ true preferences. We also assess the robustness of our results to different specifications. For instance, we replicate our analysis for a sub-sample consisting of only urban municipalities, as students living in these municipalities may have a significantly larger school choice set. An important caveat of this analysis is that, even though we include a rich set of variables that intend to capture several dimensions of the determinants of parental preferences, unobserved variables could still exist. 5 Main results In this section, we present the empirical results from the parental preference estimations. Column (1) of Table 3 displays the results from the rank-ordered logit estimation, while column 12For instance, in England, parents can list between three and six schools (Burgess et al.,2015). 17 (2) displays the results of the logit estimation conditional on the first preference only. Due to the large set of variables, we only report the coefficients on the interaction with students’ disadvantaged status. The non-interacted terms can be found in Table A4. The results should be interpreted in terms of the exponentiated coefficients, which can be interpreted as an odds ratio. That is, the coefficients indicate the percent change in the odds of a particular school being ranked ahead of the base category for a unit increase in the explanatory variable, holding other variables constant. As expected, the results indicate that preferences vary among disadvantaged and nondisadvantaged families. Low-SES students show a stronger aversion to distance from school and place less importance on academic performance compared to high-SES students. They tend to select larger schools with a greater number of teaching and non-teaching staff relative to the student body. They are also less likely to apply to schools that require a co-payment or ones that had a selective admission process in place prior to the reform. This finding is surprising, given that “priority” students typically do not have to pay school fees in most schools.13 This could suggest that some families are unaware of their eligibility for these benefits and, consequently, believe they lack the resources to enroll their children in fee-charging schools. Furthermore, disadvantaged families are notably less inclined to apply to schools in which the student body comes from a higher socioeconomic status, which we call high-SES schools (note that the baseline comparison group is comprised of low-SES schools). These results may suggest that, despite the inclusion of low-income families in previously selective and high-SES schools, students from these families may still not feel fully integrated or welcomed within the school or its community (Bell,2009). Another contributing factor could be that families typically rely on their social networks when making school choices, which can make these decisions more persistent. Interestingly, we observe that disadvantaged families’ choices are more strongly influenced by other non-academic attributes of the school, which are often omitted from parental preference analyses due to data availability constraints. Low-SES parents tend to favor schools that offer more “classical” sports, have a foreign name, or possess a religious affiliation. These families are also more likely to prioritize schools with a more favorable school climate, reflected in their preference for schools with fewer reported violent incidents, schools where students report facing less discrimination and exclusion, and schools where students demonstrate higher levels of selfefficacy. Surprisingly, disadvantaged families place less value on the sports infrastructure and extracurricular activities offered by the school compared to non-disadvantaged families. This result may also be related to the socioeconomic level of the feasible set of schools. When we examine the number of sports offered, distinguishing between classical sports (e.g., soccer, basketball, 13In our sample, 92% of the schools participate in the SEP policy, a national targeted voucher implemented in Chile in 2008, which increased the funding for disadvantaged students by 50%. Participating schools cannot charge out-of-pocket tuition to disadvantaged or ”priority” students. 18 volleyball, and table tennis) and niche or elite sports (e.g., hockey, golf, swimming, etc.), we find that low-income parents choose schools with more classical sports, while high-income parents choose schools with more elite sports. Elite sports include those that require special and costly infrastructure and more specialized instructors. As previously discussed, because this model might, to some extent, conflate parental preferences with parents’ local constraints, as a next step, we standardize the independent variables to account for the possibility that the observed preferences could also be influenced by variations in school availability. Table 4 reports the results for these estimations for both the full ranked list (column 1) and first choices only (column 2). Again, we only report the coefficients on the interaction with the disadvantaged status of the student (the non-interacted terms can be found in Table A5). Overall, we find similar patterns: all else being equal, increasing the distance to school reduces the odds that a disadvantaged parent prefers that school compared to a non-disadvantaged parent. The odds that disadvantaged families rank a previously fee-charging school or selective school are 15 and 9 percent lower, respectively. In addition, they are less likely to rank highperforming schools and their probability of ranking a medium-SES school above a low-SES school is almost 80 percent lower than high-SES parents. Other aspects of the data reinforce the idea that disadvantaged families value schools and communities where they feel more welcomed. A decrease of 1 standard deviation in the indicator of parental perception of school violence makes low-SES parents 2% less likely to rank a school compared to a high-SES parent. Meanwhile, disadvantaged parents are 5% more likely to choose a school with a religious affiliation. An increase of 1 SD in a school’s indicator of exclusion or discrimination among students, likewise, is associated with a 2% drop in disadvantaged families’ odds of ranking that school. All in all, these results show that parents from more vulnerable contexts are concerned to a significant degree about various non-academic outcomes, such as their children’s safety, and exhibit less interest in academic aspects than their more affluent counterparts. These findings have important implications for our understanding of parental preferences, especially among low-SES families. Indeed, this is a crucial insight, given evidence that non-academic outcomes and school’s test scores may be only weakly related (Beuermann et al.,2023). 5.1 Robustness checks Data-driven education markets. We also consider an alternative definition of education markets, following Kutscher et al. (2023). Specifically, we construct data-driven education markets, assuming two municipalities are part of the same school district if:14 14The threshold of 7.5% was chosen because, on the one hand, it needed to be less than 10% to allow for a higher level of aggregation than at the municipality level. This choice was informed by our data, which revealed that approximately 90% of pre-K students attend schools within their municipality of residence. On the other hand, we wanted to avoid very large markets, as we are aware that parents with children in primary education 19 1. 7.5% or more of the students who live in municipality iattend a school in municipality j, or vice versa, and 2. the travel distance by car between the centroids of the municipalities iand jis less than 2 hours. Figure 6 illustrates the comparison between municipalities and this new definition of datadriven markets in three major metropolitan areas of Chile. The black lines delineate the borders of the data-driven markets, while the white lines show municipal borders. We also replicate our analysis including a small buffer zone on the edge of the cities, as shown in Figure 6, which effectively removes students living close to the border of two markets, for whom we might be defining the set of eligible schools incorrectly. Table 5 displays the results of these regressions. Columns (1) and (2) report the coefficients from the parental preference estimations with the data-driven markets, for both the full list of ranked schools and the top choice. Columns (3) and (4) report the coefficients for these same estimations for the data-driven markets but with the buffer described above. The results remain consistent. Urban municipalities. To ensure that rural-urban differences do not drive our findings, we re-estimate the regression models excluding rural districts from the sample. We define rural municipalities as municipalities in which at least half of schools are rural. Students living in these municipalities have a significantly smaller school choice set. Our results, reported in Table 6, remain robust to this exercise. Ranked preferences without top choice. We re-estimated the benchmark model omitting each applicant’s top school choice. The results can be found in ??. The premise is that disadvantaged families might not understand how the DA algorithm works and may want to “secure” a vacancy for their children by listing as their top choice their neighborhood school, which is familiar to them. However, the resulting estimates did not differ substantially from the main results, suggesting that low-SES parents’ top choices follow similar selection patterns as their lower-ranked schools. 6 Conclusions Centralized admission mechanisms offer an equitable way of assigning students to schools because each pupil is treated equally in the assignment process. Families list their preferences, and students are allocated to schools based on the available seats and government-established priorities. This system eliminates schools’ discriminatory selective practices and enhances school tend to select schools close to their homes. 20 Figure 6: Municipalities and data-driven markets Gran Valparaíso Gran Concepción Gran Santiago 100 km N 0.0 0.2 0.4 0.6 0.8 % of disadv. students Notes: This figure displays the municipalities in the most populous region of Chile, encompassing three major metropolitan areas. Each municipality is color-coded based on the percentage of disadvantaged students in pre-K in 2019 across both public and private voucher schools. Black lines delineate the borders of the data-driven markets we constructed. On the left, a zoomed-in map highlights the area enclosed by a black square on the main map. The top left panel shows the market borders without buffers, while the bottom left panel displays market borders with a 500m buffer, which effectively removes from our sample students living close to the borders between education markets. It should be noted that the buffer only excludes students close to the border, not schools. choice, particularly for families from disadvantaged backgrounds. Yet, empirical evidence indicates such mechanisms do not necessarily lead to reduced socioeconomic school segregation. In this paper, we first confirm previous findings that the introduction of the centralized admission system has not significantly diminished school segregation. We then explore parental preferences among disadvantaged and non-disadvantaged families. We discover that the choices made by low-SES families are more strongly associated with non-academic factors. Low-SES parents tend to prioritize schools with a more favorable school environment, better indicators of student self-efficacy, fewer reported violent events, and a religious affiliation. They also tend to select schools that are close to where they live, have lower average test scores, a lower socioeconomic composition of the student body, and ones that had less selective admission processes before the reform. Hence, our results suggest that the elimination of barriers to school admission alone may have only a limited effect on the actual distribution of students. In addition to information fric21 References Abdulkadiro˘glu, Atila, Parag A Pathak, and Alvin E Roth, “Strategy-proofness versus efficiency in matching with indifferences: Redesigning the NYC high school match,” American Economic Review, 2009, 99 (5), 1954–78. , , Jonathan Schellenberg, and Christopher R Walters, “Do parents value school effectiveness?,” American Economic Review, 2020, 110 (5), 1502–39. Ajayi, Kehinde F, Willa H Friedman, and Adrienne M Lucas, “When Information is Not Enough: Evidence from a Centralized School Choice System,” Working Paper, National Bureau of Economic Research 2020. Allende, Claudia, Francisco Gallego, Christopher Neilson et al., “Approximating the equilibrium effects of informed school choice,” Technical Report 2019. Arteaga, Felipe, Adam J Kapor, Christopher A Neilson, and Seth D Zimmerman, “Smart matching platforms and heterogeneous beliefs in centralized school choice,” The Quarterly Journal of Economics, 2022, 137 (3), 1791–1848. Bell, Courtney A, “All choices created equal? The role of choice sets in the selection of schools,” Peabody journal of education, 2009, 84 (2), 191–208. Beuermann, Diether W, C Kirabo Jackson, Laia Navarro-Sola, and Francisco Pardo, “What is a good school, and can parents tell? Evidence on the multidimensionality of school output,” The Review of Economic Studies, 2023, 90 (1), 65–101. Burgess, Simon, Ellen Greaves, Anna Vignoles, and Deborah Wilson, “What parents want: School preferences and school choice,” The Economic Journal, 2015, 125 (587), 1262– 1289. Carrasco, Alejandro, “‘Selecci´on escolar’y desigualdad educacional en Chile:¿ qu´e tan coactiva es la regulaci´on que la proh´ıbe?,” 2014. Denice, Patrick and Betheny Gross, “Choice, preferences, and constraints: Evidence from public school applications in Denver,” Sociology of Education, 2016, 89 (4), 300–320. Elacqua, G., M. Schneider, and J. Buckley, “School choice in Chile: Is it class or the classroom,” Journal of Policy Analysis and management, 2006, 23 (3), 577–601. Elacqua, Gregory, Isabel Jacas, Thomas Krussig, Carolina Mendez, Christopher Neilson, Alonso Roman, and Sammara Soares, “Centralized Student Assignment Systems: Implementation Guide,” IDB Technical Note 02184, 2021. 28 Epple, Dennis, Richard E Romano, and Miguel Urquiola, “School vouchers: A survey of the economics literature,” Journal of Economic Literature, 2017, 55 (2), 441–92. Glazerman, Steven and Dallas Dotter, “Market signals: Evidence on the determinants and consequences of school choice from a citywide lottery,” Educational Evaluation and Policy Analysis, 2017, 39 (4), 593–619. Hastings, Justine S and Jeffrey M Weinstein, “Information, school choice, and academic achievement: Evidence from two experiments,” The Quarterly journal of economics, 2008, 123 (4), 1373–1414. , Richard Van Weelden, and Jeffrey Weinstein, “Preferences, information, and parental choice behavior in public school choice,” Working Paper, National Bureau of Economic Research 2007. Hastings, Justine, Thomas J Kane, and Douglas O Staiger, “Heterogeneous preferences and the efficacy of public school choice,” Working Paper, National Bureau of Economic Research 2009. Honey, Ngaire and Alejandro Carrasco, “A new admission system in Chile and its foreseen moderate impact on access for low-income students,” Educational Evaluation and Policy Analysis, 2022, p. 01623737221093374. Hsieh, Chang-Tai and Miguel Urquiola, “The effects of generalized school choice on achievement and stratification: Evidence from Chile’s voucher program,” Journal of Public Economics, 2006, 90 (8-9), 1477–1503. Kutscher, Macarena, Shanjukta Nath, and Sergio Urzua, “Centralized Admission Systems and School Segregation: Evidence from a National Reform,” forthcoming in Journal of Public Economics, 2023. Lauen, Lee Lauen, “Contextual explanations of school choice,” Sociology of Education, 2007, 80 (3), 179–209. Long, J Scott and Jeremy Freese,Regression models for categorical dependent variables using Stata, Vol. 7, Stata press, 2006. Luflade, Margaux, “The value of information in centralized school choice systems (job market paper),” Job Market Paper, Duke University 2017. McFadden, Daniel, “Conditional logit analysis of qualitative choice behavior,” in Paul Zarembka, ed., Fontiers in Econometrics, New York: Academic press, 1974, pp. 105–142. Mizala, Alejandra and Pilar Romaguera, “School performance and choice: the Chilean experience,” Journal of Human Resources, 2000, pp. 392–417. 29 Santos, Humberto and Gregory Elacqua, “Socioeconomic school segregation in Chile: Parental choice and a theoretical counterfactual analysis,” Commission for Latin America and the Caribbean, 2016, p. 124. Valenzuela, Juan Pablo, Cristian Bellei, and Danae de los R´ıos, “Socioeconomic school segregation in a market-oriented educational system. The case of Chile,” Journal of education Policy, 2014, 29 (2), 217–241. 30 Appendix A1 Government priority classification The only indicator of socioeconomic status available for pre-K students is the government SES index (vulnerability index). This indicator, however, has some limitations. Students categorized as vulnerable have priority in the algorithm. Hence, as discussed in the main text, families may have reacted endogenously to the policy by, for example, obtaining vulnerable status in order to gain more favorable treatment in school admission. We explore this possibility in Figure A3, which displays the test for pre-trend differences in the percentage of vulnerable students enrolled in pre-K in the context of an event-study analysis. We find no evidence of strategic behavior among families before and after the SAE was implemented. Figure A1: Event study for the difference in percentage of disadvantaged students in pre-K at the regional level 95% confidence interval -.05 -.04 -.03 -.02 -.01 0 .01 .02 .03 Change in % of disadvantaged students in PK (p.p) -5 -4 -3 -2 -1 0 1 Years to SAE implementation Notes: This figure displays the estimates and corresponding 95% confidence intervals of the coefficients βτin the following specification yrt =γr+δtP0 τ=−4βτDτ rt +εrt, where yrt is the percentage of disadvantaged students enrolled in pre-K during year tin region r.Drt equals one for regions where SAE was implemented, γrare region fixed effects and δtis a time fixed effect. 31 A2 Segregation measures The Duncan index measures the percentage of disadvantaged students who have to be reallocated across schools for equal representation of students from all socioeconomic backgrounds within the district. The index ranges from 0 to 1, with higher scores indicating greater segregation. Formally, the Duncan index for a specific year and school district can be computed as follows: 1 2 J X j=1  dj d−nj n , where djis the number of disadvantaged students in school j,njis the number of nondisadvantaged students in school j, and dand ndenotes the total number of disadvantaged and non-disadvantaged students in the district, respectively. The Exposure index, meanwhile, reports the proportion of students in the average disadvantaged student’s school in a given district are non-disadvantaged. A low exposure index indicates that students of different socioeconomic statuses attend separate schools. The formula is as follows: J X j dj d×nj tj!(1) where djis the number of disadvantaged students in school j,njis the number of nondisadvantaged students in school j,ddenotes the total number of disadvantaged students in the district, and tjis the total population at school j. Figure A2 displays the spatial distribution of the Duncan and Exposure indices at the municipality level for 2015-2019. We can see that there is high variation in both indices, mostly due to the fact that there is a large dispersion in municipalities’ characteristics across regions. 32 Figure A2: Spatial Distribution of Segregation Measures: 2015-2019 0 1 2 3 Density 0.2 .4 .6 .8 1 Index Duncan Exposure Notes: This figure displays municipality-level values of the Duncan and Exposure indices between 2015 and 2019. 33 A3 SAE and School Segregation: Identification We take advantage of the gradual implementation of the policy across regions to estimate its impact on school segregation. In particular, we employ a difference-in-differences strategy following Kutscher et al. (2023): Yirt =γi+λt+δSAEirt +X′ ir +εirt,(2) where Yirt is the school segregation in municipality iin region rand year t,SAEirt is the treatment variable, which takes a value of one if the SAE program was implemented in the region rin year t, and zero otherwise; γiand λtare municipality and year fixed effects. X′ ir is a vector of municipality variables, including the percentage of private schools and the total school population in municipality i, prior to SAE implementation. The coefficient of interest is δ, which captures the effect of the centralized admission system. The identification of Equation 2 depends on several assumptions. First, the implementation of the policy should be exogenous to pre-existing levels of school segregation. The only consideration when the policy was instituted across regions was the size of each region’s student population.15 Second, there should not be any responses in anticipation of the treatment. As discussed previously, we did not find evidence that families adjusted their government vulnerability classification in anticipation of the policy (see discussion in Appendix A1). Although we cannot test this concern, we do not think that families reacted by moving to a new location because the main effect of the policy was to increase the available school options for disadvantaged families. Finally, we rely on the conventional common trends assumption. Figure A3 in the appendix suggests that the treatment and control regions had similar trends in school segregation in the absence of SAE. 15The law established a fixed calendar for the scaling up of the policy, with the only consideration being the participating student population, from 20% in 2016 to 100% in 2019. There were no considerations related to school segregation. 34 Figure A3: Event study – indices -.02 -.01 0 .01 .02 Change in Index (p.p) -4 -3 -2 -1 0 Years since SAE implementation Duncan Exposure Notes: This figure displays the estimates and corresponding 95% confidence intervals of the coefficients βτin the following specification imrt =γm+δt+P0 τ=−4βτDτ mrt +εmrt, where imrt is the respective index (Duncan, Exposure) for municipality m, in region r, in year t.Dmrt equals one for regions where SAE was implemented, γmare municipality fixed effects and δtis a time fixed effect. Standard errors are clustered at the region level. Table A1: SAE effect on segregation indices (1) (2) (3) (4) Duncan Index Exposure Index SAE -0.005 -0.006 -0.004 -0.002 (0.008) (0.008) (0.008) (0.006) Constant 0.287∗∗∗ 0.316∗∗∗ 0.335∗∗∗ 0.301∗∗∗ (0.002) (0.023) (0.002) (0.018) Observations (Municipality ×Year) 1,680 1,680 1,680 1,680 R20.784 0.784 0.849 0.849 Municipality FE Y Y Y Y Year FE Y Y Y Y Controls N Y N Y Notes: This table displays the results from estimating Equation 2. Clustered standard errors at the region level are displayed in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 35 Figure A4: Event study – indices, data-driven markets -.02 0 .02 .04 .06 Change in Index (p.p) -4 -3 -2 -1 0 Years since SAE implementation Duncan Exposure Notes: This figure displays the estimates and corresponding 95% confidence intervals of the coefficients βτin the following specification imrt =γm+δt+P0 τ=−4βτDτ mrt +εmrt, where imrt is the respective index (Duncan, Exposure) for data-driven market m, in region r, in year t.Dmrt equals one for regions where SAE was implemented, γrare market fixed effects and δtis a time fixed effect. Standard errors are clustered at the region level. 36 Table A2: SAE effect on segregation indices, data-driven markets (1) (2) (3) (4) Duncan Index Exposure Index SAE -0.013 -0.015 0.003 0.005 (0.009) (0.009) (0.008) (0.007) Constant 0.292∗∗∗ 0.325∗∗∗ 0.328∗∗∗ 0.297∗∗∗ (0.002) (0.022) (0.002) (0.024) Observations (Market ×Year) 1,290 1,290 1,290 1,290 R20.778 0.778 0.842 0.842 Municipality FE Y Y Y Y Year FE Y Y Y Y Controls N Y N Y Notes: This table displays the results from estimating Equation 2, estimated with data-driven markets instead of municipalities. Clustered standard errors at the region level are displayed in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 37 Figure A7: Pre-K schools by type, 2013–2022 0 500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000 5,500 6,000 Number of schools serving PK students 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 Public Non-fee voucher Fee voucher No info voucher Private School type: Notes: This figure displays how schools that offer pre-K are distributed based on their administrative classification, including public, voucher (with or without monthly charges), and private. 44 Figure A8: SAE applicants and their enrollment status following the admission process .05 .15 .25 .35 .45 .55 .65 .75 .85 .95 % of applicants 2017 2018 2019 2020 2021 2022 Pre-Kinder Enrolled in SAE-assigned school Enrolled, not in SAE-assigned school Not enrolled .05 .15 .25 .35 .45 .55 .65 .75 .85 .95 % of applicants 2017 2018 2019 2020 2021 2022 Kinder Enrolled in SAE-assigned school Enrolled, not in SAE-assigned school Not enrolled .05 .15 .25 .35 .45 .55 .65 .75 .85 .95 % of applicants 2017 2018 2019 2020 2021 2022 1st grade Enrolled in SAE-assigned school Enrolled, not in SAE-assigned school Not enrolled Notes: These figures display the enrollment status of pre-K, kindergarten, and first grade students who took part in SAE, indicating whether they enrolled in their SAE-assigned schools, enrolled in other schools, or did not enroll at all. 45 Table A4: Families’ preferences (cont. from Table 3) (1) (2) Full list of ranked schools On the 1st ranked school Distance 0.0004∗∗∗ -0.0406∗∗∗ (0.0001) (0.0015) N of enrolled students 0.0004∗∗∗ 0.0006∗∗∗ (0.0000) (0.0000) N of teaching staff N of enrolled students -12.9520∗∗∗ -10.4445∗∗∗ (0.1850) (0.3368) N of non-teaching staff N of enrolled students -10.5894∗∗∗ -13.0454∗∗∗ (0.4208) (0.7732) N of Sport infrastructure 0.0307∗∗∗ 0.0639∗∗∗ (0.0041) (0.0076) N of Art extacurr. offered 0.0329∗∗∗ 0.0411∗∗∗ (0.0014) (0.0026) N of classic sports offered 0.0088∗∗∗ 0.0055 (0.0019) (0.0035) N of niche sports offered 0.0568∗∗∗ 0.0577∗∗∗ (0.0023) (0.0042) Public 0.1823∗∗∗ 0.1559∗∗∗ (0.0077) (0.0147) PIE 0.1564∗∗∗ 0.1914∗∗∗ (0.0057) (0.0106) Foreign name 0.0095 -0.0500∗∗∗ (0.0079) (0.0152) Religious 0.0062 0.0478∗∗∗ (0.0048) (0.0089) Any monthly fee 0.0480∗∗∗ -0.0766∗∗∗ (0.0071) (0.0136) Had a selective admission 0.2407∗∗∗ 0.3556∗∗∗ (0.0056) (0.0105) Reading test scores 0.0359∗∗∗ 0.0890∗∗∗ (0.0065) (0.0121) Math test scores 0.2070∗∗∗ 0.2399∗∗∗ (0.0059) (0.0109) Med-low SES school 0.5557∗∗∗ 0.3446∗∗∗ (0.0155) (0.0264) Medium SES school 1.1126∗∗∗ 0.8376∗∗∗ (0.0171) (0.0296) Med-high SES school 1.4092∗∗∗ 1.2558∗∗∗ (0.0192) (0.0339) High SES school 1.5061∗∗∗ 1.6199∗∗∗ (0.0242) (0.0418) Parental percep. of sch. violence -0.0965∗∗∗ -0.1304∗∗∗ (0.0041) (0.0077) Exclusion 0.3968∗∗∗ 0.4062∗∗∗ (0.0250) (0.0465) Parental college expectations 1.6737∗∗∗ 1.8716∗∗∗ (0.0336) (0.0622) Self-efficacy -0.0527∗∗∗ -0.0833∗∗∗ (0.0046) (0.0085) Particip. in extracurricular activities 0.0085∗∗ 0.0266∗∗∗ (0.0036) (0.0068) Observations (individuals ×choice set) 4,392,543 3,988,127 Pseudo-R20.127 0.160 Notes: This table displays families’ preferences following Section 4. Standard errors are displayed in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 46 Table A5: Families’ preferences with standardized school characteristics (cont. from Table 4) (1) (2) Full list of ranked schools On the 1st ranked school STD Distance -0.5651∗∗∗ -0.8479∗∗∗ (0.0030) (0.0062) STD N of enrolled students 0.2621∗∗∗ 0.3175∗∗∗ (0.0022) (0.0040) STD N of teaching staff N of enrolled students -0.2983∗∗∗ -0.2121∗∗∗ (0.0073) (0.0136) STD N of non-teaching staff N of enrolled students 0.0093∗-0.0277∗∗ (0.0055) (0.0107) STD N of Sport infrastructure 0.0188∗∗∗ 0.0365∗∗∗ (0.0024) (0.0045) STD N of Art extacurr. offered 0.0558∗∗∗ 0.0704∗∗∗ (0.0024) (0.0045) STD N of classic sports offered 0.0055∗∗ -0.0026 (0.0025) (0.0048) STD N of niche sports offered 0.0468∗∗∗ 0.0402∗∗∗ (0.0023) (0.0043) Public 0.0543∗∗∗ 0.0355∗∗ (0.0076) (0.0146) PIE 0.0607∗∗∗ 0.0965∗∗∗ (0.0058) (0.0109) Foreign name 0.0057 -0.0605∗∗∗ (0.0080) (0.0154) Religious 0.0051 0.0422∗∗∗ (0.0048) (0.0091) Any monthly fee 0.0971∗∗∗ -0.0314∗∗ (0.0072) (0.0140) Had a selective admission 0.2686∗∗∗ 0.3680∗∗∗ (0.0057) (0.0108) STD Reading test scores 0.0365∗∗∗ 0.0912∗∗∗ (0.0051) (0.0096) STD Math test scores 0.1725∗∗∗ 0.1934∗∗∗ (0.0046) (0.0087) Med-low SES school 0.7700∗∗∗ 0.5313∗∗∗ (0.0153) (0.0261) Medium SES school 1.5298∗∗∗ 1.2130∗∗∗ (0.0164) (0.0286) Med-high SES school 1.9713∗∗∗ 1.7488∗∗∗ (0.0185) (0.0332) High SES school 2.2075∗∗∗ 2.2453∗∗∗ (0.0235) (0.0412) STD Parental percep. of sch. violence -0.0573∗∗∗ -0.0823∗∗∗ (0.0035) (0.0068) STD Exclusion 0.0404∗∗∗ 0.0463∗∗∗ (0.0036) (0.0068) STD Parental college expectations 0.0216∗∗∗ 0.0865∗∗∗ (0.0048) (0.0093) STD Self-efficacy -0.0567∗∗∗ -0.0765∗∗∗ (0.0033) (0.0063) STD Particip. in extracurricular activities 0.0363∗∗∗ 0.0544∗∗∗ (0.0029) (0.0055) Observations (individuals ×choice set) 4,373,266 3,967,881 Pseudo-R20.156 0.211 Notes: This table displays families’ preferences following Section 4. Standard errors are displayed in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 47 Table A6: Families’ preferences with data-driven markets (cont. fromTable 5) Data-driven markets Data-driven markets w/ buffer (1) (2) (3) (4) Full list of ranked schools On the 1st ranked school Full list of ranked schools On the 1st ranked school Distance -0.0738∗∗∗ -0.1741∗∗∗ -0.0868∗∗∗ -0.1952∗∗∗ (0.0007) (0.0017) (0.0008) (0.0018) N of enrolled students 0.0005∗∗∗ 0.0006∗∗∗ 0.0005∗∗∗ 0.0006∗∗∗ (0.0000) (0.0000) (0.0000) (0.0000) N of teaching staff N of enrolled students -10.6928∗∗∗ -7.1179∗∗∗ -10.9379∗∗∗ -7.3231∗∗∗ (0.1870) (0.3363) (0.1975) (0.3540) N of non-teaching staff N of enrolled students -10.4376∗∗∗ -13.1592∗∗∗ -10.0972∗∗∗ -12.8901∗∗∗ (0.4220) (0.7845) (0.4477) (0.8311) N of Sport infrastructure 0.0339∗∗∗ 0.0695∗∗∗ 0.0316∗∗∗ 0.0697∗∗∗ (0.0041) (0.0076) (0.0043) (0.0081) N of Art extacurr. offered 0.0276∗∗∗ 0.0354∗∗∗ 0.0290∗∗∗ 0.0363∗∗∗ (0.0014) (0.0026) (0.0015) (0.0028) N of classic sports offered 0.0078∗∗∗ 0.0061∗0.0073∗∗∗ 0.0036 (0.0019) (0.0035) (0.0020) (0.0037) N of niche sports offered 0.0517∗∗∗ 0.0523∗∗∗ 0.0506∗∗∗ 0.0505∗∗∗ (0.0023) (0.0042) (0.0024) (0.0044) Public 0.2192∗∗∗ 0.1667∗∗∗ 0.2037∗∗∗ 0.1523∗∗∗ (0.0076) (0.0144) (0.0081) (0.0153) PIE 0.1531∗∗∗ 0.1863∗∗∗ 0.1525∗∗∗ 0.1926∗∗∗ (0.0057) (0.0105) (0.0060) (0.0111) Foreign name -0.0087 -0.0618∗∗∗ -0.0077 -0.0650∗∗∗ (0.0080) (0.0153) (0.0085) (0.0162) Religious 0.0237∗∗∗ 0.0567∗∗∗ 0.0338∗∗∗ 0.0620∗∗∗ (0.0048) (0.0089) (0.0050) (0.0094) Any monthly fee 0.0038 -0.1191∗∗∗ 0.0349∗∗∗ -0.0804∗∗∗ (0.0070) (0.0135) (0.0075) (0.0143) Had a selective admission 0.2439∗∗∗ 0.3330∗∗∗ 0.2448∗∗∗ 0.3262∗∗∗ (0.0055) (0.0103) (0.0059) (0.0109) Reading test scores 0.0501∗∗∗ 0.1095∗∗∗ 0.0381∗∗∗ 0.0995∗∗∗ (0.0065) (0.0122) (0.0069) (0.0128) Math test scores 0.1936∗∗∗ 0.2279∗∗∗ 0.1981∗∗∗ 0.2300∗∗∗ (0.0059) (0.0110) (0.0062) (0.0116) Med-low SES school 0.4802∗∗∗ 0.2567∗∗∗ 0.4689∗∗∗ 0.2612∗∗∗ (0.0157) (0.0268) (0.0163) (0.0279) Medium SES school 1.0230∗∗∗ 0.6966∗∗∗ 0.9909∗∗∗ 0.6949∗∗∗ (0.0171) (0.0298) (0.0179) (0.0312) Med-high SES school 1.3729∗∗∗ 1.1465∗∗∗ 1.3215∗∗∗ 1.1225∗∗∗ (0.0192) (0.0341) (0.0201) (0.0358) High SES school 1.4999∗∗∗ 1.5370∗∗∗ 1.4472∗∗∗ 1.5035∗∗∗ (0.0240) (0.0416) (0.0252) (0.0438) Parental percep. of sch. violence -0.1052∗∗∗ -0.1456∗∗∗ -0.1141∗∗∗ -0.1532∗∗∗ (0.0041) (0.0077) (0.0043) (0.0081) Exclusion 0.4536∗∗∗ 0.4400∗∗∗ 0.4611∗∗∗ 0.4461∗∗∗ (0.0252) (0.0470) (0.0268) (0.0498) Parental college expectations 1.7039∗∗∗ 1.8500∗∗∗ 1.6912∗∗∗ 1.8173∗∗∗ (0.0338) (0.0632) (0.0358) (0.0666) Self-efficacy -0.0470∗∗∗ -0.0810∗∗∗ -0.0379∗∗∗ -0.0740∗∗∗ (0.0046) (0.0086) (0.0049) (0.0090) Particip. in extracurricular activities 0.0142∗∗∗ 0.0434∗∗∗ 0.0159∗∗∗ 0.0491∗∗∗ (0.0036) (0.0067) (0.0038) (0.0071) Observations (individuals ×choice set) 7,998,524 7,504,891 7,293,587 6,849,561 Pseudo-R20.131 0.172 0.136 0.178 Notes: This table displays families’ preferences following Section 4. Standard errors are displayed in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 48 Table A7: Families’ preferences without rural municipalities (cont. from Table 6) (1) (2) Full list of ranked schools On the 1st ranked school Distance 0.0009∗∗∗ -0.0275∗∗∗ (0.0001) (0.0017) N of enrolled students 0.0004∗∗∗ 0.0005∗∗∗ (0.0000) (0.0000) N of teaching staff N of enrolled students -12.0320∗∗∗ -9.9359∗∗∗ (0.2016) (0.3779) N of non-teaching staff N of enrolled students -11.5219∗∗∗ -14.2443∗∗∗ (0.4592) (0.8709) N of Sport infrastructure 0.0344∗∗∗ 0.0608∗∗∗ (0.0042) (0.0080) N of Art extacurr. offered 0.0335∗∗∗ 0.0441∗∗∗ (0.0015) (0.0027) N of classic sports offered 0.0117∗∗∗ 0.0074∗∗ (0.0019) (0.0037) N of niche sports offered 0.0570∗∗∗ 0.0564∗∗∗ (0.0023) (0.0044) Public 0.1925∗∗∗ 0.1743∗∗∗ (0.0081) (0.0156) PIE 0.1513∗∗∗ 0.1920∗∗∗ (0.0059) (0.0110) Foreign name -0.0041 -0.0590∗∗∗ (0.0082) (0.0157) Religious 0.0050 0.0458∗∗∗ (0.0050) (0.0093) Any monthly fee 0.0544∗∗∗ -0.0708∗∗∗ (0.0072) (0.0140) Had a selective admission 0.2311∗∗∗ 0.3402∗∗∗ (0.0058) (0.0108) Reading test scores 0.0427∗∗∗ 0.1023∗∗∗ (0.0068) (0.0129) Math test scores 0.2148∗∗∗ 0.2545∗∗∗ (0.0062) (0.0117) Med-low SES school 0.5323∗∗∗ 0.3019∗∗∗ (0.0187) (0.0330) Medium SES school 1.0848∗∗∗ 0.7557∗∗∗ (0.0201) (0.0358) Med-high SES school 1.3838∗∗∗ 1.1638∗∗∗ (0.0222) (0.0402) High SES school 1.4581∗∗∗ 1.4969∗∗∗ (0.0266) (0.0472) Parental percep. of sch. violence -0.0976∗∗∗ -0.1236∗∗∗ (0.0043) (0.0082) Exclusion 0.4289∗∗∗ 0.4554∗∗∗ (0.0267) (0.0508) Parental college expectations 1.7461∗∗∗ 2.1206∗∗∗ (0.0368) (0.0704) Self-efficacy -0.0659∗∗∗ -0.1015∗∗∗ (0.0049) (0.0092) Particip. in extracurricular activities 0.0247∗∗∗ 0.0491∗∗∗ (0.0038) (0.0072) Observations (individuals ×choice set) 4,154,485 3,781,446 Pseudo-R20.121 0.149 Notes: This table displays families’ preferences following Section 4. Standard errors are displayed in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 49 Table A8: Families’ preferences dropping first ranked school (cont. from Table 7) Full list minus the 1st preference (1) (2) All municipalities Urban municipalities Distance 0.0012∗∗∗ 0.0014∗∗∗ (0.0001) (0.0001) N of enrolled students 0.0004∗∗∗ 0.0004∗∗∗ (0.0000) (0.0000) N of teaching staff N of enrolled students -14.3754∗∗∗ -13.2550∗∗∗ (0.2259) (0.2416) N of non-teaching staff N of enrolled students -9.7500∗∗∗ -10.7150∗∗∗ (0.5072) (0.5435) N of Sport infrastructure 0.0167∗∗∗ 0.0231∗∗∗ (0.0049) (0.0050) N of Art extacurr. offered 0.0320∗∗∗ 0.0320∗∗∗ (0.0017) (0.0017) N of classic sports offered 0.0115∗∗∗ 0.0149∗∗∗ (0.0022) (0.0023) N of niche sports offered 0.0636∗∗∗ 0.0646∗∗∗ (0.0027) (0.0028) Public 0.2039∗∗∗ 0.2142∗∗∗ (0.0092) (0.0095) PIE 0.1441∗∗∗ 0.1372∗∗∗ (0.0068) (0.0070) Foreign name 0.0365∗∗∗ 0.0192∗∗ (0.0094) (0.0096) Religious -0.0154∗∗∗ -0.0156∗∗∗ (0.0057) (0.0059) Any monthly fee 0.0914∗∗∗ 0.1001∗∗∗ (0.0084) (0.0085) Had a selective admission 0.2115∗∗∗ 0.2039∗∗∗ (0.0068) (0.0069) Reading test scores 0.0036 0.0117 (0.0077) (0.0081) Math test scores 0.2157∗∗∗ 0.2195∗∗∗ (0.0070) (0.0073) Med-low SES school 0.6571∗∗∗ 0.6183∗∗∗ (0.0195) (0.0228) Medium SES school 1.2701∗∗∗ 1.2296∗∗∗ (0.0213) (0.0243) Med-high SES school 1.5504∗∗∗ 1.5171∗∗∗ (0.0237) (0.0268) High SES school 1.4734∗∗∗ 1.4140∗∗∗ (0.0303) (0.0328) Parental percep. of sch. violence -0.0918∗∗∗ -0.0946∗∗∗ (0.0049) (0.0051) Exclusion 0.4080∗∗∗ 0.4406∗∗∗ (0.0300) (0.0316) Parental college expectations 1.6790∗∗∗ 1.7011∗∗∗ (0.0404) (0.0434) Self-efficacy -0.0460∗∗∗ -0.0581∗∗∗ (0.0055) (0.0058) Particip. in extracurricular activities -0.0043 0.0112∗∗ (0.0043) (0.0045) Observations (individuals ×choice set) 4,263,879 4,046,135 Pseudo-R20.131 0.123 Notes: This table displays families’ preferences following Section 4. Standard errors are displayed in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 50