Policy Research Working Paper 10720 Right to Education Forced Migration and Child Education Outcomes Juan F. Vargas Sandra V. Rozo Development Economics Development Research Group March 2024 Policy Research Working Paper 10720 Abstract About a third of the 7.7 million Venezuelans who have on school enrollment, dropout/promotion rates and stan- left their country due to political and economic turmoil dardized test scores. The identification relies on a plausibly have settled in neighboring Colombia. The extent to which exogenous measure of the predicted migration shock faced by the Colombian schooling system can absorb the massive each Colombian municipality every year. The findings show demand for education of Venezuelan children is key for that the migration shock increased the enrollment of Vene- their future trajectory of human capital accumulation, as zuelan students in both public and private schools and in all well as that of Colombian students in receiving communi- school grades, but also generated negative spillovers related ties. This paper estimates the effect of Venezuelan migration to failing promotion rates and increasing dropout. This on educational outcomes of children living in settlement paper documents that these negative effects are explained municipalities in Colombia, distinguish between the effect by the differential enrollment capacity of schools, as well of the migration shock on native and migrant students. as by the deterioration of key school inputs. Specifically, it estimates the effect of the migration shock This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at sandrarozo@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team RIGHT TO EDUCATION: FORCED MIGRATION AND CHILD EDUCATION OUTCOMES JUAN F. VARGAS¶ AND SANDRA V. ROZOΠ JEL Codes: F22, I25 Keywords: Migration, Education, Colombia, Venezuela We thank Juan Blyde, Mat´ ıas Busso, Ana Mar´ an ıa Ib´ ˜ez and Anna Maria Mayda for their advice and thorough discussion. We also thank Dante Contreras, M´ ıa, Marisol Rodriguez and onica Garc´ seminar participants at the Inter-American Development Bank’s Workshop Understanding the Impacts of Migration in Latin America and the Caribbean for helpful comments. Esteban Aleman and Nathalie Basto provided excellent research assistance. We gratefully acknowledge the financial support the of Inter-American Development Bank, and thank the Colombian Ministry of Education for kindly sharing their data. ¶ University of Turin (ESOMAS) and Collegio Carlo Alberto. E-mail: juan.vargas@unito.it. Π Research Group, World Bank. E-mail: sandrarozo@worldbank.org. 1 1. Introduction Over the past ten tears, economic and political turmoil in Venezuela, as well as a surge in criminal violence, have induced large migration flows of Venezuelans. By September 2023, about a third of the 7.7 million Venezuelans who had fled the country because of the crisis had registered in neighboring Colombia. Of these, more than 1.8 mil- lion had obtained a residence permit.1 About 40% of Venezuelans in Colombia are minors.2 Since forced migration most likely interrupted their education process, the implied surge in the demand for schooling constitutes an important challenge for the Colombian education system. Indeed, the extent to which local schools can absorb and accommodate Venezuelan children will determine the accumulation of skills and human capital of the younger generations of both migrants and native kids in host communities. In this paper, we estimate the impact of the Venezuelan migration shock on schooling outcomes in receiving Colombian municipalities. In particular, we use administrative data to study the effect of the Venezuelan migration on school enrollment, dropout rates, promotion rates, and test scores. We use individual-level administrative data to distinguish between native and migrant children, as well as in terms of gender and school characteristics such as their public or private ownership. Our empirical strategy exploits two sources of exogenous variation in cumulative Venezue- lan migration inflows at the municipality-year level. Cross-sectional municipal variation comes from the share of population within each receiving municipality that was born Venezuela and arrived in 1993, well before the start of the political and economic cri- sis of that country. Annual variation comes from the number of Venezuelans arriving to Colombia each year, as the crises worsened during the late Chavez and under the 1 Source: United Nations’ Regional Interagency Coordination Platform. 2 See https://migravenezuela.com/informes/caracterizacion-de-la-ninez-y-adolescencia- migrante-en-colombia/(lastaccessed15/01/2024). 2 Maduro administration. The interaction of these two sources of variation corresponds to a plausibly exogenous predicted migrant shock measure.3 Our identification strategy exploits the fact that crisis-induced migrants tend to move disproportionately to municipalities where they have preexisting networks, formed be- fore the beginning of the migration wave (Bartel, 1989; Greenwood and McDowell, 1986). Our identification assumption is that predicted migration shock is correlated with actual migration but do not affect educational outcomes independently or through any other channel. This is plausible after controlling for municipality-specific trends parametrized by key pre-determined municipality characteristics, which we choose us- ing machine learning techniques. We find that the migration shock increased the enrollment of both male and female foreign students. This effect is mainly driven by public schools, and it is stronger for younger kids, who enroll in primary school grades. We also find, however, negative ex- ternalities of the increased enrollment in terms of promotion rates and dropout rates. While falling promotion rates and increasing dropout occurs for both native and for- eign students, the effects are larger for the former. Migrants, on the other hand, seem to exert more effort as suggested by their performance in standardized tests. Finally, we document that the mechanisms that explain these negative effects of the migration shock have to do with the deterioration of key school inputs in public schools. For in- stance, we show that the migration shock reduces the number of teachers and increases the ratio of pupils to teachers (and thus class size). Our results highlight important policy implications, which we discuss in the conclusion. 3 Importantly, given the absence of longitudinal data on the places where Venezuelan migrants have settled over time, we use our predicted migration measure in a reduced-form specification instead of as an instrument for the endogenous actual settlement of Venezuelans. However, a visual comparison of Figures 2B. and 2D. suggests that the spatial distribution of the census-observed and the predicted number of cumulative Venezuelans in 2018 is very similar. Indeed, the correlation is 0.67. 3 Our paper contributes to the recent strand of the literature that studies the effect of ex- posure to refugees on educational outcomes of native children. Interestingly, most such papers find that the interaction with refugees does not affect the outcomes of native students. For example, van der Werf (2019) finds that the inflow of Indochinese refugees in the U.S. at the end of the Vietnam War did not affect native children’s academic ¨ achievement. In a similar fashion, Figlio and Ozek (2017) find no effects on native students Florida public schools of Haitian migrants who fled after the 2010. Assaad et al. (2019) find not effect of Syrian refugees on educational outcomes of Jordanian children. Ohinata and Van Ours (2013) also find not effect of having refugees in the classroom on outcomes of Dutch students.4 Our paper also contributes to an extensive literature that studies how the characteristics of peers affect students’ outcomes. Most of these papers find that that low-ability and disruptive peers (e.g. exposed to native violence) have negative impacts in student achievement (Lavy et al., 2012; Figlio, 2007; ¨ Gould et al., 2009; Figlio and Ozek, 2017; Carrell and Hoekstra, 2010). 2. Context 2.1. The Venezuelan crisis. The beginning of the Venezuelan political crisis can be avez as president on December 6, 1998. Chavez’ traced back to the election of Hugo Ch´ socialist regime was characterized by constitutional amendments, land expropriations, the implementation of populist social programs, nationalizations, and restrictions on ´ private businesses (Crasto and Alvarez, 2017). These policies were continued –and in as Maduro, who was elected president of Venezuela some cases strengthened- by Nicol´ 4 Other studies do find non-zero effects. Brunello and Rocco (2013) find that native kids obtain lower test scores in a cross-country setting and Jensen and Rasmussen (2011) confirm these findings for the case of Denmark. While Gould et al. (2009) finds that native children face a lower probability of finishing high-school in the long run in Israel, Hunt (2017) finds the opposite for native black students in the U.S. 4 in 2013. Since then, shortages of food and basic necessities became common, and looting began to occur systematically throughout the country (Revista Semana, 2017). Moreover, insecurity became endemic, repression of the opposition became common, and systematic human rights violations by public authorities were repeatedly reported by the international media (see El Nuevo Herald, 2014; BBC News, 2017; BBC, 2016). External factors such as plummeting oil prices since 2014 and international sanctions on Venezuela, have exacerbated the crisis.5 This situation triggered large waves of out-migration by Venezuelans, who most often moved to neighboring Colombia. According to the official statistics, since that start of the humanitarian crisis caused by Chavez’ and Maduro’s regimes, about 7.7 million Venezuelans have left their country. Of these, about 2.5 million have settled in Colom- bia. Initially, Venezuelan migrants consisted mainly of wealthy Venezuelans and entrepreneurs who came to invest in Colombia and fled to save their capital from expropriations and ´ from high inflation (Revista Semana, 2017; Crasto and Alvarez, 2017). As the crisis intensified, however, the core of Venezuelan migration shifted to the less educated pop- ulation, who report fleeing to escape violent crime, political repression, and to look for basic necessities for survival (NPR, 2018). Indeed, according to recent characteriza- tions of Venezuelan migrants based on the Colombian household surveys of 2015 and 2016, over 80% of registered migrants have not completed a high school education, at least half are 25 years old or less, and they are balanced in terms of gender (see OLR, 2017). 5 Following the repression of protests by the Maduro administration, several countries led by the U.S. and the European Union established in 2017 sanctions to individuals associated with Maduro, including politicians, military personnel and private citizens. The sanctions were soon extended to include private companies and, in 2019, entire industries associated with mining and banking activities. The United Nations High Commissioner for Human Rights has warned that sanctions could have worsened the precarious situation of Venezuelans. 5 2.2. Colombia’s education system and policy response. The education system in Colombia comprises one year of preschool, five years of primary education, four years of lower secondary education and two years of upper secondary education. In 2014, 87% of the schools in Colombia were public and out of those, 78% were located in rural areas (OECD, 2016). All children between five and fifteen years old are legally required to attend preschool plus nine years of compulsory basic schooling. However, it is estimated that 20% of the students do not continue studying beyond primary school (OECD, 2016), and only 65% of boys and 77% of girls complete lower secondary education (Radinger et al., 2018). The Colombian government has implemented several initiatives to facilitate the inte- gration of Venezuelan children in public schools. For example, Decree 1288 of 2018 simplified the process for Venezuelans to validate their educational through standard- ized tests. This policy was designed with the objective of allowing migrant children to enroll in school grades according to both their age and prior academic achievement. In addition, also since 2018, the Colombian government allowed Venezuelan children to attend public schools regardless of the immigration status of their households. Unfortunately, before 2018 (the end of our sample period, limited by data availabil- ity) the Colombian government did not keep systematic track of school enrollment by student nationality. However, our administrative individual-level dataset does distin- guish between Colombian and foreign students, and crucially, according to the 2018 population census almost 9 out of 10 settled foreigners in Colombia are Venezuelans.6 6 If we compute this share at the municipality level the average is 0.74 and the standard deviation 0.29. This suggests that the biggest cities in population, such as Bogota, have a somewhat smaller share. 6 3. Data 3.1. Outcomes. To measure our outcome variables, we will combine two adminis- trative datasets. First, we will compute school-level enrollment, dropout rates and promotion rates using the administrative registry of all the students in Colombia, en- rolled in either public or private schools. This dataset is called R166-SIMAT and its source is the Colombian Ministry of Education.7 It is available for the period 2012- 2018, which shapes our sample period. Importantly, R166-SIMAT includes the student ID that allows us to distinguish be- tween Colombian and foreign students in order to explore the effects of the Venezuelan migration shock on both native and migrant students. Specifically, we identified as ‘migrant’ those students with IDs different to the standard ID that the government issues to underage natives. These include special residence permits, visas, and border mobility cards. We also classified as migrant students the individuals who enrolled using a provisional ID, provided by the municipal Secretary of Education to undoc- umented children who want to enroll in a public institution. The vast majority of undocumented children are foreigners, most of whom are Venezuelans (see section 2. Using R166-SIMAT we can construct the following school (or school/grade)-level vari- ables: 1. (Log) Enrollment: the (log of the) total number of students enrolled per school (or school/grade) at the beginning of each academic year. 7 The name of the dataset originated in the Ministry’s Resolution 166 of 2004, which created the National Enrollment System (SIMAT, from the Spanish acronym) that mandated all education in- stitutions to report to the Ministry individual-level enrollment each year as well as the condition of each student at the end of the academic year. Importantly, this registry excludes schools that have nontraditional education models. This is the case of some indigenous communities in rural areas. It also excludes public institutions for adult education and literacy,and training colleges. 7 At the end of the academic year, total enrollment is broken into four categories: stu- dents who transferred to a different school during the academic year, student who dropped out from the school (and did not transfer to any school during that year), stu- dent who were promoted to the next grade (or graduated from school), and students who failed the grade. Using this break-up, we can compute the following additional outcomes: 2. Dropout rate: the share of students who dropped out from each school during the academic year over the initial school enrollment for that year. Note that this corresponds to the intra-annual dropout rate as it measures the proportion of students that leave the school during an academic year. Also importantly, this is a true measure of dropout, as we net out the students who changed schools during the academic year. Formally, for grade g of school s and year t, we compute the dropout rate (DR) as: 11 g =1 dgst DRst = 11 g =1 egst where dgst is the number of dropouts from grade g of school s and year t and egst is enrollment at the beginning of that grade/school/year (net of school switch- ers). 3. Promotion rate: the share of students who were promoted to the next grade relative to the initial enrollment (net of school switchers). Formally, we compute the promotion rate P R as: 11 g =1 pgst P Rst = 11 g =1 egst 8 where egst is defined as above and pgst is the number of students promoted in that grade/school/year. R166-SIMAT also includes a registry of public school teachers for the sub-period 2014- 2018. We use these data to explore potential mechanisms (albeit for public schools only) related to key supply measures such as the teacher/students ratio and the quality of teachers. The second administrative dataset contains information on the scores obtained by students in the official high-school exit exam (called Saber 11) for the period 2005– 2018. It comes from the Colombian Institute for the Promotion of Higher Education. From it, we computed math and language test scores. To facilitate the interpretation and make scores comparable across years, we standardized the test scores to have mean zero and standard deviation one each year. Table 1 reports descriptive statistics of the main outcome variables. 3.2. Other data. We will also employ data on the total number of Venezuelans ar- on riving annually in Colombia, available from the national migration agency (Migraci´ Colombia). These data come from the information recorded at official migration points and thus, it does not include any illegal or unregistered migration. footnoteThis imply that the 2.5 million figure of Venezuelans in Colombia is likely un- derrepresented. Finally, we have access to a large number of pre-determined municipal level controls (based on the 1993 and 2005 population censuses) that we use to control for differential trends interacted by time-invariant municipal characteristics that may help predict the evolution of educational outcomes. Table 2 reports the descriptive statistics on these covariates. 9 4. Empirical Strategy As forced migrants do not choose their arrival municipalities randomly, we cannot use a mean comparison to identify their effects on educational outcomes in host mu- nicipalities. Such a comparison would likely be biased. For example, if migrants go disproportionally to more prosperous municipalities and prosperity is positively associ- ated with better school performance, such the bias would be positive. More generally, it is reasonable to assume that the decision of where to locate is associated with mu- nicipal characteristics that, in turn, are correlated with the quality of education. Our empirical strategy, consequently, exploits the fact that, as the political crises inten- sified in Venezuela, migrants tend to move disproportionately to municipalities where they have networks, family, or acquaintances. This has been shown to be the case in a variety of contexts.8 In particular, we estimate the following augmented specification, which distinguishes the effects of the migration shock across nationality (native versus migrant students) and gender (females versus males). (4.1) Ysmdt = λs + γd×t + θ1 Pr.Ven.Shock1993 mdt + θ2 Nativesmdt + θ3 Femalesmdt + θ4 [Pr.Ven.Shock1993 1993 mdt × Nativesmdt ] + θ5 [Pr.Ven.Shockmdt × Femalesmdt ]+ θ6 [Pr.Ven.Shock1993 mdt × Nativesmdt × Femalesmdt ] + θ7 [Nativesmdt × Femalesmdt ]+ δ (c × φt ) + smdt c∈Xmd where Ysmdt is any of several educational outcomes in school s from municipality m of department d and year t; λs and γd×t are respectively school fixed effects and department×year fixed effects. These control, respectively, for any time-invariant 8See for example Bartel (1989) and Greenwood and McDowell (1986). Rozo and Vargas (2021) use a similar empirical strategy to estimate the effect of the Venezuelan migration shock on electoral outcomes in Colombia. 10 school-level heterogeneity that may be correlated with educational outcomes and for any aggregate shock that may affect in the same way all the municipalities of the same department. Pr .Ven .Shock 1993 mdt is our predicted cumulative migration inflow to munic- ipality m of department d and year t, based on the 1993 census (before Hugo CHavez was elected as president in Venezuela).9 Nativesmdt (Femalesmdt ) identifies the sub- group of Colombian (Female) students in school s. Xmd is a vector of pre-determined municipality-specific characteristics which we interact with the year fixed effects rep- resented by φt . This interaction effectively controls for municipal-specific changes over time, parametrized by the set of control included in Xmd . It is worth noting that the municipality characteristics included in this set are not chosen in an ad hoc way. Rather, following Belloni et al. (2014), the controls are selected using machine learning techniques. In this way we are agnostic about which municipality characteristics are more related to educational outcomes in areas that have hosted Venezuelan migrants. Finally, the error term, smdt , is estimated allowing for serial correlation within munic- ipalities. Our measure of the predicted migration shock follows the standard practice in the literature (see Card, 2001 and Altonji and Card, 1991 for the pioneer approaches and Lewis and Peri, 2015 for a review of the literature on applications) and exploits the dis- proportionate levels of cumulative migrant inflows to areas with previous settlements of similar identity groups. Specifically, our measure is constructed as: (4.2) 1 Pr.Ven.Shock1993 mdt = 1993 1993 Tot. Ven. Inflowt × Venezuelan Sharemd ×100 Populationmd 9The distributions of Venezuelans across Colombian municipalities according to the 1993 census is reported in Figure 2A.. 11 where Tot. Ven. Inflowt is the aggregate number of (legal) Venezuelans entering Colom- bia every year (as recorded by migration authorities), Population1993 md is the total pop- ulation of municipality m in 1993, which is kept fixed prior to the migration shock to avoid further endogeneity concerns. Venezuelan Share1993 md is the share of Venezuelans living in municipality m according to the 1993 population census to the total number of Venezuelans living in Colombia.10 Venezuelan Pop1993 Venezuelan Share1993 md = md 1993 m Venezuelan Pop md For robustness, we aggregate our outcome variables at the municipality level and re- estimate equation 4.1 changing the school fixed effects for municipality fixed effects. 4.1. Challenges to identification. Our identification relies on the plausibly exo- geneity of our predicted migration shock. We now discuss potential threats to this assumption, and how we have dealt with them. First, it is worth noting that we are not using the predicted cumulative inflows as an instrument of the actual inflows of Venezuelans to each municipality/year. This is because there are no administrative records of where do the arriving Venezuelan na- tionals settle on a yearly basis. We therefore use the predicted shock to estimate a reduced-form equation. The magnitude of our results should therefore be interpreted as only capturing the numerator of a standard 2SLS estimator (and thus likely as lower bounds of the latent IV estimate). The first assumption that we rely on for identification is, therefore, that the predicted migration is strongly correlated with the (unobserved) actual Venezuelan migration. 10We use the 1993 census because by the time of the next census (2005) the Venezuelan political crises -and thus migration to Colombia- had started under the rule of Hugo Chavez. Our results are however largely unchanged if we use the 2005 census to compute the predicted migration flow of each municipality. Indeed, the correlation of the 1993 and the 2005-based measures if 0.93. 12 We can test this assumption for 2018, when the last population census took place in Colombia. Figures 2B. and 2D. show the municipal distribution of, respectively, the observed number of the Venezuelans and the predicted cumulative figure for 2018 (the predicted 2018 inflow is reported in Figure 2C.). Reassuringly, not only the distribu- tion looks very similar, but also their correlation is 0.67. We are thus confident that our predicted migration shock has predictive power.11 Secondly, the 1993-based predicted inflow measure needs not to be correlated with con- temporaneous schooling outcomes through any channel different than actual Venezue- lan migration. Regarding this assumption, it is worth noting that, because our esti- mates include fixed effects by municipality as well as by department×year, they are confounded neither by time-invariant differences across municipalities nor by annual aggregate department-level shocks.12 This is, however, not enough to achieve identifica- tion. It may well be the case that pre-shock migrants disproportionally settle in places with characteristics that explain future educational outcomes. Indeed, as noted by Goldsmith-Pinkham et al. (2019), identification in the Bartik/Shift-Share-type instru- ments comes mainly from the cross-sectional (“share”) variation, so it is important to check the extent to which the initial shares (of migrants) are correlated with potential confounders prior to the current migration wave. To this end, following Belloni et al. (2014), we use machine learning to select the most robust determinants of Venezuelan settlements according to the 1993 census and include in our main specification the interaction between each of these and the year fixed effects. By doing so, we flexibly control for municipal-specific trends, parametrized by a large set of pre-determined characteristics that predict early settlements.13 11In other words if we were to rely on a 2SLS strategy, the first stage would most likely be strong and the instrument relevant. 12The just over 1,100 Colombian municipalities are distributed across 32 departments. 13Even when the share of early migrants is not exogenous, Borusyak et al. (2018) show that identi- fication can be achieved if the aggregate shocks are as good as random, a condition that is satisfied when: i) one controls by observable municipal characteristics weighted by shock exposure and ii) there 13 One additional recent criticism to the validity of using early migrants networks to study the impacts of migration is that in settings in which migration is serially correlated, past migration causes both current outcomes and current migration, and thus, the short and long run effects of migration are confounded (Jaeger et al., 2018). Our empiri- cal strategy is not sensitive to this threat because the inflows crisis-driven Venezuelan migrants are not stable in time, they are sudden and large in scale as a consequence of the intensification of the internal and the Venezuelan crises. Moreover, Venezuelan migration to Colombia prior to the Venezuelan crisis was negligible. Overall, we are confident that our estimates can be interpreted identifying the causal (potentially lower bound) effect of the Venezuelan migration shock on educational out- comes. 5. Results 5.1. Main results. We start by studying the effect of the Venezuelan migration shock on educational outcomes aggregated at the school level and averaging across all schools (Table 3). We then separate the result across public and private schools (Tables 4 and 5 respectively); between primary and secondary school grades (Tables 6 and 7 respec- tively); and across schools situated in relatively more urban or rural municipalities (Tables 8 and 9 respectively). Panel A in each table includes no controls, and Panel B includes the set of controls optimally selected by the machine learning algorithm proposed by Belloni et al. (2014). All the results are robust to the inclusion of the controls, both in magnitude and in terms of statistical significance. To facilitate the interpretation of the findings, all these tables report the marginal is a large number of observed shocks per period and a large number of periods. We meet both these criteria. 14 effects of the migration shock on each type of student (across gender and national- ity). Because our main specification interacts the predicted migration shock with the school-level subgroup of native student and females (and it is saturated with all the underlying double interactions), interpreting the regression output is time-consuming. For reference, in the Appendix we report Tables A.1 to A.7, which are the regression output counterpart of Tables 3 to 9. In those tables, the coefficient associated with the non-interacted migration shock (θ1 in equation 4.1) is the effect of the shock for foreign males. The effect of the shock on foreign females is the sum of the former and the co- efficient associated to the interaction between the shock and the Female indicator (θ5 ). The effect on native males is the sum of the coefficient associated with the shock (θ1 ) and that of its interaction with the Native indicator (θ4 ). Finally the effect on native females is the sum of the coefficient associated with the shock and those associated with the two double interactions (θ4 and θ5 ) and that of the triple interaction (θ6 ). We compute these sums (and their corresponding standard errors), and for simplicity only refer to the marginal effects henceforth. Starting with the effect of the Venezuelan migration shock across the aggregation of all school types (Table 3), we find very intuitive results for the case of enrollment. On average, the shock increased the school enrollment of both migrant men and women, but not of natives (Column 1). Focusing on Panel B, which includes the optimal set of controls interacted with the year fixed effects, we find that a one-standard-deviation increase in the predicted migration shock (= 22.89, see Panel B of Table 1) increases the enrollment of migrant male students by 9.38 students (= 22.89 × 0.0041 × 100), and that of migrant female students by 8.47 students.14 These effects are sizeable: they represent about 4% of total average enrollment (see Table 1). The estimated coeffi- cients of the marginal effects of the migration shock on the enrollment of natives (men 14We multiply the interaction of the coefficient and the standard deviation of the predicted measure by 100 because of the log-level nature of the specification for the case of enrollment. 15 and women) are very close to zero. We also find that the shock decreased promotion rates across the board (Column 2), making it harder for both native and migrant students to advance in their educational cycle. In terms of the economic size of the effect, a one-standard-deviation increase in the predicted migration shock decreases the promotion rate of foreign male students by 0.54% (= 22.89 × 0.0235) and that of native males by 0.5%. These effects are rather small, and represent, respectively 2.4% and 4.7% of the group-specific standard deviation of the promotion rate (reported at the bottom of the Table). The effect of a one-standard-deviation increase in the predicted migration shock on the promotion rate of native female students is a decrease in 0.43% (4.6% of the group standard de- viation). For the case of foreign females, the estimate is however not significant, but the magnitude is 66% of the estimated effect for native females. Note that the magnitude of the effect is larger for native students as compared to migrants. One potential explanation is that migrants who claim to have successfully completed a higher school level before arriving to Colombia –but do not have the documentation to prove it–are allowed to be promoted just by taking a test that is administered by the local Secretary of Education of the municipality where their new school is located (Decree 1288 of 2018). However, as discussed next, the effect of the shock on dropout rates is also larger (and indeed the gap is much wider) for local students, and this could not be accounted by this or any explanation about the insti- tutional environment. The fact that the Venezuelan migration shock decreases promotion rates is consistent with an interpretation of school congestion: if school inputs remain constant in the short run (our sample period cover 7 years), the documented large increase in enroll- ment is likely to harm the learning process of both migrant and native students. This interpretation finds further support in Column 3 of Table 3, where we study the effects 16 of the Venezuelan migration shock on dropout rates. As mentioned in section 3, this outcome already accounts for school switchers, so it should be interpreted a school- system dropout. We find that the Venezuelan migration shock increased dropout rates for all types of students. A one-standard-deviation increase in the predicted migration shock increases the dropout rate of foreign males (females) in 1.6 (1.4)% of the group- specific standard deviation. It also increases the dropout rate of native males (females) in 3.6 (4.2) of the group-specific standard deviation. While it is worrisome that the shock increases dropout rates and that it does so across the board, notice that the magnitude of the effect is substantially larger for native students. This finding is important for the design of school retention policies. Finally, we study the effect of the migration shock on the scores of the standardized end-of-school-test, a proxy of educational quality. We do so for the case of math scores (Column 4) and language scores (Column 5). The migration-driven increased school enrollment does not seem to affect school performance, as all the coefficients are close to zero. The only exception is a positive and significant effect on the performance of migrant male students in the language test. A one-standard-deviation increase in the predicted migration shock increases this outcome in 5% of a standard deviation.15 This may be explain by foreign students exerting higher effort, or by institutional reasons such as the fact that many Venezuelan children arrive in Colombia with completed high school, but repeat the final high school grade and take the state exam so that they can be given a local high school diploma. 5.2. Results by school types. We move to studying migration impacts on educa- tional outcomes in different types of school. First, we distinguish between public and private schools, and report the results in Tables 4 and 5, respectively. In both Tables, the results reported in Column 1 imply that the documented increase 15According to the education literature this is about 25% of a year worth of education. 17 in the enrollment of foreign students of both genders occur both in public and private schools. Anecdotal evidence suggests that many Venezuelans seek private education, even if their economic conditions are precarious. Still, since enrolling in a public school is free, relative to the total average enrollment the effect is twice as large for public schools than for private schools. On the other hand, the documented reduction in promotion rates, and the increase in dropout rates, are both entirely driven by public schools (Column 2 and 3 of both Tables, respectively). The estimated coefficients of the effect of the migration shock on promotion/dropout rates in private schools is not only statistically insignificant, but also much smaller in magnitude relative to the estimates for public schools. This heterogeneity is largely consistent with the different capacity of public and private schools to react to increases in enrollment by providing more or better school inputs, a mechanism that we test formally in the next section. The effects of the migration shock on school performance of high school seniors is en- couraging, and it is concentrated on foreign students. In public schools, the migration shock increases school performance of both foreign females (in math and language) and foreign males (in language only), but has no effect on natives (Columns 4 and 5 of Table 4). In private schools, only foreign males benefit from the Venezuelan migration flows, and they do so both in math and language. The second dimension of heterogeneity is the focus on primary versus secondary school grades. Do the documented effects of the shock depend on the schooling cycle and therefore on age? We explore this on Tables 6 and 7, which report the estimated effects of the shock in primary school grades and in secondary grades respectively.16 Interestingly, the shock-induced enrollment increase of foreign students is present in 16Note that these tables do not report effects on test scores since, for the entire sample period under study (2012-2018), these are available only for high-school seniors (their end-of-school test). 18 both primary and secondary grades, suggesting that migrant kids cover a wide age spectrum (Column 1). This is consistent with the anecdotal evidence discussed in section 2. Moreover, the negative effect of the shock on promotion rates seems to be driven by primary school grades in all sub-groups except that of foreign females, who seem to have their promotion rate affected in secondary school grades (Column 2). Finally, the estimates for dropout rates are rather imprecise, but the coefficients are positive in both cases and larger in magnitude for primary school grades (Column 3). While inconclusive, this evidence is consistent with the migration shock affecting disproportionally primary school grades. The third and final dimension of heterogeneity if whether the school is located in a rel- atively more urban or more rural areas. We implement this classification by identifying schools located in municipalities above and below the median of the ratio of rural to total municipal population. These data come for the Colombian Statistics agency. The descriptive statistics of this variable are reported in Panel B of Table 1. Table 8 reports the results for the subsample of schools located in urban areas, and Table 9 does so for the subsample of rural areas. Once again, it is reassuring to corroborate that larger migration cumulative inflows of Venezuelans increase school enrollment of both migrant men and women in both urban and rural areas (Column 1). Interestingly, however, in rural areas the migration shock seems to have positive spillovers on local children, as the enrollment of native males and females also increases significantly (Table 9). The other outcomes have imprecisely estimated effects, which prevent us from drawing strong conclusions about their heterogeneity across the urban/rural dimension. 5.3. Dynamics. Because of the cumulative nature of the migration shock (see Fig- ure 1), we also study the yearly dynamics of the effect of the year-by-year cumulative migration inflows coming from Venezuela on the outcomes of interest. Figures 3 to 7 19 report the marginal effects of an event-study specification that interacts all the compo- nents of equation 4.1 with year dummies, together with their 95% confidence interval.17 As usual in this type of specifications, standard errors are somewhat large. However, interesting patterns do emerge. For instance, Figure 3 shows that the effect of the mi- gration shock on the enrollment of both foreign men and women has virtually the same magnitude from 2014 to 2018, and it is consistently larger than the effect for natives (and always significantly different from zero). Moreover, the enrollment of both native males and females does increase with the migration shock, but only in the middle of the sample period, from 2014 to 2017. In fact, it follows an inverse U pattern. Figure 4 corroborates the aforementioned findings for the case of promotion rates, and further highlights that promotion rates decrease with the shock for all students, but es- pecially natives, almost every year of the sample period and following a U-pattern that peaks in 2015. In addition, Figure 5 suggests that the discussed increase in dropout rates –for native students only–following the migration shock, is driven by its behavior in 2013 to 2015. The fact that promotion drops and dropout gains on native students are smaller in magnitude (and mostly statistically insignificant) in the latter sample period is en- couraging, as it suggests that schools managed to adjust to the challenges posed by the enrollment boost, and that they actually did so during the period that faced the largest migration flows. Importantly, such adjustment allowed schools to reduce the negative spillovers of the increased enrollment. Finally, as reported in Figures 6 and 7, the shock had a very precisely measured null effect on test scores of native students during every one year of the sample period. Instead, for the case of foreigners the effect was rather volatile, but migrant males 17Because they plot the marginal effects for each subgroup of interest, the figures show no omitted year of reference. 20 seem to have benefited the most from the migration shock in terms of school perfor- mance, especially in language tests (Figure 7).The estimates of both outcomes for all the subgroups stabilize at zero starting in 2017 (2016 for the case of the math test). 6. Potential mechanisms We find that the massive recent migration of Venezuelan children to Colombia in- creased school enrollment but also decreased promotion rates and increased dropout rates, especially so for native children in public schools. These findings are consistent with an interpretation in which exogenous enrollment surges generate school conges- tion and harm the learning environment of both native and foreign students. However, perhaps migrant students exert more effort to offset this threat, as suggested by their performance in standardized tests. In this section, we test this idea in several different, but complementary, ways. First, we explore the extent to which the observed increases in the enrollment of mi- grant students differentially affect the outcomes that we study in this paper. To that end, we compute the school-specific capacity of absorbing new students. Unfortunately, however, there is no data on school-level vacancies. Moreover, by law, public schools need to accommodate any new enrollment demand. We therefore compute an ‘en- rollment capacity gap’ measure that is specific to 2016. This is the last sample year prior to the largest Venezuelan inflow shock (see Figure 1). In order to compute each school’s enrollment gap, we calculate the largest historical observed enrollment of the school, and subtract from it the 2016 enrollment. Finally, we run our main specifica- tion (equation 4.1) in the subsample of schools located each of the four enrollment gap quartiles. In other words, we explore the heterogeneous effects of our main results by the extent to which a school can absorb new students, as suggested by the size of their historical enrollment. 21 The results are reported on Table 10. There, it is very salient that the aforementioned decrease in promotion rates and increase in dropout rates is, by and large, explained by schools located in the first quartile of the enrollment capacity gap (Panel D) and to a lesser extent (in terms of the magnitude of the coefficients) by the schools located in the second quartile (Panel C). In other words, the schools at the bottom half of the enrollment capacity gap (i.e. those that have less ability to absorb new students just prior to the large migration shock) are the ones negatively affected by the exogenous shift in the demand for school places. We also examine the effect of the migration shock on key school inputs, specifically associated with the number and quality of teachers and class size. To that end we estimate the following simpler version of equation 4.1: (6.1) Ysmdt = θ[Pr.Ven.Shock1993 mdt ] + λs + γd×t + δ (c × φt ) + smdt c∈Xmd where Ysmdt is either the (log of) the number of teachers in school s or the pupil-to- teacher ratio (class size) It is also worth noting that, unfortunately, these data are only available for public schools and from 2014 onwards. The results are reported in Table 11, which looks at the effect on all (public) schools (Panel A), primary schools (Panel B), secondary schools (Panel C), schools located in urban areas (Panel D) and those located in rural areas (Panel E). We find that the migration shock reduces the number of school teachers and increases class size (the ratio of students to teachers). These are important findings as they suggest that (public) school inputs deteriorate with the migration shock, which may explain at least in part why both foreign and native students are promoted less and also dropout more. 22 7. Conclusion For several years, Venezuela has faced a humanitarian crisis generated by economic and political turmoil. Public safety has deteriorated and the access to basic supplies and medications is largely restricted. This situation has pushed almost 8 million Venezue- lans to leave their country, and most of them have settled across South America. By late 2023, the country that had received the vast majority of Venezuelan migrants is Colombia, where the official statistics point to 2.5 million migrants, over 5% of the its own native population. This constitutes a shock of unprecedented magnitude, that has affected most economic and social outcomes in receiving municipalities. The extent to which the labor market, the housing market, the health, and the ed- ucation systems can absorb such a large shock without causing large externalities to local communities (thus generating backlash and public outcry) depends on the policy response of the national and the local governments. These will shape the sectoral and geographical adaptation to the migration wave, and thus, the capacity of local com- munities to offset potential negative externalities and boost the positive spillovers. We find that plausibly exogenous predicted cumulative Venezuelan migration inflows have large effects on the enrollment of migrants, and even positive enrollment spillovers for local children in rural areas. However, we also find that, perhaps because the enroll- ment surge and the consequential congestion of resources (including but not limited to teachers) the migration shock negatively affects school promotion and increases dropout rates. While this occurs for both migrant and local students, these negative spillovers are substantially larger for natives. One potential reason is that migrant students ex- ert more effort at school. This is consistent with out findings that, after the migration shock, both foreign men and women perform better in the national end-of-school exam, but the same is not true for natives. 23 Importantly, we also find that while the enrollment boost occurs in all types of schools (public and private, urban and rural, primary and secondary), the negative spillovers are mainly taking place in public schools, and especially in primary school grades. This is likely driven by the differential capacity of public and private schools to react to the demand shock by increasing key inputs such as the number of teachers. Indeed, we find that the migration shock reduced the number of teachers in public schools, while at the same time increased class-size in detriment of more targeted learning experiences for both migrant and native students. The lack of short-term investments and response of public schools are perhaps what make these results contrast to a large literature that has found no effects of migration flows on the schooling outcomes of native students.18 Understanding the effect of the recent surge in migration flows from Venezuela on selected outcomes such as those related with the capacity of children to accumulate human capital, and understanding the potential mitigating effects of different policy responses is of foremost policy importance. This will help achieve a smooth and bene- ficial absorption of the Venezuelan community into Colombia. This paper contributes to this policy agenda, perhaps the most important that Colombia will face in the next decade, as the crisis in Venezuela intensifies. 18We discuss such papers in the introduction. 24 References Altonji, J. G. and D. Card (1991): “The effects of immigration on the labor market outcomes of less-skilled natives,” in Immigration, Trade, and the Labor Market, University of Chicago Press, 201–234. Assaad, R. A., T. Ginn, and M. Saleh (2019): “Impact of Syrian Refugees on Education Out- comes in Jordan,” . Bartel, A. P. (1989): “Where Do the New U.S. Immigrants Live?” Journal of Labor Economics, 7, 371–91. BBC (2016): ““Una pena de muerte disimulada”: emica Operaci´ la pol´ on on de Liberaci´ del Pueblo, la mano dura del gobierno de Venezuela contra el crimen,” Accessed from: http://www.bbc.com/mundo/noticias-america-latina-38126651. BBC News (2017): “Venezuela profile - Timeline,” Accessed from: http://www.bbc.com/news/world-latin-america-19652436. Belloni, A., V. Chernozhukov, and C. Hansen (2014): “High-dimensional methods and infer- ence on structural and treatment effects,” Journal of Economic Perspectives, 28, 29–50. Borusyak, K., P. Hull, and X. Jaravel (2018): “Quasi-Experimental Shift-Share Research Designs,” NBER Working Paper N. 24997. Brunello, G. and L. Rocco (2013): “The effect of immigration on the school performance of natives: Cross country evidence using PISA test scores,” Economics of Education Review, 32, 234–246. Card, D. (2001): “Immigrant Inflows, Native Outflows, and the Local Labor Market Impacts of Higher Immigration,” Journal of Labor Economics, 19, 22–64. Carrell, S. and M. L. Hoekstra (2010): “Externalities in the classroom: How children exposed to domestic violence affect everyone’s kids,” American Economic Journal: Applied Economics, 2, 211–228. ´ Crasto, T. C. and M. R. Alvarez on venezolana: causas, (2017): “Percepciones sobre la migraci´ na como destino, expectativas de retorno,” Migraciones. Publicaci´ Espa˜ on del Instituto Universitario de Estudios sobre Migraciones, 133–163. El Nuevo Herald (2014): “Oleada de venezolanos emigra escapando de la inseguridad y la crisis,” Accessed from: http://www.elnuevoherald.com/noticias/mundo/america-latina/venezuela- es/article2036640.html. 25 ¨ Figlio, D. and U. Ozek (2017): “Unwelcome Guests? The Effects of Refugees on the Educational Outcomes of Incumbent Students,” . Figlio, D. N. (2007): “Boys named Sue: Disruptive children and their peers,” Education, 2, 376–394. ¨ Figlio, D. N. and U. Ozek (2017): “Unwelcome Guests? The Effects of Refugees on the Educational Outcomes of Incumbent Students,” Working Paper 23661, National Bureau of Economic Research. Goldsmith-Pinkham, P., I. Sorkin, and H. Swift (2019): “Bartik Instruments: What, When, Why, and How,” NBER Working Paper N. 24408. Gould, E. D., V. Lavy, and M. Daniele Paserman (2009): “Does immigration affect the long- term educational outcomes of natives? Quasi-experimental evidence,” The Economic Journal, 119, 1243–1269. Greenwood, M. J. and J. McDowell (1986): “The Factor Market Consequences of U.S. Immi- gration,” Journal of Economic Literature, 24, 1738–72. Hunt, J. (2017): “The impact of immigration on the educational attainment of natives,” Journal of Human Resources, 52, 1060–1118. Jaeger, D. A., J. Ruist, and J. Stuhler (2018): “Shift-share instruments and the impact of immigration,” NBER Working Paper N. 24285. Jensen, P. and A. W. Rasmussen (2011): “The effect of immigrant concentration in schools on native and immigrant children’s reading and math skills,” Economics of Education Review, 30, 1503–1515. Lavy, V., M. D. Paserman, and A. Schlosser (2012): “Inside the black box of ability peer effects: Evidence from variation in the proportion of low achievers in the classroom,” The Economic Journal, 122, 208–237. Lewis, E. and G. Peri (2015): “Immigration and the Economy of Cities and Regions,” in Handbook of Regional and Urban Economics, vol. 5, 625–685. NPR (2018): “Venezuela’s Deepening Crisis Triggers Mass Migration Into Colombia,” Accessed from: https://www.npr.org/sections/parallels/2018/02/20/587242391/venezuelas-deepening- crisis-triggers-mass-migration-into-colombia. OECD (2016): Education in Colombia, OECD Publishing. Ohinata, A. and J. C. Van Ours (2013): “How immigrant children affect the academic achievement of native Dutch children,” The Economic Journal, 123, F308–F331. 26 ısticas de los migrantes de Venezuela a Colombia,” Tech. rep., Observatorio OLR (2017): “Caracter´ Laboral de la Universidad del Rosario. Radinger, T., A. Echazarra, G. Guerrero, and J. P. Valenzuela (2018): OECD Reviews of School Resources: Colombia 2018, OECD Publishing. Revista Semana (2017): exodo de los venezolanos a Colombia,” “El triste ´ Accessed from: http://www.semana.com/nacion/articulo/venezolanos-en-colombia-y-colombianos- repatriados/519657. Rozo, S. V. and J. F. Vargas (2021): “Brothers or invaders? How crisis-driven migrants shape voting behavior,” Journal of Development Economics, 150, 102636. van der Werf, C. (2019): “The Impact of Refugees on Native Students? Academic Achievement,” . 27 Figure 1. Evolution of Venezuelan migration to Colombia 28 (A.) Distribution of Venezuelans accord- (B.) Distribution of Venezuelans accord- ing to the 1993 census ing to the 2018 census (C.) Predicted Distribution of Venezue- (D.) Predicted Distribution of Cumula- lan Inflows for 2018 tive Venezuelan Inflows for 2018 29 Figure 3. Marginal yearly effect on enrollment 30 Figure 4. Marginal yearly effect on promotion rate 31 Figure 5. Marginal yearly effect on dropout rate 32 Figure 6. Marginal yearly effect on math test scores 33 Figure 7. Marginal yearly effect on language test scores 34 Table 1. Descriptive statistics: main variables Variable Obs Mean Std. Dev. Min Max Panel A: School Level Enrollment 119,061 470.661 651.445 1.000 10320.000 Promotion Rate 119,061 94.290 10.241 0.000 100.000 Dropout Rate 119,061 3.214 8.874 0.000 100.000 Saber 11 Math STD. Scores 60,348 0.097 1.003 -3.946 7.803 Saber 11 Language STD. Scores 60,348 0.099 0.997 -4.027 5.013 Public Schools 119,061 0.602 0.489 0.000 1.000 N. of Teachers 44,890 32.852 32.940 1.000 347.000 Pupil/Teachers Ratio 44,890 21.260 39.266 0.120 2332.000 Temporal Teachers (%) 44,890 21.787 25.812 0.000 100.000 Capacity Gap at 2016 16769 60.644 130.314 0.000 2875.000 Panel B: Municipality Level Predicted Cum. Venezuelans Inflow 6,874 6.746 22.891 0 643.313 Indicator for Urban Municipality 6,874 0.494 0.500 0 1.000 35 Table 2. Descriptive Statistics: pre-determined municipal controls Variable Year Obs. Mean Standard Deviation Category Per capita GDP (Millions) 2005 1,097 6.381 6.632 Economic Growth Night Light Density 1995 1,048 3.968 7.466 Economic Growth GINI 1993 1,043 0.456 0.0378 Poverty and Inequality Subsidized Health System Cov. (%Pop.with UBN) 1998 1,136 0.716 0.411 Poverty and Inequality Unsatisfied Basic Needs (UBN, % Households ) 1993 1,035 52.98 19.21 Poverty and Inequality Number of Financial Institutions 1995 1,046 1.754 8.922 Institutions Number of Tax Collection Offices 1995 1,046 36.05 182.4 Institutions Informal Labor* (% Household) 2005 1,114 0.949 0.0571 Labor Market Municipal Tax Income (Millions) 1995 1,098 1,033 16,066 Government Finance Mun. Public Expenditure (Thousands) 1995 1,098 2,909 28,866 Government Finance Central Gov.Transfers (Millions) 1995 1,098 1,168 5,348 Government Finance Homicide Rate (per 100,000 Indv.) 1995 1,048 52.92 66.89 Conflict and Violence Hectares of Coca Crops 1999 1,124 142.5 960.2 Conflict and Violence N. of Terrorist Attacks 1993 1,124 0.657 2.628 Conflict and Violence Notes: *Informal Labor is a dummy variable equal to one if less than 100% of the economically active population within a household does not contribute to the pension system. 36 Table 3. Average effect of the Venezuelan migration shock on educa- tional outcomes – All schools (marginal effects) (1) (2) (3) (4) (5) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Math Std. Language Std. Score Score Panel A: Without Controls Foreign Males 0.0043*** -0.0244*** 0.0146** 0.0017 0.0040*** (0.0010) (0.0090) (0.0067) (0.0012) (0.0012) Native Males 0.0005 -0.0228** 0.0146** -0.0000 0.0003 (0.0009) (0.0091) (0.0069) (0.0006) (0.0005) Foreign Females 0.0040*** -0.0135 0.0122* 0.0014 0.0016* (0.0010) (0.0095) (0.0068) (0.0013) (0.0009) Native Females 0.0004 -0.0199** 0.0162** 0.0003 0.0005 (0.0009) (0.0090) (0.0070) (0.0004) (0.0004) R-Squared 0.8739 0.2568 0.2102 0.7949 0.7786 Panel B: With Controls Foreign Males 0.0041*** -0.0235*** 0.0135** 0.0016 0.0039*** (0.0010) (0.0088) (0.0066) (0.0012) (0.0012) Native Males 0.0003 -0.0219** 0.0135* -0.0001 0.0002 (0.0008) (0.0090) (0.0069) (0.0006) (0.0005) Foreign Females 0.0037*** -0.0126 0.0111* 0.0013 0.0015 (0.0010) (0.0094) (0.0067) (0.0013) (0.0010) Native Females 0.0002 -0.0190** 0.0150** 0.0002 0.0004 (0.0008) (0.0088) (0.0070) (0.0003) (0.0004) R-Squared 0.8740 0.2573 0.2106 0.7951 0.7788 Native Males Mean 4.431 93.64 3.334 0.268 0.0731 Native Males SD 1.671 10.71 8.673 1.004 0.977 Native Females Mean 4.354 95.25 2.800 -0.102 0.0900 Native Females SD 1.707 9.383 8.178 0.907 0.964 Foreign Males Mean 1.101 90.79 6.241 0.811 0.511 Foreign Males SD 1.120 22.49 19.12 1.852 1.764 Foreign Females Mean 1.082 92.10 5.834 0.236 0.406 Foreign Females SD 1.104 21.18 18.71 1.697 1.736 Observations 334,160 334,160 334,160 120,478 120,478 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. 37 Table 4. Average effect of the Venezuelan migration shock on educa- tional outcomes – Public schools (marginal effects) (1) (2) (3) (4) (5) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Math Std. Language Std. Score Score Panel A: Without Controls Foreign Males 0.0053*** -0.0294*** 0.0128** 0.0015 0.0041** (0.0013) (0.0096) (0.0064) (0.0013) (0.0020) Native Males 0.0013 -0.0312*** 0.0119** 0.0000 0.0003 (0.0012) (0.0090) (0.0059) (0.0006) (0.0005) Foreign Females 0.0051*** -0.0193* 0.0126* 0.0022** 0.0022** (0.0013) (0.0114) (0.0068) (0.0009) (0.0010) Native Females 0.0018 -0.0176** 0.0102 0.0004 0.0007* (0.0013) (0.0088) (0.0064) (0.0003) (0.0004) R-Squared 0.8980 0.2539 0.2167 0.6896 0.6729 Panel B: With Controls Foreign Males 0.0051*** -0.0286*** 0.0129** 0.0014 0.0041** (0.0013) (0.0093) (0.0064) (0.0012) (0.0020) Native Males 0.0011 -0.0305*** 0.0120** -0.0001 0.0002 (0.0011) (0.0089) (0.0059) (0.0006) (0.0004) Foreign Females 0.0049*** -0.0185* 0.0126* 0.0021** 0.0021** (0.0013) (0.0112) (0.0067) (0.0009) (0.0010) Native Females 0.0016 -0.0170* 0.0103 0.0003 0.0006 (0.0013) (0.0088) (0.0064) (0.0003) (0.0004) R-Squared 0.8982 0.2544 0.2169 0.6907 0.6735 Native Males Mean 4.706 91.76 3.961 -0.0329 -0.249 Native Males SD 1.787 11.32 8.874 0.743 0.751 Native Females Mean 4.648 94.07 3.176 -0.400 -0.262 Native Females SD 1.820 9.700 8.194 0.626 0.691 Foreign Males Mean 1.247 87.48 8.391 -0.0445 -0.231 Foreign Males SD 1.167 25.09 21.43 1.389 1.535 Foreign Females Mean 1.221 89.29 7.842 -0.527 -0.319 Foreign Females SD 1.147 23.81 21.09 1.319 1.475 Observations 203,326 203,326 203,326 81,082 81,082 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. 38 Table 5. Average effect of the Venezuelan migration shock on educa- tional outcomes – Private schools (marginal effects) (1) (2) (3) (4) (5) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Math Std. Language Std. Score Score Panel A: Without Controls Foreign Males 0.0027** -0.0111 0.0045 0.0031** 0.0034*** (0.0011) (0.0121) (0.0156) (0.0014) (0.0009) Native Males 0.0000 -0.0060 0.0008 0.0003 -0.0000 (0.0018) (0.0151) (0.0187) (0.0013) (0.0010) Foreign Females 0.0022** 0.0022 -0.0019 -0.0009 -0.0010 (0.0010) (0.0149) (0.0173) (0.0025) (0.0012) Native Females -0.0006 -0.0067 0.0032 0.0003 -0.0003 (0.0019) (0.0151) (0.0184) (0.0013) (0.0008) R-Squared 0.8363 0.2059 0.2087 0.7915 0.7415 Panel B: With Controls Foreign Males 0.0025** -0.0110 0.0046 0.0032** 0.0035*** (0.0011) (0.0119) (0.0153) (0.0014) (0.0008) Native Males -0.0002 -0.0059 0.0009 0.0005 0.0001 (0.0017) (0.0149) (0.0184) (0.0013) (0.0010) Foreign Females 0.0020** 0.0023 -0.0018 -0.0007 -0.0009 (0.0010) (0.0148) (0.0169) (0.0025) (0.0012) Native Females -0.0008 -0.0066 0.0032 0.0005 -0.0002 (0.0018) (0.0149) (0.0181) (0.0014) (0.0009) R-Squared 0.8364 0.2063 0.2091 0.7916 0.7417 Native Males Mean 4.009 96.50 2.373 0.906 0.756 Native Males SD 1.373 8.971 8.263 1.174 1.046 Native Females Mean 3.905 97.06 2.225 0.510 0.813 Native Females SD 1.405 8.565 8.119 1.074 1.040 Foreign Males Mean 0.871 96.01 2.841 1.788 1.359 Foreign Males SD 0.999 16.30 14.10 1.833 1.621 Foreign Females Mean 0.853 96.75 2.519 1.125 1.251 Foreign Females SD 0.988 14.76 13.27 1.656 1.632 Observations 130,834 130,834 130,834 39,396 39,396 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. 39 Table 6. Average effect of the Venezuelan migration shock on educa- tional outcomes – Primary school grades (marginal effects) (1) (2) (3) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Panel A: Without Controls Foreign Males 0.0038*** -0.0257*** 0.0175*** (0.0009) (0.0080) (0.0067) Native Males 0.0010 -0.0187** 0.0109 (0.0008) (0.0075) (0.0070) Foreign Females 0.0036*** -0.0115 0.0085 (0.0009) (0.0088) (0.0070) Native Females 0.0008 -0.0167** 0.0120* (0.0008) (0.0078) (0.0072) R-Squared 0.8691 0.2449 0.2063 Panel B: With Controls Foreign Males 0.0036*** -0.0247*** 0.0162** (0.0008) (0.0077) (0.0065) Native Males 0.0007 -0.0177** 0.0095 (0.0007) (0.0074) (0.0070) Foreign Females 0.0034*** -0.0105 0.0071 (0.0009) (0.0087) (0.0069) Native Females 0.0006 -0.0157** 0.0106 (0.0008) (0.0077) (0.0072) R-Squared 0.8693 0.2453 0.2068 Native Males Mean 4.005 94.61 2.941 Native Males SD 1.442 10.24 8.492 Native Females Mean 3.892 95.91 2.470 Native Females SD 1.462 9.077 7.976 Foreign Males Mean 0.911 91.56 5.933 Foreign Males SD 1.017 22.45 19.27 Foreign Females Mean 0.884 92.70 5.529 Foreign Females SD 0.997 21.20 18.82 Observations 313,195 313,195 313,195 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. 40 Table 7. Average effect of the Venezuelan migration shock on educa- tional outcomes – Secondary school grades (marginal effects) (1) (2) (3) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Panel A: Without Controls Foreign Males 0.0023*** -0.0152 0.0032 (0.0007) (0.0111) (0.0075) Native Males 0.0005 -0.0147 0.0066 (0.0007) (0.0116) (0.0074) Foreign Females 0.0025*** -0.0212* 0.0190** (0.0008) (0.0118) (0.0089) Native Females 0.0005 -0.0104 0.0067 (0.0007) (0.0110) (0.0077) R-Squared 0.8784 0.2680 0.2098 Panel B: With Controls Foreign Males 0.0022*** -0.0143 0.0027 (0.0007) (0.0110) (0.0075) Native Males 0.0005 -0.0139 0.0061 (0.0007) (0.0117) (0.0075) Foreign Females 0.0024*** -0.0203* 0.0185** (0.0008) (0.0117) (0.0088) Native Females 0.0004 -0.0095 0.0062 (0.0007) (0.0110) (0.0077) R-Squared 0.8784 0.2684 0.2099 Native Males Mean 4.611 91.17 4.077 Native Males SD 1.292 11.50 8.145 Native Females Mean 4.614 93.83 3.322 Native Females SD 1.291 9.446 7.659 Foreign Males Mean 0.880 88.99 6.790 Foreign Males SD 0.990 25.97 21.11 Foreign Females Mean 0.861 90.95 6.290 Foreign Females SD 0.980 23.91 20.44 Observations 203,560 203,560 203,560 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. 41 Table 8. Average effect of the Venezuelan migration shock on educa- tional outcomes – Urban areas (marginal effects) (1) (2) (3) (4) (5) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Math Std. Language Std. Score Score Panel A: Without Controls Foreign Males 0.0041*** -0.0205* 0.0112 0.0014 0.0036*** (0.0010) (0.0108) (0.0078) (0.0011) (0.0011) Native Males -0.0006 -0.0164 0.0079 -0.0004 -0.0000 (0.0011) (0.0113) (0.0089) (0.0007) (0.0005) Foreign Females 0.0038*** -0.0103 0.0079 0.0008 0.0013 (0.0010) (0.0115) (0.0080) (0.0015) (0.0010) Native Females -0.0008 -0.0130 0.0086 0.0000 0.0002 (0.0012) (0.0113) (0.0089) (0.0004) (0.0004) R-Squared 0.8683 0.2497 0.2033 0.8051 0.7836 Panel B: With Controls Foreign Males 0.0040*** -0.0201* 0.0106 0.0014 0.0036*** (0.0009) (0.0105) (0.0077) (0.0011) (0.0011) Native Males -0.0007 -0.0160 0.0074 -0.0004 -0.0000 (0.0011) (0.0112) (0.0089) (0.0007) (0.0005) Foreign Females 0.0037*** -0.0099 0.0073 0.0008 0.0013 (0.0009) (0.0114) (0.0080) (0.0015) (0.0010) Native Females -0.0010 -0.0126 0.0080 -0.0000 0.0002 (0.0012) (0.0112) (0.0089) (0.0004) (0.0004) R-Squared 0.8684 0.2502 0.2037 0.8052 0.7837 Native Males Mean 4.631 93.74 3.309 0.387 0.214 Native Males SD 1.627 10.50 8.572 1.024 0.981 Native Females Mean 4.561 95.24 2.847 -0.00142 0.227 Native Females SD 1.660 9.275 8.158 0.934 0.974 Foreign Males Mean 1.138 91.13 5.880 0.897 0.603 Foreign Males SD 1.137 21.80 18.32 1.870 1.755 Foreign Females Mean 1.119 92.42 5.545 0.311 0.470 Foreign Females SD 1.120 20.45 18 1.694 1.747 Observations 253,107 253,107 253,107 95,866 95,866 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. 42 Table 9. Average effect of the Venezuelan migration shock on educa- tional outcomes – Rural areas (marginal effects) (1) (2) (3) (4) (5) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Math Std. Language Std. Score Score Panel A: Without Controls Foreign Males 0.0077*** -0.0314 -0.0051 0.0016 0.0060 (0.0020) (0.0239) (0.0184) (0.0036) (0.0064) Native Males 0.0110*** -0.0343 0.0033 -0.0006 -0.0003 (0.0019) (0.0220) (0.0153) (0.0010) (0.0011) Foreign Females 0.0073*** -0.0100 -0.0019 0.0046* 0.0014 (0.0018) (0.0239) (0.0181) (0.0024) (0.0041) Native Females 0.0112*** -0.0180 0.0043 0.0005 0.0007 (0.0019) (0.0224) (0.0156) (0.0010) (0.0014) R-Squared 0.8978 0.2908 0.2464 0.6467 0.5996 Panel B: With Controls Foreign Males 0.0076*** -0.0302 -0.0065 0.0015 0.0059 (0.0020) (0.0237) (0.0182) (0.0036) (0.0064) Native Males 0.0109*** -0.0331 0.0018 -0.0008 -0.0005 (0.0019) (0.0219) (0.0152) (0.0009) (0.0011) Foreign Females 0.0072*** -0.0088 -0.0033 0.0045* 0.0013 (0.0018) (0.0238) (0.0178) (0.0023) (0.0041) Native Females 0.0111*** -0.0168 0.0028 0.0003 0.0005 (0.0018) (0.0223) (0.0154) (0.0010) (0.0014) R-Squared 0.8979 0.2910 0.2467 0.6477 0.6003 Native Males Mean 3.888 93.34 3.400 -0.180 -0.459 Native Males SD 1.669 11.23 8.939 0.774 0.752 Native Females Mean 3.789 95.28 2.670 -0.492 -0.442 Native Females SD 1.708 9.671 8.230 0.667 0.704 Foreign Males Mean 0.934 89.25 7.876 -0.339 -0.713 Foreign Males SD 1.027 25.33 22.33 1.084 1.384 Foreign Females Mean 0.912 90.64 7.175 -0.834 -0.504 Foreign Females SD 1.007 24.20 21.67 1.353 1.248 Observations 81,049 81,049 81,049 24,612 24,612 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. 43 Table 10. Effect of the Venezuelan migration shock on educational outcomes by quartile of enrollment capacity gap (1) (2) (3) (4) (5) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Math Std. Language Std. Score Score Panel A: Capacity gap Q4 Foregin Males 0.0040*** 0.0019 -0.0149 0.0011 0.0031 (0.0009) (0.0156) (0.0174) (0.0013) (0.0021) Native Males -0.0045* -0.0002 -0.0131 -0.0001 -0.0001 (0.0024) (0.0121) (0.0136) (0.0009) (0.0006) Foreign Females 0.0032*** 0.0180 -0.0218 0.0010 0.0025 (0.0009) (0.0202) (0.0188) (0.0016) (0.0025) Native Females -0.0045* 0.0050 -0.0160 0.0003 0.0004 (0.0026) (0.0133) (0.0134) (0.0004) (0.0004) Native Males Mean 5.768 91.97 3.533 0.135 -0.0486 Native Males SD 1.212 8.473 4.955 0.833 0.817 Native Females Mean 5.760 94.22 2.878 -0.236 -0.0489 Native Females SD 1.125 6.607 4.570 0.757 0.810 Foreign Males Mean 1.387 88.58 7.697 0.521 0.238 Foreign Males SD 1.204 23.53 20.09 1.741 1.653 Foreign Females Mean 1.339 90.26 7.153 -0.0141 0.197 Foreign Females SD 1.179 22.33 19.72 1.580 1.644 Panel B: Capacity gap Q3 Foregin Males 0.0007 -0.0176 0.0086 0.0013 0.0033*** (0.0008) (0.0143) (0.0105) (0.0018) (0.0011) Native Males -0.0017 -0.0185** 0.0100 0.0001 0.0001 (0.0012) (0.0088) (0.0075) (0.0005) (0.0006) Foreign Females 0.0009 -0.0162 0.0162** 0.0026** 0.0010* (0.0007) (0.0129) (0.0083) (0.0010) (0.0006) Native Females -0.0021 -0.0164 0.0132* 0.0004 0.0002 (0.0014) (0.0104) (0.0075) (0.0004) (0.0006) Native Males Mean 4.700 93.70 2.965 0.211 0.00213 Native Males SD 1.260 8.254 5.581 0.990 0.977 Native Females Mean 4.640 95.60 2.355 -0.130 0.0352 Native Females SD 1.284 6.618 5.048 0.886 0.965 Foreign Males Mean 0.984 90.87 6.038 0.976 0.717 Foreign Males SD 1.044 23.18 19.41 1.930 1.837 Foreign Females Mean 0.957 92.33 5.640 0.402 0.489 Foreign Females SD 1.028 21.53 19.09 1.727 1.755 Panel C: Capacity gap Q2 Foregin Males 0.0020* -0.0269* 0.0293** 0.0017 -0.0041 (0.0011) (0.0154) (0.0138) (0.0025) (0.0035) Native Males 0.0011 -0.0315** 0.0345*** 0.0001 0.0001 (0.0012) (0.0126) (0.0106) (0.0007) (0.0007) Foreign Females 0.0019 -0.0172 0.0249** -0.0028 -0.0030 (0.0013) (0.0116) (0.0107) (0.0023) (0.0026) Native Females 0.0006 -0.0271** 0.0358*** 0.0006 0.0003 (0.0012) (0.0110) (0.0110) (0.0007) (0.0008) Native Males Mean 3.492 95.74 2.337 0.429 0.221 Native Males SD 1.549 8.735 6.684 1.152 1.105 Native Females Mean 3.379 96.79 1.984 0.0433 0.251 Native Females SD 1.583 7.653 6.369 1.028 1.081 Foreign Males Mean 0.773 92.77 4.827 1.016 0.574 Foreign Males SD 0.922 21.10 17.76 1.796 1.597 Foreign Females Mean 0.780 93.79 4.522 0.428 0.556 Foreign Females SD 0.926 19.92 17.33 1.686 1.676 Panel D: Capacity gap Q1 Foregin Males 0.0076*** -0.0466*** 0.0309*** 0.0033 0.0097*** (0.0018) (0.0117) (0.0093) (0.0026) (0.0028) Native Males 0.0051*** -0.0345*** 0.0247*** -0.0005 0.0009* (0.0018) (0.0125) (0.0095) (0.0007) (0.0005) Foreign Females 0.0071*** -0.0328*** 0.0278*** 0.0010 0.0034 (0.0017) (0.0123) (0.0094) (0.0024) (0.0030) Native Females 0.0050*** -0.0299*** 0.0247** -0.0003 0.0009* (0.0018) (0.0115) (0.0096) (0.0006) (0.0005) Native Males Mean 4.504 95.01 2.653 0.496 0.302 Native Males SD 1.466 8.390 6.470 1.139 1.088 Native Females Mean 4.394 96.21 2.287 0.101 0.322 Native Females SD 1.521 7.279 6.094 1.040 1.075 Foreign Males Mean 1.043 92.62 5.040 1.080 0.768 Foreign Males SD 1.104 20.21 17.19 1.940 1.866 Foreign Females Mean 1.032 93.67 4.620 0.451 0.641 Foreign Females SD 1.093 18.91 16.47 1.837 1.873 Observations 304,320 304,320 304,320 118,509 118,509 R-squared 0.8841 0.2239 0.1593 0.7955 0.7791 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. 44 Table 11. Average effect of the Venezuelan migration shock on school inputs (1) (2) Dependent Var. Ln(# Teachers) Pupil/Teacher Ratio Panel A: All schools (with controls) Predicted Ven. Shock -0.0021*** 0.1899*** (0.0007) (0.0691) R-squared 0.9826 0.6087 Observations 42,765 42,765 Dep. Var. Mean 2.762 21.26 Dep. Var. SD 1.477 39.27 Panel B: Primary school grades Predicted Ven. Shock -0.0007 0.0825* (0.0005) (0.0454) R-squared 0.9658 0.6654 Observations 42,412 42,412 Dep. Var. Mean 2.152 23.29 Dep. Var. SD 1.159 27.21 Panel C: Secondary school grades Predicted Ven. Shock -0.0010*** 0.0241 (0.0003) (0.0177) R-squared 0.9644 0.5960 Observations 32,195 32,195 Dep. Var. Mean 2.601 24.11 Dep. Var. SD 1.003 22.30 Panel D: Urban schools (with controls) Predicted Ven. Shock -0.0018** 0.1370* (0.0008) (0.0704) R-squared 0.9724 0.5815 Observations 25,135 25,135 Dep. Var. Mean 3.271 23.04 Dep. Var. SD 1.266 42.80 Panel E: Rural schools (with controls) Predicted Ven. Shock -0.0030** 0.2952*** (0.0013) (0.0951) R-squared 0.9853 0.6849 Observations 17,555 17,555 Dep. Var. Mean 2.046 18.76 Dep. Var. SD 1.455 33.51 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: Hom. Rate × 2015, N. of Terrorist Attacks × 2015, N. of Terrorist Attacks × 2018, Informal Labor × 2014, Informal Labor × 2017, Informal Labor × 2017, Night Light Density × 2015, UBN × 2018. i APPENDIX Table A.1. Average effect of the Venezuelan migration shock on edu- cational outcomes – All schools (regression output) (1) (2) (3) (4) (5) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Math Std. Language Std. Score Score Predicted Ven. Shock 0.0041*** -0.0235*** 0.0135** 0.0016 0.0039*** (0.0010) (0.0088) (0.0066) (0.0012) (0.0012) Native 4.1801*** 3.0021*** -3.2734*** 0.0289 0.1482*** (0.0989) (0.3391) (0.3439) (0.0370) (0.0342) Native×Pred. Ven. Shock -0.0038*** 0.0016 0.0000 -0.0017* -0.0037*** (0.0014) (0.0059) (0.0046) (0.0009) (0.0010) Female -0.0627*** 1.2665*** -0.3641** -0.5436*** -0.0562 (0.0093) (0.2388) (0.1501) (0.0537) (0.0605) Female×Pred. Ven. Shock -0.0003** 0.0109* -0.0024 -0.0003 -0.0024 (0.0002) (0.0058) (0.0024) (0.0019) (0.0017) Native×Female -0.0239*** 0.3168** -0.1628 0.1445*** 0.0399 (0.0079) (0.1589) (0.1306) (0.0531) (0.0581) Native×Female 0.0002 -0.0079** 0.0039 0.0005 0.0027* × Pred. Ven. Shock (0.0002) (0.0039) (0.0027) (0.0017) (0.0016) R-squared 0.8740 0.2573 0.2106 0.7951 0.7788 Native Males Mean 4.431 93.64 3.334 0.268 0.0731 Native Males SD 1.671 10.71 8.673 1.004 0.977 Native Females Mean 4.354 95.25 2.800 -0.102 0.0900 Native Females SD 1.707 9.383 8.178 0.907 0.964 Foreign Males Mean 1.101 90.79 6.241 0.811 0.511 Foreign Males SD 1.120 22.49 19.12 1.852 1.764 Foreign Females Mean 1.082 92.10 5.834 0.236 0.406 Foreign Females SD 1.104 21.18 18.71 1.697 1.736 Observations 334,160 334,160 334,160 120,478 120,478 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. ii Table A.2. Average effect of the Venezuelan migration shock on edu- cational outcomes – Public schools (regression output) (1) (2) (3) (4) (5) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Math Std. Language Std. Score Score Predicted Ven. Shock 0.0051*** -0.0286*** 0.0129** 0.0014 0.0041** (0.0013) (0.0093) (0.0064) (0.0012) (0.0020) Native 4.4985*** 4.3857*** -4.7909*** 0.1190*** 0.1666*** (0.1034) (0.2851) (0.2920) (0.0460) (0.0529) Native × Pred. Ven. Shock -0.0040** -0.0019 -0.0009 -0.0015 -0.0039** (0.0017) (0.0085) (0.0064) (0.0013) (0.0018) Female -0.0515*** 1.7446*** -0.5195** -0.4516*** -0.0246 (0.0085) (0.2768) (0.2016) (0.0585) (0.0746) Female × Pred. Ven. Shock -0.0002 0.0101 -0.0003 0.0007 -0.0019 (0.0002) (0.0066) (0.0029) (0.0015) (0.0023) Native ×Female -0.0185** 0.4666** -0.2351 0.0688 -0.0066 (0.0094) (0.2235) (0.1909) (0.0589) (0.0745) Native ×Female 0.0006*** 0.0034 -0.0014 -0.0003 0.0024 × Pred. Ven. Shock (0.0002) (0.0036) (0.0032) (0.0016) (0.0023) R-squared 0.8982 0.2544 0.2169 0.6907 0.6735 Native Males Mean 4.706 91.76 3.961 -0.0329 -0.249 Native Males SD 1.787 11.32 8.874 0.743 0.751 Native Females Mean 4.648 94.07 3.176 -0.400 -0.262 Native Females SD 1.820 9.700 8.194 0.626 0.691 Foreign Males Mean 1.247 87.48 8.391 -0.0445 -0.231 Foreign Males SD 1.167 25.09 21.43 1.389 1.535 Foreign Females Mean 1.221 89.29 7.842 -0.527 -0.319 Foreign Females SD 1.147 23.81 21.09 1.319 1.475 Observations 203,326 203,326 203,326 81,082 81,082 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. iii Table A.3. Average effect of the Venezuelan migration shock on edu- cational outcomes – Private schools (regression output) (1) (2) (3) (4) (5) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Math Std. Language Std. Score Score Predicted Ven. Shock 0.0025** -0.0110 0.0046 0.0032** 0.0035*** (0.0011) (0.0119) (0.0153) (0.0014) (0.0008) Native 3.6962*** 0.8736*** -0.9418*** -0.0599 0.1195*** (0.1584) (0.1909) (0.1857) (0.0663) (0.0428) Native × Pred. Ven. Shock -0.0027 0.0051 -0.0037 -0.0028*** -0.0034*** (0.0019) (0.0058) (0.0050) (0.0008) (0.0012) Female -0.0668*** 0.5288*** -0.1745 -0.6320*** -0.0796 (0.0178) (0.1424) (0.1262) (0.0799) (0.0841) Female × Pred. Ven. Shock -0.0004 0.0133** -0.0064** -0.0040 -0.0044*** (0.0004) (0.0054) (0.0032) (0.0028) (0.0014) Native ×Female -0.0456*** 0.0337 0.0175 0.1978** 0.0966 (0.0127) (0.1401) (0.1352) (0.0823) (0.0792) Native ×Female -0.0001 -0.0140*** 0.0087** 0.0040 0.0041*** × Pred. Ven. Shock (0.0004) (0.0050) (0.0035) (0.0027) (0.0015) R-squared 0.8364 0.2063 0.2091 0.7916 0.7417 Native Males Mean 4.009 96.50 2.373 0.906 0.756 Native Males SD 1.373 8.971 8.263 1.174 1.046 Native Females Mean 3.905 97.06 2.225 0.510 0.813 Native Females SD 1.405 8.565 8.119 1.074 1.040 Foreign Males Mean 0.871 96.01 2.841 1.788 1.359 Foreign Males SD 0.999 16.30 14.10 1.833 1.621 Foreign Females Mean 0.853 96.75 2.519 1.125 1.251 Foreign Females SD 0.988 14.76 13.27 1.656 1.632 Observations 130,834 130,834 130,834 39,396 39,396 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. iv Table A.4. Average effect of the Venezuelan migration shock on edu- cational outcomes – Primary school grades (regression output) (1) (2) (3) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Predicted Ven. Shock 0.0036*** -0.0247*** 0.0162** (0.0008) (0.0077) (0.0065) Native 3.8177*** 3.2320*** -3.2290*** (0.0841) (0.3415) (0.3264) Native×Pred. Ven. Shock -0.0029** 0.0070 -0.0067 (0.0012) (0.0048) (0.0046) Female -0.0674*** 1.0542*** -0.3126* (0.0075) (0.2419) (0.1609) Female×Pred. Ven. Shock -0.0002 0.0142*** -0.0091*** (0.0002) (0.0054) (0.0029) Native×Female -0.0559*** 0.2119 -0.1423 (0.0069) (0.1740) (0.1437) Native×Female 0.0001 -0.0121*** 0.0102*** × Pred. Ven. Shock (0.0002) (0.0035) (0.0032) R-squared 0.8693 0.2453 0.2068 Native Males Mean 4.005 94.61 2.941 Native Males SD 1.442 10.24 8.492 Native Females Mean 3.892 95.91 2.470 Native Females SD 1.462 9.077 7.976 Foreign Males Mean 0.911 91.56 5.933 Foreign Males SD 1.017 22.45 19.27 Foreign Females Mean 0.884 92.70 5.529 Foreign Females SD 0.997 21.20 18.82 Observations 313,195 313,195 313,195 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. v Table A.5. Average effect of the Venezuelan migration shock on edu- cational outcomes – Secondary school grades (regression output) (1) (2) (3) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Predicted Ven. Shock 0.0022*** -0.0143 0.0027 (0.0007) (0.0110) (0.0075) Native 4.1614*** 3.1487*** -3.5184*** (0.0985) (0.3911) (0.3705) Native×Pred. Ven. Shock -0.0018 0.0005 0.0034 (0.0011) (0.0070) (0.0050) Female -0.0426*** 2.0774*** -0.6452*** (0.0108) (0.3082) (0.1673) Female×Pred. Ven. Shock 0.0002 -0.0060 0.0158*** (0.0002) (0.0060) (0.0045) Native×Female 0.0406*** 0.5387** -0.0893 (0.0107) (0.2164) (0.1623) Native×Female -0.0002 0.0103** -0.0157*** × Pred. Ven. Shock (0.0003) (0.0045) (0.0043) R-squared 0.8784 0.2684 0.2099 Native Males Mean 4.611 91.17 4.077 Native Males SD 1.292 11.50 8.145 Native Females Mean 4.614 93.83 3.322 Native Females SD 1.291 9.446 7.659 Foreign Males Mean 0.880 88.99 6.790 Foreign Males SD 0.990 25.97 21.11 Foreign Females Mean 0.861 90.95 6.290 Foreign Females SD 0.980 23.91 20.44 Observations 203,560 203,560 203,560 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. vi Table A.6. Average effect of the Venezuelan migration shock on edu- cational outcomes – Urban areas (regression output) (1) (2) (3) (4) (5) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Math Std. Language Std. Score Score Predicted Ven. Shock 0.0040*** -0.0201* 0.0106 0.0014 0.0036*** (0.0009) (0.0105) (0.0077) (0.0011) (0.0011) Native 4.2357*** 2.6759*** -2.9175*** 0.0253 0.1340*** (0.1154) (0.3532) (0.3625) (0.0399) (0.0359) Native×Pred. Ven. Shock -0.0047*** 0.0041 -0.0032 -0.0018* -0.0036*** (0.0018) (0.0066) (0.0052) (0.0009) (0.0009) Female -0.0632*** 1.2371*** -0.2763* -0.5479*** -0.0838 (0.0112) (0.2774) (0.1614) (0.0556) (0.0605) Female×Pred. Ven. Shock -0.0003* 0.0102* -0.0033 -0.0006 -0.0023 Cum. Ven. Inflows (0.0002) (0.0059) (0.0026) (0.0020) (0.0017) Native×Female -0.0146 0.2210 -0.1742 0.1260** 0.0600 (0.0091) (0.1699) (0.1399) (0.0555) (0.0591) Native×Female 0.0001 -0.0069* 0.0040 0.0010 0.0026 × Pred. Ven. Shock (0.0002) (0.0036) (0.0027) (0.0019) (0.0016) R-squared 0.8684 0.2502 0.2037 0.8052 0.7837 Native Males Mean 4.631 93.74 3.309 0.387 0.214 Native Males SD 1.627 10.50 8.572 1.024 0.981 Native Females Mean 4.561 95.24 2.847 -0.00142 0.227 Native Females SD 1.660 9.275 8.158 0.934 0.974 Foreign Males Mean 1.138 91.13 5.880 0.897 0.603 Foreign Males SD 1.137 21.80 18.32 1.870 1.755 Foreign Females Mean 1.119 92.42 5.545 0.311 0.470 Foreign Females SD 1.120 20.45 18 1.694 1.747 Observations 253,107 253,107 253,107 95,866 95,866 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013. vii Table A.7. Average effect of the Venezuelan migration shock on edu- cational outcomes – Rural areas (regression output) (1) (2) (3) (4) (5) Dependent Var. Ln(Enrollment) Promotion Rate Dropout Rate Math Std. Language Std. Score Score Predicted Ven. Shock 0.0076*** -0.0302 -0.0065 0.0015 0.0059 (0.0020) (0.0237) (0.0182) (0.0036) (0.0064) Native 3.9437*** 4.3602*** -4.7548*** 0.2053* 0.3977*** (0.1225) (0.4471) (0.4435) (0.1245) (0.1460) Native×Pred. Ven. Shock 0.0033 -0.0029 0.0083 -0.0023 -0.0064 (0.0029) (0.0141) (0.0122) (0.0034) (0.0063) Females -0.0540*** 1.3224*** -0.7132** -0.4696*** 0.3454* (0.0163) (0.3561) (0.2949) (0.1743) (0.1902) Female×Pred. Ven. Shock -0.0004 0.0214 0.0033 0.0030 -0.0046 (0.0007) (0.0154) (0.0105) (0.0041) (0.0063) Native×Female -0.0542*** 0.5544 0.0009 0.1484 -0.3369* (0.0170) (0.3593) (0.2983) (0.1747) (0.1901) Native×Female 0.0006 -0.0051 -0.0022 -0.0019 0.0056 × Pred. Ven. Shock (0.0007) (0.0155) (0.0110) (0.0042) (0.0064) R-squared 0.8979 0.2910 0.2467 0.6477 0.6003 Native Males Mean 3.888 93.34 3.400 -0.180 -0.459 Native Males SD 1.669 11.23 8.939 0.774 0.752 Native Females Mean 3.789 95.28 2.670 -0.492 -0.442 Native Females SD 1.708 9.671 8.230 0.667 0.704 Foreign Males Mean 0.934 89.25 7.876 -0.339 -0.713 Foreign Males SD 1.027 25.33 22.33 1.084 1.384 Foreign Females Mean 0.912 90.64 7.175 -0.834 -0.504 Foreign Females SD 1.007 24.20 21.67 1.353 1.248 Observations 81,049 81,049 81,049 24,612 24,612 Clustered standard errors by municipality in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All columns include School FE, Year FE and Department × Year FE. Selected controls include the following variables interacted with year dummies: N. of Terrorist Attacks × 2012, 2014 and 2016, Informal Labor × 2018, UBN × 2013.