The World Bank Economic Review, 37(3), 2023, 437–459 https://doi.org10.1093/wber/lhad008 Article Taking Cover: Human Capital Accumulation in the Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 Presence of Shocks and Health Insurance Paulino Font-Gilabert Abstract Using the expansion of a large-scale health-insurance program in Mexico and variation in local rainfall levels, this study explores whether the program-induced increase in healthcare coverage protected the cognitive attain- ment of primary school children in the event of adverse rainfall shocks. Results show that the universalization of healthcare mitigated the negative effect of atypical rainfall on test scores, particularly in more marginalized and rural areas. An analysis of the mechanisms at play shows a reduced incidence of sickness among children, lower demand for their time, and higher stability in household consumption among program-eligible families exposed to rainfall shocks. JEL classification: I13, I15, I25, O12 Keywords: universal healthcare, resilience to shocks, cognitive attainment, Seguro Popular 1. Introduction Universal health coverage has recently evidenced its potential to protect the world’s population against global health shocks (Aarabi, Haghighi, and Takian 2020). While the WHO pushes for its expansion as a major goal for health reform to address inequalities in health and financial outcomes (WHO and World Bank 2017, 2020), little is known about the protective effect that social health insurance can have in other domains. This study seeks to investigate whether access to public health insurance protects children’s perfor- mance in school during adversity. The study therefore contributes to a growing area of research evaluat- ing the extent to which different public programs can mitigate the negative effect of weather and other environmental shocks on human capital formation. For instance, conditional cash transfers ease the neg- ative effect of rainfall shocks on educational attainment in Mexico (Adhvaryu et al. 2018) and Colombia (Duque, Rosales-Rueda, and Sanchez Torres 2019), a rural employment scheme in India mitigates the Paulino Font-Gilabert is a research affiliate at King’s College London, 16 De Crespigny Park, London SE5 8AF, UK; his email address is paulino.font_gilabert@kcl.ac.uk. For helpful comments and discussions, the author thanks Sonia Bhalotra, seminar participants at ISER, the ICID-SITES-IFAD conference in Rome, the Public Health and Development Workshop at the University of Gothenburg, the EUDN PhD workshop on Development Economics at CERDI, the ESPE 2019 Annual Conference, the Young Economists’ Meeting at Masaryk University, and the EEA-ESEM 2019 Congress. The author also thanks three anonymous referees for providing valuable feedback, and gratefully acknowledges funding from the Economic and Social Research Council – Research Centre on Micro-Social Change. A supplementary online appendix is available with this article at The World Bank Economic Review website. The author declares that he has no relevant or material financial interests that relate to the research described in this paper. © The Author(s) 2023. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. 438 Font-Gilabert impact of heat waves on children’s cognition (Garg, Jagnani, and Taraz 2017), and the introduction of air conditioning in schools alleviates the effect of heat exposure on test scores in the United States (Park et al. 2020).1 The health-insurance program under study is Seguro Popular (or Popular Health Insurance, hereafter also referred to as SP). The SP was the result of a reform of the Mexican health system in response to a political debate after national estimates showed that more than 50 percent of the health expenditures in Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 the country were out of pocket, with 2 to 4 million families estimated to be suffering from catastrophic and impoverishing health expenditures each year (Knaul et al. 2006). Starting in 2002 as a pilot program, it offered a comprehensive package of health services to individuals outside the social security system and, after 10 years of program expansion and more than 52 million new affiliations, it achieved its target of establishing universal healthcare. This study analyzes the capacity of universal healthcare in protecting the cognitive development of children in the event of negative shocks. To do so, it combines information from a yearly nationwide standardized test delivered to all students in certain grades in primary education, the expansion in health coverage induced by the reform of the health system and the creation of Seguro Popular, and rainfall precipitation measured at the school-locality level in a region where climatic conditions are influenced by El Niño/Southern Oscillation (ENSO).2 The results show that while adverse rainfall shocks reduce mathematics and verbal attainment by 0.022 and 0.020 standard deviations respectively, a 1 standard deviation increase in healthcare coverage mitigates 55 percent and 52 percent of the negative effect.3 The estimated results are driven by schools located in more marginalized and rural areas. Moreover, the impact of the shocks differs by intensity and nature, with dry periods imposing a higher burden on the process of learning and during which health coverage offsets the highest proportion of the adverse effect. On the other hand, robustness specifications suggest a null impact of health coverage on cognitive attainment in the absence of shocks. An exploration of the underlying mechanisms using household survey data shows that when hit by rainfall shocks, access to SP reduces the incidence of sickness among children from eligible families, de- creases the demand for children’s time, and protects household’s consumption levels. While negative rain- fall shocks increase by 6.6 percentage points the probability of children being sick, and by 14.1 percentage points their probability of being involved in domestic chores, each additional year of SP eligibility reduces these probabilities by 1.5 and 3.5 percentage points respectively (significant at the 5 percent level). Sim- ilarly, rainfall shocks are associated with a reduction of 16 percent in consumption expenditures among program-eligible households (18 percent in rural households and similar to the 16.7 percent reduction estimated by Bobonis (2009) for a sample of rural households in Mexico). Each additional year of SP availability, however, reduces by 4 percent (3 percent in rural areas) the negative effect. Overall, the story that emerges from the findings is one of positive spillovers on education from public investments in health. It provides evidence of the capacity of universal healthcare in building resilience in cognitive attainment against negative shocks experienced during childhood, and contributes to our understanding of some of the mechanisms at play. This study also speaks to the literature evaluating the effect of healthcare coverage on children’s edu- cational outcomes. Most of the evidence to date comes from the Medicaid program and the CHIP (Chil- 1 Similar studies have also looked at the protective effect of health programs on health at-birth. For instance, vitamin A sup- plementation at-birth reduces the adverse effects of exposure to a tornado on infant health in Bangladesh (Gunnsteinsson et al. 2019), and public health improvements in West Africa weaken the link between dust storms and child mortality (Adhvaryu et al. 2019). 2 An irregular climatic phenomenon that has been shown to affect precipitation levels in Mexico. 3 The effect of rainfall shocks on mathematics test scores is equivalent to erasing more than one-fourth of the gains from interventions that provide instructional materials, or more than one-sixth of the gains from teacher training programs (see McEwan (2015) for a review of randomized educational experiments in developing countries). The World Bank Economic Review 439 dren’s Health Insurance Program) in the United States, which target families and children in poverty and under specific vulnerable conditions (see for instance Levine and Schanzenbach (2009) and Cohodes et al. (2016)). Instead, I estimate the effect of healthcare coverage in a context of high regional imbalances and exploit a nationwide policy to implement universal health coverage. One other study has evaluated the link between health insurance and education in Mexico (Alcaraz et al. 2016), finding a positive associ- ation between healthcare coverage, school enrollment, and educational performance at the municipality Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 level. This study expands on previous findings by assessing the capacity of healthcare coverage in building resilience in children’s performance in school, and by investigating some of the mechanisms by which uni- versal healthcare might help children and their families endure adverse environmental shocks. This study is similar in spirit to Liu (2016), who using survey data shows that the expansion in health coverage across rural China increased the probability of children being enrolled in school following a household health shock. To avoid the potential endogeneity of health shocks and risk-sharing networks among neighboring households, I focus on rainfall shocks experienced at the locality level, and focus on children’s performance in school (instead of enrollment) using administrative data. Because adverse rainfall shocks are one of the most prevalent disturbances experienced among the poor (Dinkelman 2013), the results of this study are highly relevant to a large share of the population of the world. Climate instability has consolidated as one of the major threats to developmental gains, including gains in human capital, and there is international consensus to develop and implement policies that mitigate its negative effects on the population (Field et al. 2012). To the best of my knowledge, this is the first study to assess the capacity of universal healthcare in mitigating the effect of negative shocks on children’s cognitive performance. 2. Seguro Popular: Health Insurance for the Poor Before the creation of Seguro Popular, social health insurance was administered by two main institutions that still exist today. On the one hand, the Mexican Social Security Institute (IMSS), covering the workers of the private sector; and the Institute for Social Security and Services for State Workers (ISSSTE), covering public employees.4 , 5 Those families not integrated into any of the former institutions could seek healthcare assis- tance through the conditional cash transfer program and main anti-poverty program in the country (Progresa/Oportunidades), or in the Coverage Expansion Program (PAC), which consisted of mobile healthcare teams visiting the most isolated regions and communities in the country.6 All other workers in the informal sector and individuals detached from the labor market could seek medical care in either health facilities managed by the Ministry of Health (SSA) or in the private sector. In both cases, medical attention and prescription drugs were at the expense of the user. As a result, the health system left half of the population uninsured. While Mexico ranked 51 out of 191 countries in the overall attainment in health in the World Health Report 2000, its health system placed 144 with respect to its fairness in finan- cial contribution (WHO 2000). National-level estimates showed that more than 50 percent of the health expenditures were out of pocket and that between 2 and 4 million families suffered from catastrophic and impoverishing health expenditures each year (Knaul et al. 2006). The low levels of financial protection in health were one of the major catalysts for the creation of Seguro Popular, which was introduced in 2002 as a pilot program and became the central pillar of the reform of the health system of 2003. The new law, effective from January 1, 2004, created the System of 4 Also playing a more marginal role, the Mexican Petroleums (PEMEX), covering workers in the oil industries. 5 These institutions also administered other benefits such as pensions, disability benefits, and severance payments. 6 The Progresa program started in 1997 in rural areas and was renamed Oportunidades in 2002 when it expanded to urban areas. In 2014, the program’s name changed to Prospera. The Coverage Expansion Program or Programa de Ampliación de Cobertura (PAC) started in 1996. 440 Font-Gilabert Social Protection in Health (or SPSS in its acronym in Spanish) to provide health coverage and financial protection in health to all citizens with no access to social security and to consolidate universal healthcare and the right to health (Knaul et al. 2006).7 , 8 The services offered, listed in the Universal Catalog of Health Services (CAUSES), expanded as the program consolidated across the territory, and included the most cost-effective health interventions and the leading causes for outpatient and hospital utilization in the country (Bonilla-Chacín and Aguilera 2013). Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 The health reform also sought to increase the funds of the public health system and to reduce the inequalities in public health spending across insurance schemes and regions (Kurowski et al. 2012). See also figs S1.1 and S1.2 in the supplementary online appendix. The push for universal healthcare resulted in the construction of new patient clinics and hospitals, with the proportion of the Ministry of Health budget devoted to investments in healthcare infrastructure increasing from 3.8 percent in 2000 to 9.1 percent in 2006 (Frenk, Gomez-Dantes, and Knaul 2009). Moreover, the gap in the availability of medical personnel between individuals covered by the social security and those that were not decreased substantially (Knaul et al. 2012), as did the difference in the numbers of hospitals and beds between poor and rich municipalities (Conti and Ginja 2023). The financial resources of SP come mostly from the federal government and the states.9 Although initially only families in the first two deciles of the income distribution were exempt from any payments, in practice very few households ever paid (Knaul et al. 2012).10 Furthermore, the reform introduced incentives for the states to expand coverage, as historical health budget allocations were replaced with a premium based on the number of affiliates (Bonilla-Chacín and Aguilera 2013).11 In 2012 and after having enrolled 52.6 million individuals, Mexico achieved universal health coverage. 3. Data This study combines an extensive array of publicly available information obtained from different institu- tions, all described in greater detail below. 3.1. School and Academic Performance I measure cognitive attainment with a yearly national standardized test: the National Evaluation of Aca- demic Achievement in Schools (or ENLACE in its acronym in Spanish). Since its implementation in 2006, the test has evaluated the mathematical and verbal abilities of students in grades 3 to 6 in primary ed- ucation and 7 to 9 in lower secondary education. The data are available from Mexico’s Ministry of Education (SEP), with school results disaggregated by grade and subject. This study focuses on the eval- uation of schools in primary education, for which disaggregated results by subject can be obtained for all the grades and years in which the test was implemented (2006–2013).12 The information provided includes test score results in the different subjects under evaluation; the distribution of students falling into different categories of proficiency: inadequate, fair, good, or excellent results; the number of students sitting the test; the number of students considered to have been involved in copying, dictating answers, 7 The self-employed, the underemployed, the unemployed, those detached from the labor market, and their families. 8 The requirements to enroll in SP are proof of residence, Mexican ID, and lack of access to health insurance. 9 The contributions to SP from the states are a subsidy in nature. These are set as a fraction of the total expected cost in health services per capita (which vary by state) adjusted by differentials in regional wages, and capped at a maximum of 30 percent of the total per-family expected cost. 10 Knaul et al. (2006) show that by the end of 2011, only 1 percent of the families were paying the family premium. 11 Previously, the states’ budget for the health system was based on their infrastructure and healthcare personnel in the late 1990s, adjusted for inflation and mortality levels (Bonilla-Chacín and Aguilera 2013). 12 The ENLACE was replaced by another standardized test (PLANEA), which was then canceled in 2019 due to budget constraints. The World Bank Economic Review 441 or other fraudulent practices; and the level of marginalization experienced in the school’s location.13 The evaluation date is scheduled in advance of the start of the academic year, and the test is simultaneously administered to all schools during the national evaluation week (typically towards the end of the school year). These data are complemented with school information held in the Estadística 911 (Statistic 911). The 911 is an administrative questionnaire that all schools in Mexico are required to fill at the beginning and Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 the end of the school year, detailing information on students, teachers, and other school characteristics. With the information provided, I calculate the number of students per teacher, the share of female pupils, and the dropout rate (the proportion of students that left the school throughout the academic year), all for the grades evaluated in ENLACE, and an indicator for whether the head of the school has teaching responsibilities. I also obtained a list of all the schools participating in different educational programs operating during the study period from the Ministry of Education.14 Summary statistics of the variables are shown in table 1. To derive the geolocation of the schools, I use the 2013 school census, provided by INEGI (the National Institute of Statistics and Geography), the Statistics 911, and the ENLACE evaluation. The geographic information that all three sources provide is the state, municipality, and locality code in which the school is located (following the Unique Catalogue of Geostatistical State, Municipal, and Local Areas). With this information, each school is matched to its respective locality.15 , 16 The final sample excludes those schools with inconsistent geographic information and in the top percentile of the share of students considered to have cheated during the test.17 Moreover, I restrict the analysis to those schools observed in all periods, and with 15 or more students evaluated. After applying this filter, the sample consists of 49,751 schools, observed uninterruptedly for eight years. 3.2. Health-Insurance Coverage Administrative records on affiliation to Seguro Popular are provided by the Ministry of Health (SSA), containing the number of affiliates to the program by municipality and quarter.18 I measure the expansion of SP across the country by dividing the number of beneficiaries in a municipality by its population size. Yearly population at the municipality level is calculated assuming linear growth between the two census years of 2005 and 2010, and with projections of municipality population estimated by the National Population Council (CONAPO) after 2010. Figure 1 displays the evolution of the affiliation to SP and its coverage at the national level. It shows that by 2013, the program was covering almost half of the Mexican population, with 55 million beneficiaries. Figure 2 displays the regional expansion of the coverage rate. 13 The census authorities in Mexico create and maintain a marginalization index that reflects the different levels of devel- opment observed throughout the country and at different administration levels. At the smallest regional disaggregation (AGEB or Basic Geostatistical Area), it is calculated with different measures related to education and literacy, access to services, child mortality, and the quality of housing, depending on whether it is an urban or a rural location. 14 The PES (Programa de Escuela Segura) or Safe School Program, the PETC (Programa Escuelas de Tiempo Completo) or Extending School Hours Program, and the PEC (Programa Escuelas de Calidad) or Quality Schools Program. 15 Although the ENLACE evaluation provides information to track the localities in which the schools are based, the infor- mation is not always consistent across all years (in part due to changes in the coding system). Therefore, I prefer to use the school census of 2013 to infer the geographical location of schools, and on the few occasions that this is missing, infer it from the Statistics 911 and the ENLACE evaluation when the codes provided are consistent across all evaluation years. 16 A locality in Mexico refers to the lowest of the three subnational divisions contemplated by the law. 17 Equivalent to excluding those schools where more than 58 percent of students have invalid test results. 18 A municipality in Mexico refers to the second-level administrative division of the country, and it is equivalent to a county in the United States. 442 Font-Gilabert Table 1. Summary Statistics: Sample of Schools (ENLACE) Mean SD Mathematics results Math score 524.36 71.12 Math (% inadequate) 20.38 16.30 Math (% fair) 49.27 14.40 Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 Math (% good+) 30.35 21.63 Verbal results Verbal score 516.35 63.84 Verbal (% inadequate) 19.93 15.56 Verbal (% fair) 49.59 13.96 Verbal (% good+) 30.48 21.24 School characteristics Share of girls 0.49 0.07 Students per teacher 26.51 8.07 Head of school also teaches 0.29 0.46 Students evaluated 136.86 114.42 Dropout rate (pp) 3.41 4.11 Very low marginalization 0.41 0.49 Low marginalization 0.18 0.38 Medium marginalization 0.12 0.33 High marginalization 0.24 0.43 Very high marginalization 0.04 0.21 School programs Safe School 0.17 0.38 Extending School Hours 0.02 0.12 Quality Schools 0.26 0.44 Seguro Popular coverage Seguro Popular coverage rate (pp) 34.33 24.22 Seguro Popular coverage increase (pp) 36.31 17.34 Rainfall shocks Rain shock (total) 0.29 0.45 Rain excess 0.14 0.35 Rain shortage 0.14 0.35 # schools 49,751 # periods (years 2006–2013) 8 Observations 398,008 Source: Author’s own calculations on data from the National Evaluation of Academic Achievement in Schools (ENLACE), the Statistics 911, the State and Municipal Database System, the Ministry of Education, the Ministry of Health, the National Oceanic and Atmospheric Administration, and the census. In the sample of schools, the average coverage rate during the study period is around 34 percent, and the average observed expansion is 36.3 percentage points (standard deviation of 17.3) (table 1). In addition to the coverage rate, I calculate the start date of the program in each municipality. Following previous studies (Bosch and Campos-Vazquez 2014), the quarter of program implementation is defined as when at least 10 individuals enroll in SP.19 With this definition, figs S1.3 and S1.4 in the supplementary 19 The reason is that some of the municipalities, especially at the beginning of the program, show a very low affiliation (zero or close to zero) for several quarters, making it difficult to infer whether the program was operational during that period. The World Bank Economic Review 443 Figure 1. National Affiliation to Seguro Popular Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 Source: Ministry of Health. Note: The coverage rate is defined as the number of affiliates divided by the total population. online appendix display the timing and the pace at which municipalities joined the program. As fig. S1.4 shows, most of the municipalities had already joined SP by 2008. 3.3. Rainfall Shocks I use rainfall data from the National Oceanic and Atmospheric Administration (NOAA). They offer monthly hydrometeorology information from 1950 to 2013 for all of North America in grid cells of approximately 6 km width (1/16◦ ). The data set improves on previously available information in the re- duction of transboundary discontinuities and with an adjustment of orographic precipitations in Mexico (see Livneh et al. (2015) for a more detailed discussion). I measure monthly precipitations at the school- locality level by constructing a linear-distance-weighted rainfall variable using all the data points located within a 20 km radius of each locality centroid. I define the existence of a rainfall shock as when the precipitations gathered in a given locality in the 12 months preceding the academic evaluation are below or above 1 standard deviation from the historical regional mean. With the use of a relative instead of an absolute measure of rainfall I make sure that I am not comparing localities that are more prone to gather higher levels of rainfall with localities that typically receive much less rain. Instead, the measure captures the effect of locality-specific departures from their normal precipitation levels. This definition of rainfall shock has shown to best explain the evolution of agricultural income in Mexico (Adhvaryu et al. 2018; Bobonis 2009). Figure 3 displays the geographical distribution of rainfall shocks with the previous definition for the state of Puebla in 2006 and for localities with at least one school in the final sample. Triangles depict periods when the rainfall gathered in a locality exceeded by 1 standard deviation the 444 Font-Gilabert Figure 2. Geographical Evolution of Seguro Popular Coverage Rate (Percentage) Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 Source: Author’s own calculations on affiliation data from the Ministry of Health, census data (years 2005 and 2010), and projections of municipality population from the National Population Council (CONAPO). Note: The coverage rate is defined as the number of affiliates divided by the population size in each municipality. historical regional rainfall mean (rain excess), crosses represent rainfall levels below 1 standard deviation from the historical records (rain shortage), and grey dots represent stable precipitations. 3.4. MxFLS (Mexican Family Life Survey) To inspect the potential mechanisms by which availability of health insurance might interact with shocks and cognitive attainment I draw on the Mexican Family Life Survey (MxFLS). The MxFLS is a multi- thematic longitudinal household survey representative of the Mexican population at the national, urban, rural, and regional levels, interviewing around 8,400 households in 150 locations (Rubalcava and Teruel 2006, 2013). Relevant to this study, the survey gathers information relating to children’s health, time use, household economic resources, and availability and access to health insurance. I focus the analysis on children aged 6 to 14 during the third wave of the survey (carried out between 2009 and 2011). Summary statistics of the children and their families are shown in table 2. On average, children are 10 years old and have had access to Seguro Popular in their municipality of residence (conditional on eligibility) for 4.6 years (standard deviation of 1.66 years). School enrollment is high (with 96 percent of children attending school), and the incidence of child labor is low (only 3 percent work for pay, 3 percent work in agriculture, and 1 percent work in the family business). On the other hand, the share of children with other household responsibilities, which include domestic chores (56 percent) and caring for elderly or sick members in the household or other children (16 percent), is high. The World Bank Economic Review 445 Figure 3. Pre-exam Locality-Level Rainfall in the State of Puebla (2006) Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 Source: Author’s own calculations on data from the National Evaluation of Academic Achievement in Schools (ENLACE) and the National Oceanic and Atmospheric Administration. Note: Each mark in the map depicts a locality in which there is at least one school evaluated in ENLACE. Rainfall excess is defined as precipitations in the 12 months preceding the test evaluation above 1 standard deviation from the regional historical mean, rainfall shortage is defined as precipitation levels below 1 standard deviation, and stable precipitations as rainfall within 1 standard deviation. 446 Font-Gilabert Table 2. Summary Statistics: Sample of Children (MxFLS) Mean SD Child variables Age 10.11 2.60 Female 0.50 0.50 Indigenous language 0.16 0.36 Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 Sick 0.08 0.27 Attending school 0.96 0.19 Caring for others 0.16 0.37 Household chores 0.56 0.50 Work for pay 0.03 0.16 Work family business 0.01 0.09 Work in agriculture 0.03 0.18 Seguro Popular exposure (years) 4.58 1.66 Rainfall shock 0.29 0.45 Mother’s education No education (mother) 0.14 0.35 Primary (mother) 0.38 0.48 Secondary (mother) 0.31 0.46 High school + (mother) 0.17 0.38 Father’s education No education (father) 0.30 0.46 Primary (father) 0.30 0.46 Secondary (father) 0.23 0.42 High school + (father) 0.17 0.37 Household variables Male hh head 0.78 0.42 Age hh head 44.59 13.07 Married hh head 0.70 0.46 Household size 6.02 2.45 Owns house 0.66 0.47 Tubed water 0.25 0.43 Toilet 0.75 0.43 Cooks with wood or coal 0.39 0.49 Favorable floor material 0.36 0.48 Favorable roof material 0.77 0.42 Rural 0.46 0.50 Observations 5,720 Source: Author’s own calculations on data from the Mexican Family Life Survey (MxFLS), the Ministry of Health, and the National Oceanic and Atmospheric Administration. 3.5. Other Information on health facilities and medical personnel, on the share of eligible individuals at the program start, on pre-program indicators relating to primary education (pass rate and completion rate), and on the marginalization level of municipalities is obtained from SIMBAD (State and Municipal Database System). Also, I compute a measure of regional political alignment with state and municipal election results with data from CIDAC (Development Research Centre). I use these variables to analyze the determinants of the rollout of the SP health-insurance program across the country, discussed in greater detail in the Robustness section. 4. Empirical Strategy To identify the extent to which health insurance can mitigate the impact of rainfall shocks on children’s cognitive performance I exploit rainfall disturbances in the school locality and the expansion of Seguro The World Bank Economic Review 447 Popular (SP) across municipalities. Using the share of the population covered at a given point in time in a municipality I estimate the following equation: yslmt = β1 Rlmt + β2 SPmt + β3 SPmt Rlmt + α Zslmt + ζ Xmt + δt μr + as + slmt , (1) where yslmt are the evaluation results of primary school s in locality l of municipality m during the school Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 year t, Rlmt is a rainfall shock dummy that equals 1 when precipitation gathered in the school locality during the 12 months preceding the academic evaluation is above or below 1 standard deviation from the regional historical mean, SPmt is the coverage rate of Seguro Popular in municipality m measured at the end of the year in which the academic year started,20 and SPmt Rlmt is the interaction of the two terms. I use the intensive instead of the extensive margin in health coverage because the school test-scores data are only available from the academic year 2005/06, and SP rollout began in year 2002. While I observe most of the expansion in the SP coverage rate, there are no pre-SP school data for most schools. The equa- tion also includes a vector of school characteristics Zslmt to control for the ratio of students per teacher, the share of girls, whether the school principal has teaching duties, the marginalization level of the school area,21 three dummy variables indicating whether the school participates in educational programs in year t (i.e., Safe School, Extending School Hours, or the Quality School program), and the share of students marked as carrying fraudulent practices during the test. The variable Xmt is a vector of covariates includ- ing population size,22 the homicide rate, and the transfers per capita from the Oportunidades/Progresa program23 at the municipality level. The regression further includes state-year fixed effects δ t μr to account for yearly disturbances common to all schools in a given state, and school fixed effects as , which capture time-invariant characteristics of the school, its location, and the environment. The coefficients of interest are β 1 and β 3 : the impact of rainfall shocks on school performance and the capacity of social healthcare to mitigate this effect, respectively. While the arrival of SP was quasi- exogenous at the municipality level (Bosch and Campos-Vazquez 2014; Conti and Ginja 2023), the mu- nicipality coverage rate (which depends on how many individuals enroll) is potentially not. The inclusion of SPmt in the regression, therefore, controls for the potential non-random take-up of the program. In other words, years with stable precipitation control for non-random SP expansion. The variable β 1 is identified through the exogenous variation in rainfall shocks, while β 3 , the mitigation effect, is identified by the combination of rainfall shocks and the variation in the health-insurance coverage induced by the program expansion. A concern for the identification of β 3 is whether there exists heterogeneity in how communities cope with rainfall shocks that is correlated with the take-up of SP. I show that the main driver of SP expansion is the rate of eligible individuals, and so the inclusion of school and state-year fixed effects captures the majority of the relevant heterogeneity in program expansion and capacity to mitigate shocks. I also show that (once controlling for school and state-year fixed effects) the evolution of SP coverage is orthogonal to the experience of rainfall shocks. In addition, I estimate the regression equa- tion for different subgroups that are likely to be similar in the characteristics affecting both enrollment to SP and rainfall mitigation capacity. The robustness section explores the determinants of the timing and expansion of SP and conducts some tests to inspect the evolution of the SP coverage rate. In addi- tion, table S1.1 in the supplementary online appendix provides evidence of the exogeneity of shocks, as 20 For example, for the academic year 2005/06, the healthcare coverage rate used is the one observed at the end of 2005. 21 In five categories: very low, low, medium, high, and very high marginalization. 22 Divided into seven categories: (a) less than 5 K, (b) between 5 K and 20 K, (c) between 20 K and 50 K, (d) between 50 K and 100 K, (e) between 100 K and 200 K, (f) between 200 K and 500 K, and (g) higher than 500 K. 23 The Mexican conditional cash transfer program for education. 448 Font-Gilabert Table 3. Test Score Results (1) (2) (3) (4) Score (SD) Inadequate (pp) Fair (pp) Good+ (pp) Panel A: Mathematics Rainfall shock −0.022*** 0.647*** −0.253* −0.394** (0.008) (0.148) (0.139) (0.171) Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 Rainfall shock × SP 0.007*** −0.206*** 0.091*** 0.115*** (0.002) (0.033) (0.034) (0.036) Panel B: Verbal Rainfall shock −0.020*** 0.482*** −0.176 −0.307* (0.007) (0.129) (0.151) (0.157) Rainfall shock × SP 0.006*** −0.149*** 0.058* 0.091*** (0.001) (0.027) (0.033) (0.033) School FE Yes Yes Yes Yes State-year FE Yes Yes Yes Yes Controls Yes Yes Yes Yes Observations 398,008 398,008 398,008 398,008 # schools 49,751 49,751 49,751 49,751 Source: Author’s own calculations. Note: Rainfall shock is a dummy variable that equals 1 when the precipitations gathered in the school locality during the 12 months preceding the evaluation date of National Evaluation of Academic Achievement in Schools (ENLACE) are above or below 1 standard deviation from the regional historical mean (since 1950). SP (coverage rate) is calculated by dividing the number of affiliates to Seguro Popular in the municipality by its population size and scaled up by a factor of 10, so that a value of 10 represents full coverage. Rainfall shock × SP is the interaction of the two terms. The share of girls is defined from 0 to 1, and Safe School, Extending School Hours, and Quality Schools programs are dummy variables indicating whether the school participates in any of these programs. Controls include five dummies of marginalization level of the school, the share of students with unreliable test results, seven dummies of municipality size, municipality per capita expenses in the Progresa/Oportunidades program, and the municipality homicide rate. Robust standard errors clustered at the municipality level in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01. locality-level rainfall shocks are not systematically correlated with pre-SP locality characteristics derived from census data for the years under study.24 5. Results 5.1. Basic Specification Does social health insurance mitigate the effect of negative rainfall shocks on cognitive attainment? The results of estimating equation (1) are shown in table 3. Column (1) displays the results with test scores as the dependent variable, while Columns (2) to (4) show estimates of the effect on the distribution of test achievement. In Panel A, Column (1) shows that students experiencing a negative rainfall shock during the academic year score 0.022 standard deviations lower in the mathematics test, and this reduction is significant at the 1 percent level. However, an increase of 10 percentage points in the health coverage rate mitigates the negative effect by 0.007 standard deviations (32 percent of the effect). A closer look at the distribution of test results shows that the share of students failing the evaluation (inadequate performance) increases by 0.65 percentage points in the event of a rainfall shock (Column (2)), with a 10 percentage- point increase in health coverage reducing the effect by 0.21 percentage points (both estimates significant at the 1 percent). Panel B shows that the experience of a rainfall shock has a smaller impact in the verbal section of the evaluation (−0.020 standard deviations, Column (1)), with a 10 percentage-point increase in the Seguro Popular (SP) coverage rate mitigating by 0.006 standard deviations the negative effect (both magnitudes significant at the 1 percent level). 24 Slightly different are the results observed for the distribution of rainfall shocks in 2013, where it would appear that rural and less illiterate communities were more disproportionately affected. The results of the study remain unchanged if excluding the year 2013 from the analysis. The World Bank Economic Review 449 Table 4. Test Score Results: Marginalized Schools (1) (2) (3) (4) Score (SD) Inadequate (pp) Fair (pp) Good+ (pp) Panel A: Mathematics Rainfall shock −0.022** 0.615*** −0.442** −0.173 (0.010) (0.215) (0.217) (0.216) Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 Rainfall shock × SP 0.006*** −0.187*** 0.119*** 0.068* (0.002) (0.039) (0.044) (0.040) Panel B: Verbal Rainfall shock −0.020** 0.454** −0.411** −0.042 (0.010) (0.200) (0.205) (0.198) Rainfall shock × SP 0.006*** −0.140*** 0.098** 0.042 (0.002) (0.036) (0.041) (0.038) School FE Yes Yes Yes Yes State-year FE Yes Yes Yes Yes Controls Yes Yes Yes Yes Observations 185,112 185,112 185,112 185,112 # schools 23,139 23,139 23,139 23,139 Source: Author’s own calculations. Note: Marginalized schools are those established in a locality considered to be experiencing some degree of marginalization (medium, high, or very high) according to the National Population Council (CONAPO). Rainfall shock is a dummy variable that equals 1 when the precipitations gathered in the school locality during the 12 months preceding the evaluation date of National Evaluation of Academic Achievement in Schools (ENLACE) are above or below 1 standard deviation from the regional historical mean (since 1950). Rainfall shock × SP is the interaction term of the rainfall shock with the SP coverage rate (which is calculated by dividing the number of affiliates to Seguro Popular in the municipality by its population size, and scaled up by a factor of 10 so that a value of 10 represents full coverage). Controls include the SP coverage rate, the number of students per teacher, the number of students per group, the share of girls, whether the head of the school also teaches, whether the school participates in the Safe School, Extending School Hours, or Quality Schools programs, the share of students with unreliable test results, seven dummies of municipality size, municipality per capita expenses in the Progresa/Oportunidades program, and the municipality homicide rate. Robust standard errors clustered at the municipality level in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01. 5.2. Regional Disparities Although disturbances in precipitation levels could impact students’ productivity in school in many ways, the effect of rainfall in disrupting performance may vary across areas with different levels of development and infrastructure. To assess whether there is regional heterogeneity in the impact of shocks on cognitive achievement, I divide schools by the level of marginalization of the area in which they are located.25 When schools in marginalized areas experience a negative rainfall shock, students’ achievement scores in mathematics drop by 0.022 standard deviations (Column (1) of table 4). However, each 10 percentage- point increase in the health coverage rate absorbs 27 percent of the negative effect (significant at the 1 percent level). This reduction is of 0.020 standard deviations in the verbal section, with a 10 percentage- point increase in the coverage rate mitigating 30 percent of the effect. On the other hand, rainfall shocks have no significant effect on test scores in non-marginalized areas (Column (1) of table 5). I also differentiate the effects between rural localities, small urban localities with fewer than 50,000 inhabitants, and large urban localities with more than 50,000 inhabitants. The results show that while rainfall shocks negatively affect mathematics learning in rural areas (table S1.2), they pose no statistically significant reduction in test performance in urban areas irrespective of their population size (tables S1.3 and S1.4). The estimated results in the verbal section of the national evaluation are similar. For instance, rainfall shocks in rural locations increase the verbal failure rate by 0.43 percentage points, with health insurance mitigating by 0.15 percentage points the negative effect per each 10 percentage-point increase in the health coverage rate (Column (2) of table S1.2). 25 I consider a school to be marginalized if it is established in a locality considered to be experiencing some degree of marginalization (medium, high, or very high) according to the National Population Council (CONAPO). 450 Font-Gilabert Table 5. Test Scores Results: Non-marginalized Schools (1) (2) (3) (4) Score (SD) Inadequate (pp) Fair (pp) Good+ (pp) Panel A: Mathematics Rainfall shock −0.012 0.201* 0.108 −0.309 (0.009) (0.117) (0.134) (0.202) Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 Rainfall shock × SP 0.004 −0.033 −0.079** 0.111* (0.003) (0.034) (0.039) (0.057) Panel B: Verbal Rainfall shock −0.012 0.195* 0.093 −0.288 (0.009) (0.110) (0.118) (0.181) Rainfall shock × SP 0.004 −0.029 −0.062* 0.091* (0.003) (0.033) (0.036) (0.053) School FE Yes Yes Yes Yes State-year FE Yes Yes Yes Yes Controls Yes Yes Yes Yes Observations 212,896 212,896 212,896 212,896 # schools 26,612 26,612 26,612 26,612 Source: Author’s own calculations. Note: Non-marginalized schools are those established in a locality considered to be experiencing a low degree of marginalization (low or very low) according to the National Population Council (CONAPO). Rainfall shock is a dummy variable that equals 1 when the precipitations gathered in the school locality during the 12 months preceding the evaluation date of National Evaluation of Academic Achievement in Schools (ENLACE) are above or below 1 standard deviation from the regional historical mean (since 1950). Rainfall shock × SP is the interaction term of the rainfall shock with the SP coverage rate (which is calculated by dividing the number of affiliates to Seguro Popular in the municipality by its population size, and scaled up by a factor of 10 so that a value of 10 represents full coverage). Controls include the SP coverage rate, the number of students per teacher, the number of students per group, the share of girls, whether the head of the school also teaches, whether the school participates in the Safe School, Extending School Hours, or Quality Schools programs, the share of students with unreliable test results, seven dummies of municipality size, municipality per capita expenses in the Progresa/Oportunidades program, and the municipality homicide rate. Robust standard errors clustered at the municipality level in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01. These results point out that rainfall shocks and health insurance have significant differential effects depending on the region’s characteristics. In rural areas, where precipitations are more closely linked to income generation through agricultural production (or in more marginalized areas, where there is lower infrastructure and the population is more vulnerable to shocks), the experience of atypical rainfall may result in higher stress levels for families and children. Indeed, Mexico is considered an arid or semi-arid country, and according to the National Agricultural Survey26 of 2017, the share of rain-fed agriculture in Mexico amounts to 79 percent of the total cultivated area. In urban areas, on the other hand, rainfall disturbances might not be the best measure to capture shocks (either health or income shocks) to children and their families, and the benefits from SP are less likely to be linked to its ability to build resilience against rainfall shocks. 6. Robustness As noted earlier, the rolling out and take-up of Seguro Popular might not have been random. The expan- sion of the program gave priority to states and municipalities with (a) low social-security coverage, (b) a larger number of uninsured individuals in the first six deciles of income, (c) capacity to offer the services granted, (d) a higher pool of potential affiliates, and (e) an explicit request from the state authorities, all subject to available financial resources.27 In this section I first assess the determinants of the tim- ing of SP implementation following Bosch and Campos-Vazquez (2014) and Conti and Ginja (2023) by 26 Encuesta Nacional Agropecuaria, carried out by the National Institute of Statistics and Geography (INEGI). 27 Diario oficial Viernes 04 de Julio de 2003: Acuerdo por el que la Secretaría de Salud da a conocer las reglas de operación e indicadores de gestión y evaluación del Programa Salud para Todos (Seguro Popular de Salud). The World Bank Economic Review 451 estimating the following equation: Quarterms = θ Xms + μs + ms , (2) where Quarterms is the quarter of implementation of SP in municipality m of state s, Xms is a series of sociodemographic, political, healthcare, and primary education indicators measured before the program start, and ms is the error term. The regression includes state fixed effects μs , as the timing in which the Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 states were offering the new health scheme was negotiated with the federal government. On the other hand, it was less clear which municipalities were to receive the program first. Therefore, I study the determinants of the program rollout within states but also estimate the model without state fixed effects for comparison. While I do not have information on test results before the start of the program, I measure municipality pre-program trends in education with the evolution of the primary completion rate, and with the pass rate of the grades evaluated in ENLACE. Table S1.5 displays the results of estimating equation (2). Columns (2) and (4) show the model esti- mates with state fixed effects. Municipalities with greater population size, with more eligible population, and with more medical personnel received the program first. Political alignment is also a good predictor of program implementation, as the occurrence of same political party in both the state and municipal gov- ernment predicts the implementation of SP one and a half quarters earlier than in municipalities without political alignment. Moreover, the evolution in the primary completion rate and in the pass rate of the grades evaluated in ENLACE in the five years preceding the program start is not significantly correlated with the timing of the program implementation in any of the specifications. I further inspect the determinants of the SP coverage-rate expansion after it is implemented in a mu- nicipality. Table S1.6 shows the results of estimating a variation of equation (2), where the dependent variable quarter of implementation is replaced with the increase in the coverage rate in the first, second, and third years after SP implementation. Columns (1) and (2) show that, once SP is implemented, the share of eligible individuals is the main determinant of its expansion. The coverage rate increases by 33 percentage points in the first year in a municipality where everybody is eligible. Higher marginalization, which is closely linked to eligibility, also explains higher program enrollment. Health infrastructure mea- sured as doctors per eligible population also predicts a small but significant higher coverage expansion in the first year. In the second year (Columns (3) and (4)), only eligibility and population size are able to ex- plain enrollment, and in the third year, only eligibility remains significant (Columns (5) and (6)). Political alignment and the evolution of indicators in primary education do not predict coverage expansion. Next, I test the robustness of the main results to various specifications. Results for the mathematics test are displayed in table 6, while table 7 shows the results in the verbal evaluation. I also show the es- timate of β 2 from equation (1) to examine how its estimated value change across specifications. Column (1) shows the coefficient estimates from the main specification (equation (1)). In Column (2) I add an interaction term of the presence of a rainfall shock with municipality expenses per capita on the Oportu- nidades/Progresa program. This interaction allows testing whether the estimated shock-mitigating effect from the expansion of health insurance partly reflects the mitigating effect of cash transfers in Mexico. Results do not change. Column (3) controls for the political alignment defined as same political party in the state and municipal governments. This specification accounts for the possibility that the political environment could be affecting the level of resources (including higher expenses on both education and health) in the different municipalities. The results are practically identical. In Column (4) I include all the pre-program municipality characteristics correlated with the rollout of Seguro Popular (except for the share of eligible individuals) interacted with a linear trend (see tables S1.5 and S1.6). Notice that this is a demanding test, as the information on test scores is only available from 2006 onward, and the expansion of SP could have already affected the evolution of educational achievement. The point estimates reduce in magnitude. The effect of a rainfall shock on the mathematics test scores reduces from −0.022 to −0.017 standard deviations, and the mitigating effect from 0.007 to 0.006 standard deviations (Column (4) of 452 Table 6. Test Results in Mathematics: Robustness Checks (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Math score Math score Math score Math score Math score Math score Math score Math score Math score Math score SD SD SD SD SD SD SD SD SD SD Rainfall shock −0.022*** −0.022*** −0.024*** −0.017** −0.014** — −0.023*** −0.022* −0.022*** −0.028*** (0.008) (0.008) (0.008) (0.007) (0.007) (0.008) (0.013) (0.008) (0.008) SP (coverage rate) 0.020*** 0.020*** 0.021*** 0.001 −0.005 0.014*** 0.019*** 0.020*** 0.018*** 0.020*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.003) (0.004) (0.004) Rainfall shock × SP 0.007*** 0.007*** 0.007*** 0.006*** 0.005*** – 0.007*** 0.007** 0.007*** 0.008*** (0.002) (0.002) (0.002) (0.002) (0.001) (0.002) (0.003) (0.002) (0.002) Rainfall shock × Op – −0.000 – – – – – – – – (0.001) Lead rainfall shock – – – – – −0.004 – – – – (0.008) Lead rainfall shock × SP – – – – – 0.002 – – – – (0.002) Lagged rainfall shock – – – – – – −0.010 – – – (0.007) Lagged rainfall shock × SP – – – – – – 0.004** – – – (0.002) School FE Yes Yes Yes Yes Yes Yes Yes No Yes Yes State-year FE Yes Yes Yes Yes Yes Yes Yes No Yes Yes Basic controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Political alignment No No Yes Yes Yes No No No No No Pre-SP mun. charac. × trends No No No Yes Yes No No No No No Pre-SP eligibility × trends No No No No Yes No No No No No Spatial corrected standard errors No No No No No No No Yes No No Observations 398,008 398,008 398,008 398,008 398,008 348,257 398,008 398,008 375,256 396,800 # schools 49,751 49,751 49,751 49,751 49,751 49,751 49,751 49,751 46,907 49,600 Source: Author’s own calculations. Note: Rainfall shock is a dummy variable that equals 1 when the precipitations gathered in the school locality during the 12 months preceding the evaluation date of ENLACE are above or below 1 standard deviation from the regional historical mean (since 1950). SP (coverage rate) is calculated by dividing the number of affiliates to Seguro Popular in the municipality by its population size and scaled up by a factor of 10 so that a value of 10 represents full coverage. Rainfall shock × SP is the interaction of the two terms. Rainfall shock × Op is the interaction of the rainfall shock with the municipality per capita expenses (in 1,000 pesos) in the Progresa/Oportunidades program. Controls include the number of students per teacher, the number of students per group, the share of girls, whether the head of the school also teaches, whether the school participates in the Safe School, Extending School Hours, or Quality Schools programs, five dummies of marginalization level of the school, the share of students with unreliable test results, seven dummies of municipality size, municipality per capita expenses in the Progresa/Oportunidades program, and the municipality homicide rate. Column (1) displays the main coefficient estimates. Column (2) includes the interaction term of rainfall shock with the municipality expenses per capita in the Oportunidades/Progresa program. Column (3) adds a control for political alignment. In Column (4) I further interact pre-program municipality characteristics correlated with the rollout of Seguro Popular with trends, except for the share of eligible individuals (see table S1.5 in the supplementary online appendix). Column (5) further includes the share of eligible individuals interacted with a trend. Column (6) conducts a placebo test of the effect of future rainfall shocks on current test outcomes. Column (7) includes one lag of the rainfall shock and its interaction with the SP coverage rate. Column (8) displays Conley standard spatial errors calculated using a radius of 200 km around each locality centroid. Column (9) excludes those localities in which there is no variation in rainfall shocks (either never or always experienced a rainfall shock). Column (10) replaces school-locality-level rainfall shocks with shocks measured at the municipality level. Except in Column (8), robust standard errors clustered at the municipality level in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01. Font-Gilabert Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 Table 7. Test Results in Verbal: Robustness Checks (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Verbal score Verbal score Verbal score Verbal score Verbal score Verbal score Verbal score Verbal score Verbal score Verbal score SD SD SD SD SD SD SD SD SD SD Rainfall shock −0.020*** −0.020*** −0.022*** −0.016** −0.014** – −0.020*** −0.020* −0.019*** −0.026*** (0.007) (0.007) (0.007) (0.006) (0.006) (0.007) (0.010) (0.007) (0.007) SP (coverage rate) 0.013*** 0.013*** 0.014*** 0.000 −0.004 0.008** 0.012*** 0.013*** 0.011*** 0.013*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.003) (0.004) (0.003) (0.004) (0.004) Rainfall shock × SP 0.006*** 0.006*** 0.006*** 0.005*** 0.004*** – 0.006*** 0.006** 0.006*** 0.007*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.001) (0.002) The World Bank Economic Review Rainfall shock × Op – −0.000 – – – – – – – – (0.001) Lead rainfall shock – – – – – −0.000 – – – – (0.007) Lead rainfall shock × SP – – – – – 0.001 – – – – (0.002) Lagged rainfall shock – – – – – −0.009 – – – (0.006) Lagged rainfall shock × SP – – – – – – 0.004** – – – (0.002) School FE Yes Yes Yes Yes Yes Yes Yes No Yes Yes State-year FE Yes Yes Yes Yes Yes Yes Yes No Yes Yes Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Political alignment No No Yes Yes Yes No No No No No Pre-SP mun. charac. × trends No No No Yes Yes No No No No No Pre-SP eligibility × trends No No No No Yes No No No No No Spatial corrected standard errors No No No No No No No Yes No No Observations 398,008 398,008 398,008 398,008 398,008 348,257 398,008 398,008 375,256 396,800 # schools 49,751 49,751 49,751 49,751 49,751 49,751 49,751 49,751 46,907 49,600 Source: Author’s own calculations. Note: Rainfall shock is a dummy variable that equals 1 when the precipitations gathered in the school locality during the 12 months preceding the evaluation date of ENLACE are above or below 1 standard deviation from the regional historical mean (since 1950). SP (coverage rate) is calculated by dividing the number of affiliates to Seguro Popular in the municipality by its population size and scaled up by a factor of 10 so that a value of 10 represents full coverage. Rainfall shock × SP is the interaction of the two terms. Rainfall shock × Op is the interaction of the rainfall shock with the municipality per capita expenses (in 1,000 pesos) in the Progresa/Oportunidades program. Controls include the number of students per teacher, the number of students per group, the share of girls, whether the head of the school also teaches, whether the school participates in the Safe School, Extending School Hours, or Quality Schools programs, five dummies of marginalization level of the school, the share of students with unreliable test results, seven dummies of municipality size, municipality per capita expenses in the Progresa/Oportunidades program, and the municipality homicide rate. Column (1) displays the main coefficient estimates. Column (2) includes the interaction term of rainfall shock with the municipality expenses per capita in the Oportunidades/Progresa program. Column (3) adds a control for political alignment. In Column (4) I further interact pre-program municipality characteristics correlated with the rollout of Seguro Popular with trends, except for the share of eligible individuals (see table S1.5 in the supplementary online appendix). Column (5) further includes the share of eligible individuals interacted with a trend. Column (6) conducts a placebo test of the effect of future rainfall shocks on current test outcomes. Column (7) includes one lag of the rainfall shock and its interaction with the SP coverage rate. Column (8) displays Conley standard spatial errors calculated using a radius of 200 km around each locality centroid. Column (9) excludes those localities in which there is no variation in rainfall shocks (either never or always experienced a rainfall shock). Column (10) replaces school-locality-level rainfall shocks with shocks measured at the municipality level. Except in Column (8), robust standard errors clustered at the municipality level in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01. 453 Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 454 Font-Gilabert table 6). Moreover, the correlation between SP expansion and test scores during stable precipitations becomes null, suggesting that health insurance impacted cognitive attainment only through its abil- ity to mitigate the negative effect of shocks on students’ performance in school. Column (5) further includes the share of eligible individuals at the program start interacted with a trend. This specifica- tion produces the lowest point estimates, as the coverage rate of Seguro Popular is highly correlated with population eligibility, and the program is suspected to have the largest effect in regions with a Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 higher proportion of eligible individuals. Even then, the shock-mitigating effect of SP on mathemat- ics test scores is estimated to be of 0.005 standard deviations per each 10 percentage-point increase in the coverage rate (Column (5) of table 6), and of 0.004 standard deviations on the verbal results (Column (5) of table 7), both magnitudes significant at the 1 percent level. Column (6) displays the results of a placebo test that consists of interacting future rainfall shocks with the healthcare coverage rate and shows that future rainfall does not have a significant effect on current test scores. Column (7) includes one lag of the rainfall shock and shows that the effect on test scores is mainly driven by contemporaneous disturbances. However, there is a lasting protective effect on current test scores from health coverage dur- ing past negative shocks. In the base specification, I cluster the standard errors at the municipality level. Column (8) shows standard errors adjusted for spatial correlation with the method developed in Conley (1999), and using a radius of 200 km around each locality centroid to define areas independent of admin- istrative boundaries. Column (9) excludes those localities in which there is no variation in rainfall shocks (either always or never experienced a rainfall shock), and Column (10) replaces school-locality-level rain- fall shocks with shocks measured at the municipality level. Rainfall shocks measured at the municipality level have a larger effect on school achievement, and healthcare coverage has a lower mitigating effect. However, this likely reflects the impact of a larger shock in absolute terms, as average precipitations are more stable when computed over a broader area. The results discussed above still hold. Another concern involves endogenous migration patterns (or children leaving the school more gener- ally). If rainfall shocks affect migration decisions of families and family characteristics are related to both migration decisions and child characteristics, the effect of rainfall shocks on school test scores could be biased. Using the Statistics 911 I create an indicator for the ratio of children that did not complete the academic year in the school in which they started it (the share of students that drop out), and inspect whether this indicator is related to the experience of rainfall shocks or the interaction of rainfall with SP expansion. Column (1) of table S1.7 shows that the probability of students dropping out from the school is not associated with the experience of a rainfall shock in the locality nor with the interaction of rainfall with SP expansion in the event of a rainfall sock. To rule out any additional compositional bias arising from negative shocks in the locality affecting the type of students that sit the academic evaluation, I test whether the number of students evaluated in each school is affected by the experience of a rainfall shock. The results of this test, displayed in Column (2) of table S1.7, show that neither rainfall shocks nor the interaction of rainfall with the expansion of SP have a significant effect on the number of students evaluated. Moreover, in Column (3) I assess whether rainfall shocks affect SP expansion, with the results showing that rainfall shocks do not impact enrollment behavior. 7. Mechanisms This section inspects potential channels that could help explain why rainfall shocks negatively affect children’s performance in school, and the role of access to health insurance in mitigating the effects. To do so, I move from school-level data to individual- and household-level data, described in greater detail in the Data section. I assess the impact of rainfall shocks and access to SP among children aged 6 to 14 and their families, and who were interviewed during the third wave of the Mexican Family Life Survey (between 2009 and 2012). The World Bank Economic Review 455 In the survey data it is possible to observe whether a household is enrolled in SP. However, I prefer to focus on eligibility rather than affiliation to avoid a potential self-selection bias. To infer whether a household is eligible for SP, I inspect their availability and access to formal health insurance (in which case the household is deemed ineligible). In the survey, individuals report all the different health-insurance schemes they benefit from, which include insurance from the social security: IMSS, ISSSTE, PEMEX, and other minor schemes; and other private plans (either privately purchased or offered by their employer). Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 As long as one household member has access to any form of formal health insurance, this extends to the rest of the family, and I define such a household as ineligible for Seguro Popular. All other households in which none of the members have access to formal health insurance are deemed eligible for SP (48 percent of all households in the sample). Combining eligibility and the rollout date of SP I construct a measure of access to social health insur- ance that exploits individual-level variation: years of exposure to Seguro Popular. The number of years a child had access to SP depends on the child’s age and the introduction date of the program in the child’s municipality of residence (subject to eligibility). I prefer to use years of exposure to SP instead of the cov- erage rate at the municipality to (a) obtain more natural estimates of potential differences of the impact of rainfall shocks and access to social health insurance between eligible and non-eligible households and 9b) circumvent the fact that as the program reached full expansion, the coverage rate was closer to the eligibility rate. I also estimate a model of household fixed effects to assess the impact of rainfall shocks and health insurance on household consumption by including the consumption information available in the previous survey (years 2005–2006). Notice that at the household level, exposure to SP only varies by its introduction date in the municipality of residence. Moreover, rainfall shocks are now measured at the municipality level, as opposed to shocks at the locality level, as the latter information is deemed confidential and it is not disclosed. 7.1. Specification To capture the impact of rainfall shocks on educational inputs and any potential mitigating effect arising from access to health insurance, I estimate the following equation: yim = β1 Rm + β2 SP(years)im + β3 SP(years)im Rm + ζ Xim + δt + μz + im , (3) where yim are indicators of the health and time use of children, Rm is a rainfall shock dummy reflect- ing whether the precipitations gathered in the municipality of residence in the 12 months preceding the interview date were above or below 1 standard deviation from the historical regional mean, SP(years)im controls for the number of years a child had Seguro Popular available in their municipality (which de- pends on the child’s age and the introduction date of SP in the municipality), and SP(years)im Rm is the interaction of the two terms. Similarly to before, β 1 and β 3 are the coefficients of interest. The variable Xim is a set of children, family, and regional covariates,28 δ t and μz are dummies for month and year of interview respectively, and ms is the error term. Similarly, to capture the effect of rainfall shocks on household’s economic resources I estimate the following equation: log(Cimt ) = β1 Rmt + β2 SP(years)mt + β3 SP(years)mt Rmt + Himt + δt + μz + γi + υimt , (4) 28 The child’s gender and age (categorical dummies), whether the child speaks an indigenous language, attends a public school, and assists an evening shift, the age, gender, and marital status of the household head, the total number of individuals in the household, dummies for mother’s and father’s education (no education, primary school, secondary school, and high school or higher), whether the household owns the house, has piped water inside, or a toilet, whether the household cooks with wood or coal, dummies for the quality of the roof and floor, and type of location (urban or rural). 456 Font-Gilabert where log (Cimt ) is the logarithm of the equivalised household expenditures in non-durable goods,29 SP(years)mt is now defined at the household level (and depends only on the date that SP was introduced in the municipality of residence), Himt is a vector of household characteristics relating to household compo- sition, wealth, and information about the household head,30 δ t and μz are dummies for month and year of interview respectively, γ i are household fixed effects, and υ imt is the error term. Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 7.2. Results: Mechanisms Table 8 shows the results of estimating equation (3) separately by eligibility status (SP eligible: with no formal insurance, and SP ineligible: with access to formal insurance). Robust standard errors are shown in parentheses, while clustered errors at the municipality level are displayed between brackets. In Panel A, Column (2) shows that rainfall shocks increase the probability of children eligible for SP being sick in the four weeks prior to the interview date by 6.6 percentage points. However, the availability of SP in the municipality reduces this probability by 1.5 percentage points per year of exposure. Similarly, rainfall shocks increase the probability of children looking after elderly or sick people, or other children, by 8.8 percentage points (Column (4)) and doing domestic chores by 14.1 percentage points (Column (6)). However, and similarly, the expansion and availability of financial protection in health reduces the demand for children’s time in domestic tasks when hit by rainfall shocks. The differences in the estimates between the specification that includes basic controls (Columns (1), (3), and (5)) and the specification that includes a broader set of controls (Columns (2), (4), and (6)) are small, in line with rainfall shocks being orthogonal to children and households’ characteristics. On the other hand, there are no statistically significant effects on health status and time use for those children in families ineligible for the new health- care scheme (Panel B). Dividing the sample by rural and urban locations generates a similar picture (table S1.8), where the benefits from SP availability in the event of rainfall shocks are mainly concentrated in rural areas (where the share of eligible individuals is higher and adverse weather presumably has more pervasive consequences on children and their families than in urban areas). Finally, I assess whether the experience of negative rainfall shocks affects the economic resources of the household. In the consumption regression (equation (4)), I exclude the households in the top and bottom percentile in equivalised household expenditures. Column (6) of table 9 shows that the experience of a rainfall shock reduces by 16 percent the equivalised household consumption level among households eligible for SP (with no formal insurance). However, each additional year of financial protection in health reduces the negative effect by 4 percent (significant at the 10 percent level). The point estimates of these effects are similar for households in rural areas (Column (10)). In rural areas, a negative rainfall shock reduces household consumption by 18 percent, similar to the reduction estimated in Bobonis (2009) for a sample of rural households in Mexico (16.7 percent). However, an additional year of SP availability reduces this effect by 3 percent. On the other hand, rainfall shocks do not translate into any significant reduction in household consumption among families with access to formal health insurance or living in urban areas (in which case there are no additional benefits from the expansion in health coverage). 29 That is, excluding expenses on electronic appliances, furniture, property, and acquisition of vehicles. I construct the expenditures equivalence scale for Mexican households following Teruel, Rubalcava, and Santana (2005), and assign a factor of 0.77 to children from 0 to 5 years old, 0.80 to children from 6 to 12 years old, 0.74 to children from 13 to 18 years old, and 1 to adults above 18 years of age. 30 The full list of household characteristics is the age, gender, education, and marital status of the household head, the total number of individuals living in the household, the number of children under age 5, the number of individuals between 6 and 10 years of age, between 11 and 18 years old, between 19 and 45 years old, between 46 and 60 years old, and more than 60 years old, whether the households owns the house, has piped water inside, a toilet, whether the household cooks with wood or coal, dummies for the quality of the roof and floor, type of location (urban or rural), an interviewer- reported variable on the accuracy of reported expenditures (dummy for excellent accuracy), and a dummy controlling for whether the household expenditures questionnaire was filled by the same respondent in the different waves. The World Bank Economic Review 457 Table 8. Children’s Health and Time Use by Eligibility to Seguro Popular Sick Caring Chores (1) (2) (3) (4) (5) (6) Panel A: SP eligible Rainfall shock 0.067 0.066 0.098 0.088 0.145 0.141 Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 (0.027)** (0.029)** (0.041)** (0.041)** (0.051)*** (0.052)*** [0.036]* [0.037]* [0.079] [0.073] [0.067]** [0.067]** Rainfall shock × SP(years) −0.016 −0.015 −0.020 −0.018 −0.036 −0.035 (0.006)*** (0.006)** (0.008)** (0.009)** (0.011)*** (0.011)*** [0.007]** [0.007]** [0.015] [0.015] [0.013]*** [0.013]*** Observations 2,936 2,900 2,970 2,933 2,970 2,933 R2 0.021 0.028 0.055 0.080 0.163 0.176 Panel B: Not SP eligible Rainfall shock 0.043 0.031 0.033 0.032 0.043 0.040 (0.035) (0.036) (0.043) (0.045) (0.058) (0.060) [0.042] [0.042] [0.049] [0.049] [0.069] [0.071] Rainfall shock × SP(years) −0.011 −0.008 −0.010 −0.009 −0.019 −0.017 (0.007) (0.007) (0.009) (0.009) (0.012) (0.012) [0.008] [0.008] [0.009] [0.009] [0.014] [0.014] Observations 2,827 2,797 2,859 2,829 2,859 2,829 R2 0.012 0.023 0.043 0.072 0.138 0.144 Basic controls Yes Yes Yes Yes Yes Yes Additional controls No Yes No Yes No Yes Source: Author’s own calculations using data from the Mexican Family Life Survey. Note: A child is eligible for SP if the family does not have any other form of health insurance. Sick is a binary variable equal to 1 if the child stopped doing any of their daily activities due to sickness in the past four weeks. Caring is a dummy variable recording whether the child took care of elderly or sick people and/or other children in the last week. Chores is a dummy variable equal to 1 if the child did domestic chores in the past week. Rainfall shock is a dummy variable that equals 1 when the precipitations gathered in the municipality of residence during the 12 months preceding the interview date were above or below 1 standard deviation from the regional historical mean. Rainfall shock × SP(years) is the interaction term of rainfall shock with the number of years a child had Seguro Popular available. The basic controls are the number of years a child had Seguro Popular available, dummies for year and month of interview, child’s age, and gender. Additional controls include whether the child speaks an indigenous language, four categories of father’s and mother’s education (no education, secondary, and high school or higher), the age, gender, and marital status of the household head, ownership status of dwelling, rural location, whether the household has piped water into the house, a toilet, cooks with wood or coal, and indicators of the quality of the materials of the floor and roof. Robust standard errors in parentheses. Robust standard errors clustered at the municipality level in brackets. * p < 0.1, ** p < 0.05, *** p < 0.01. 8. Conclusion As the world moves closer to achieving the Millennium Developmental Goal of universal primary com- pletion, significant challenges to ensure effective learning in the classroom remain. Poverty and marginal- ization continue to be significant predictors of human capital accumulation among children, and negative shocks experienced during childhood threaten to aggravate the existent inequalities by households’ ability to cope with them. This study shows that a state intervention to reduce inequality in healthcare access protected the cognitive attainment of primary school children in the event of negative shock. This result points towards synergies from public investments in health and education, and from higher returns to educational investments when the ability of families to endure shocks is increased. In this regard, the study shows that the expansion in health coverage mitigated the negative effect of rainfall shocks on children’s health among program-eligible households, reduced the demand for children’s time, and protected household’s consumption from fluctuations accruing from rainfall disturbances. The results add to a new stream of research that investigates whether shocks to human capital during childhood can be mitigated through different policies or interventions, by showing the capacity of universal health coverage in buffering negative environmental shocks. As rainfall disturbances are felt the most in regions with weaker infrastructure and higher dependence on climate, the discouraging evolution of weather patterns is likely to aggravate the existing gap in human 458 Font-Gilabert Table 9. Equivalised Household Expenditures (in logs) All Not SP eligible SP eligible Urban Rural (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Rainfall shock −0.084 −0.089 −0.041 −0.014 −0.129 −0.161 −0.052 −0.053 −0.153 −0.180 (0.041)** (0.041)** (0.058) (0.058) (0.059)** (0.058)*** (0.057) (0.058) (0.068)** (0.068)*** Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 [0.060] [0.059] [0.070] [0.071] [0.099] [0.097]* [0.062] [0.064] [0.075]* [0.068]** Rainfall shock × SP(years) 0.015 0.014 0.002 0.005 0.036 0.040 0.007 −0.001 0.021 0.030 (0.011) (0.012) (0.016) (0.017) (0.018)** (0.018)** (0.015) (0.016) (0.017) (0.017)* [0.014] [0.015] [0.018] [0.019] [0.018] [0.017]* [0.020] [0.020] [0.028] [0.027] Household FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Basic controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Additional controls No Yes No Yes No Yes No Yes No Yes Observations 4,703 4,602 2,388 2,338 2,300 2,249 2,507 2,446 2,196 2,156 R2 0.038 0.076 0.052 0.127 0.060 0.098 0.057 0.106 0.053 0.094 Source: Author’s own calculations using data from the Mexican Family Life Survey. Note: Household expenditures consist of all expenses in non-durable goods. The expenditures equivalence scale used is the one estimated by Teruel, Rubalcava, and Santana (2005). A household is eligible for SP if the family does not have any other form of health insurance. Rainfall shock is a dummy variable that equals 1 when the precipitations gathered in the municipality of residence during the 12 months preceding the interview date were above or below 1 standard deviation from the regional historical mean. Rainfall shock × SP(years) is the interaction term of rainfall shock with the number of years a household had Seguro Popular available. The basic controls are the number of years a household had Seguro Popular available and dummies for year and month of interview. Additional controls include the age, gender, education, and marital status of the household head, household size, and variables of household composition (number of individuals between 6 and 10 years of age, between 11 and 18 years old, between 19 and 45 years old, between 46 and 60 years old, and more than 60 years old), ownership status of dwelling, rural location, whether the household has piped water in the building, a toilet, cooks with wood or coal, quality of the materials of floor and roof, an interviewer-reported variable on the accuracy of reported expenditures (dummy for excellent accuracy), and a dummy controlling for whether the household expenditures were reported by the same respondent in the different waves. Robust standard errors in parenthesis. Robust standard errors clustered at the municipality level in brackets. * p < 0.1, ** p < 0.05, *** p < 0.01. capital by socioeconomic disadvantage. However, estimates show larger positive effects from universal healthcare in more marginalized areas. The results presented here are also relevant amid the growing number of countries expanding healthcare coverage and should be considered when carrying out a cost- benefit analysis of public investments in health. Nevertheless, the extent to which the expansion in social health insurance is accompanied by improvements in healthcare infrastructure, including its efficiency, will determine the capacity of national health systems in protecting individuals and families from financial and other health-related shocks. Data Availability The data underlying this article are available in the article and in its online supplementary material. References Aarabi, M., H. Haghighi, and A. Takian. 2020. “The Role of Universal Health Coverage in Overcoming the COVID-19 Pandemic.” The BMJ Opinion. Accessed February 25, 2020. Adhvaryu, A., P. Bharadwaj, J. Fenske, A. Nyshadham, and R. Stanley. 2019. “Dust and Death: Evidence from the West African Harmattan.” Working Paper 25937, NBER. Adhvaryu, A., A. Nyshadham, T. Molina, and J. Tamayo. 2018. “Helping Children Catch Up: Early Life Shocks and the Progresa Experiment.” Working Paper 24848, NBER. Alcaraz, C., D. Chiquiar, M. J. Orraca, and A. Salcedo. 2016. “The Effect of Publicly Provided Health Insurance on Education Outcomes in Mexico.” World Bank Economic Review 30 (Supplement_1): S145–S156. Bobonis, G. J. 2009. “Is the Allocation of Resources within the Household Efficient? New Evidence from a Randomized Experiment.” Journal of Political Economy 117 (3): 453–503. The World Bank Economic Review 459 Bonilla-Chacín, M. E., and N. Aguilera. 2013. “The Mexican Social Protection System in Health.” UNICO Studies Series 1. Bosch, M., and R. M. Campos-Vazquez. 2014. “The Trade-Offs of Welfare Policies in Labor Markets with Informal Jobs: The Case of the “Seguro Popular” Program in Mexico.” American Economic Journal: Economic Policy 6 (4): 71–99. Cohodes, S., D. Grossman, S. Kleiner, and M. Lovenheim. 2016. “The Effect of Child Health Insurance Access on Downloaded from https://academic.oup.com/wber/article/37/3/437/7068455 by World Bank and IMF user on 14 September 2023 Schooling: Evidence from Public Insurance Expansions.” Journal of Human Resources 51 (3): 727–59. Conley, T. 1999. “GMM Estimation with Cross Sectional Dependence.” Journal of Econometrics 92 (1): 1–45. Conti, G., and R. Ginja. 2023. “Who Benefits from Free Health Insurance: Evidence from Mexico.” Journal of Human Resources 58 (1): 146–82. Dinkelman, T. 2013. “Mitigating Long-Run Health Effects of Drought: Evidence from South Africa.” Working Paper 19756, NBER. Duque, V., M. Rosales-Rueda, and F. Sanchez Torres. 2019. “How Do Early-Life Shocks Interact with Subsequent Human Capital Investments? Evidence from Administrative Data.” Working Papers 2019-17, University of Sydney, School of Economics. Field, C. B., V. Barros, T. F. Stocker, and Q. Dahe. 2012. Managing the Risks of Extreme Events and Disasters to Ad- vance Climate Change Adaptation: Special Report of the Intergovernmental Panel on Climate Change. Cambridge University Press. Frenk, J., O. Gomez-Dantes, and F. Knaul. 2009. “The Democratization of Health in Mexico: Financial Innovations for Universal Coverage.” Bulletin of the World Health Organisation 87 (7): 542–48. Garg, T., M. Jagnani, and V. P. Taraz. 2017. “Human Capital Costs of Climate Change: Evidence from Test Scores in India.” 2017 Annual Meeting, July 30-August 1, Chicago, Illinois 258018, Agricultural and Applied Economics Association. Gunnsteinsson, S., T. Molina, A. Adhvaryu, P. Christian, A. Labrique, J. Sugimoto, A. A. Shamim, and K. P. West, Jr. 2022. “Protecting Infants from Natural Disasters: The Case of Vitamin A Supplementation and a Tornado in Bangladesh”, Journal of Development Economics 158: 102914. Knaul, F. M., H. Arreola-Ornelas, O. Méndez-Carniado, C. Bryson-Cahn, J. Barofsky, R. Maguire, and M. Miranda et al. 2006. “Evidence Is Good for Your Health System: Policy Reform to Remedy Catastrophic and Impoverishing Health Spending in Mexico.” Lancet 368 (9549): 1828–41. Knaul, F. M., E. González-Pier, O. Gomez-Dantes, D. García-Junco, H. Arreola-Ornelas, M. Barraza-Lloréns, and R. Sandoval et al. 2012. “The Quest for Universal Health Coverage: Achieving Social Protection for All in Mexico.” Lancet 380 (9849): 1259–79. Kurowski, C., M. Vilar-Compte, E. C. Araujo, and A. Stanciole. 2012. “Mexico Social Protection System in Health and the Formal Sector.” Other Health Study 76736, The World Bank. Levine, P. B., and D. Schanzenbach. 2009. “The Impact of Children’s Public Health Insurance Expansions on Educa- tional Outcomes.” In Forum for Health Economics & Policy, Volume 12 (1). De Gruyter. Liu, K. 2016. “Insuring against Health Shocks: Health Insurance and Household Choices.” Journal of Health Eco- nomics 46 (March): 16–32. Livneh, B., T. J. Bohn, D. W. Pierce, F. Munoz-Arriola, B. Nijssen, R. Vose, and D. R. Cayan et al. 2015. “A Spatially Comprehensive, Hydrometeorological Data Set for Mexico, the U.S., and Southern Canada 1950–2013.” Scientific Data 2: 150042. https://www.nature.com/articles/sdata201542#citeas McEwan, P. J. 2015. “Improving Learning in Primary Schools of Developing Countries: A Meta-Analysis of Random- ized Experiments.” Review of Educational Research 85 (3): 353–94. Park, R. J., J. Goodman, M. Hurwitz, and J. Smith. 2020. “Heat and Learning.” American Economic Journal: Economic Policy 12 (2): 306–39. Rubalcava, L., and G. Teruel. 2006. “Mexican Family Life Survey, Second Round.” Technical report. Working Paper, www.ennvih-mxfls.org. ———. 2013. Mexican Family Life Survey, Third Round. Working Paper. Teruel, G., L. Rubalcava, and A. Santana. 2005. “Escalas de Equivalencia para México.” Documentos de Investigación 23. WHO. 2000. “The World Health Report 2000 - Health Systems: Improving Performance.” Technical report. WHO and World Bank. 2017. “Tracking Universal Health Coverage: 2017 Global Monitoring Report.” Technical Report. ———. 2020. “Global Monitoring Report on Financial Protection in Health 2019.” Technical Report.