ABSTRACT
This work examines whether additional time in elementary and secondary school affects economic well-being in adulthood. We leverage cross-municipality and cross-cohort variation in a reform that increased the Chilean school day by 30 percent between 1997 and 2010 and find that full-day schooling increased educational attainment, delayed childbearing, and increased earnings when students reached young adulthood. These findings are consistent with a human capital channel and demonstrate that large-scale investments in public education can yield long-term economic benefits.
I. Introduction
Policymakers frequently advocate lengthening the school day in order to promote economic growth and competitiveness, yet the relationship between the length of the school day and economic outcomes has not been fully explored.1 Despite this lack of empirical evidence, many countries have drastically lengthened the school day over the past 30 years. In the United States, the share of kindergarteners who attend a full-day of school increased from less than 20 percent in 1970 to 75 percent by 2012 (Gibbs 2014). More broadly, until the 1990s, the typical student attended school for approximately four hours a day in many middle-income and developing countries. While some of these countries have moved towards a six- to seven-hour school day, others continue to operate under the half-day model (UNESCO TERCE 2016).
We examine whether additional time in school translates into improved economic well-being in adulthood by analyzing one of the first and largest full-day schooling reforms, Chile’s Jornada Escolar Completa (JEC). The JEC increased the elementary and secondary school day in all Chilean publicly funded schools by approximately 30 percent between 1997 and 2010. Because of budgetary and logistical constraints, the timing of the introduction and expansion of full-day schooling varied across municipalities and birth cohorts. This study leverages both sources of variation to examine the effect of additional time in school on labor market outcomes in adulthood by matching the expected number of years a student would attend full-day school from school-level administrative data to nationally representative labor market data based on individuals’ year and place of birth. With these data, we compare outcomes among students born in the same region in different years and in different municipalities in the same year in a generalized difference-in-differences framework.
Our findings are threefold. First, we find that a longer school day increases educational attainment, earnings, and the likelihood of working in a skilled occupation in adulthood. The estimated earnings gains imply a 16 percent rate of return to an additional year-equivalent of schooling, in line with existing estimates on the returns to education during the 1990s and early 2000s in Chile (OECD 2013; Manacorda, Sánchez-Páramo, and Schady 2010).
Second, the existing literature has found that longer school days reduce teenage pregnancy (Berthelon and Kruger 2011). We extend this analysis to early adulthood and also document a delay in childbearing. Considering the full implications of these findings, we note that women in our sample are in their prime childbearing years; therefore, any intergenerational benefits will become apparent only in the coming decades.
Third, while access to longer school days increases educational attainment for all types of students, those from higher socioeconomic groups are more likely to work in managerial, professional, and technical occupations, whereas students from disadvantaged backgrounds are more likely to enter the workforce in less qualified occupations. These patterns suggest that students from different family backgrounds have different labor market adaptations, and disadvantaged students may face additional constraints in gaining access to skilled occupations.
Our study makes an important methodological contribution to the existing work on full-day schooling. While much of the existing literature examines a single birth cohort or jurisdiction, our approach identifies the causal effect of more time in school under relatively weak identifying assumptions regarding the timing and acceleration of the reform. Moreover, by limiting the sample to cohorts who were born before the reform was announced and defining the treatment variable based on place of birth, rather than the municipality of actual school attendance, our access measure does not rely on families choosing to move to areas that implemented the reform relatively early, attending school outside their municipality of residence, or more generally selecting a school based on full-day access.2 We further account for potential nonrandom selection into full-day schools by restricting comparisons to birth cohorts within a geographical region and controlling for local factors that may affect both reform implementation and economic outcomes.
Our empirical approach measures the effect of having access to additional schooling—the intent to treat (ITT) effect—a key policy-relevant parameter that provides the effect of offering a full-day schedule. In our setting, the ITT is equivalent to the treatment-on-the-treated (TOT) effect for students who attend a publicly funded school in the municipality in which they are born. When our sample attended school, about 90 percent of students attended a publicly funded school, 70 percent resided in the same municipality where they were born, and those who moved tended to move to cities with access to full-day schools similar to that in their birth municipalities.3 These patterns suggest the difference between the ITT and TOT is likely small for most groups in the aggregate.
Our work builds on an existing literature looking at the effects of additional time in school on student outcomes. Most of the previous research examines the short-term effects of longer school days by focusing on outcomes of current students, either during large-scale reforms that affect all elementary and secondary school students or after reforms targeting a particular age group, such as the expansion of full-day kindergarten in the United States. The results of this literature are mixed, with some studies finding no significant effect and others finding modest test score improvements (Bellei 2009; Valenzuela 2005; García 2006; Dias Mendes 2011; Llambí 2013; Orkin 2013).
Extended school days may affect outcomes other than academic performance. Longer school days provide a form of subsidized childcare and may therefore increase parental employment and family income (Berthelon, Kruger, and Oyarzún 2015; Contreras, Sepúlveda, and Cabrera 2010; Gibbs 2014; Gelbach 2002). Existing work finds longer school days in early grades confer medium-term benefits, including higher educational attainment and lower criminal justice system involvement (Cascio 2009). Older students may also benefit by an “incapacitation” mechanism that reduces risky behaviors that occur outside of school, such as teen pregnancy and involvement with the criminal justice system (Berthelon and Kruger 2011; Contreras, Sepúlveda, and Cabrera 2010). This paper contributes to previous work by evaluating the long-run effects of additional school time once students have completed their education and entered the labor market.
In addition to potentially changing students’ time use patterns and family resources, the move to longer school days increased infrastructure investments and altered other elements of the educational environment. Because these other resource changes coincide with the expansion of the school day, like all evaluations of full-day schooling reforms, we cannot fully separate the effects of additional instruction time, access to newer school facilities, changing time use patterns, or a combination of such factors. In the Chilean setting, students who have access to longer school days in early grades are also likely to have access in later grades, limiting the ability to distinguish whether benefits are driven by dynamic complementarities—in which additional learning at young ages facilitates future knowledge accumulation—or simply additional schooling at any point during one’s academic career. Despite these limitations, we can rule out some other possible channels for earnings gains. In particular, most effects are relatively small for individuals born in areas experiencing the largest increases in maternal labor supply over the reform period, inconsistent with a scenario in which schooling benefits students exclusively through a family income subsidy.
In the following, Section II discusses the existing literature on school day length and well-being. Section III describes the Chilean reform. Section IV outlines the empirical approach and data. Section V presents results, and Section VI concludes.
II. School Day Length and Student Outcomes
The time students spend in school, measured by either hours or days, depends on both individual and area characteristics that are correlated with student outcomes, making it difficult to separate different aspects of the education production function (Card and Krueger 1992; Ganimian and Murnane 2016; Patall, Cooper, and Allen 2010). These patterns caution against drawing causal conclusions from correlations between student outcomes and time spent in school. Perhaps unsurprisingly, early cross-sectional analyses find little association between the length of the school year and earnings in adulthood (Card and Krueger 1992; Heckman, Layne-Farrar, and Todd 1995).
A. Time in School and Academic Achievement
A closely related literature examines the short-term benefits from more time spent in school. This work focuses on current students and generally finds that longer school days improve academic performance. If improved achievement translates into greater educational attainment or human capital accumulation, these studies suggest that longer school days may also improve economic outcomes as students enter the labor force.
Several studies focusing on developed countries and leveraging variation in the length of the school day or year find that additional time in school improves academic performance in the short run. For example, Goodman (2014) finds that shorter school years due to additional snow days do not affect performance, but individual absences due to bad weather worsen math performance. Gibbs (2014) examines the effect of full-day classes for young students and finds full-day kindergarten improves test scores 0.3 standard deviations, with particularly large gains for Hispanic students. Pischke (2007) examines a reconfiguration of the West German academic calendar in which several states implemented multiple short school years. He finds that shorter school years worsened academic performance and increased grade repetition but had no long-term effect on employment or earnings when students were in their 20s and 30s. Finally, using cross-country data on within-student variation in subject-specific instructional time, Lavy (2015) finds an additional hour of instructional time in a given subject improves scores in that subject by 0.07–0.15 standard deviations, with a weaker relationship for students in lower-income countries.
A related literature examines the short-term effects of larger-scale reforms that gradually transition all elementary and secondary students from a part-day to a full-day schedule. Since the 1990s, several Latin American countries and cities have expanded the school day from approximately four to six or seven hours. Chile was one of the first countries to undertake such an expansion; therefore, much of the existing literature examines the Chilean reform.
The results from these reforms are mixed across countries, but they generally point to improved school performance (Glewwe et al. 2013). While some studies find worsened test scores following reforms targeted to disadvantaged or low-performing schools in Brazil (Dias Mendes 2011) and Uruguay (Llambí 2013), other studies find improvements in Colombia (Hincapie 2016; Bonilla 2011) and Buenos Aires (Llach et al. 2009). These large-scale reforms likely coincide with changes in other school resources—such as new facilities, mid-day school meals, and additional recess time—and examinations of these reforms, including the current study, cannot separate the effect of greater time spent in school from other concurrent changes to the learning environment. In general, resources that change students’ experiences by improving the quality or quantity of instructional time or improving nutritional intake improve achievement (Ganimian and Murnane 2016).
The Chilean reform extended the school day and allocated most of the additional time to greater instruction and teaching resources (DESUC 2005; Barrios-Fernández and Bovini 2021).4 Previous work has documented that the reform improved academic performance by 0.05–0.20 standard deviations (Bellei 2009; Valenzuela 2005; García 2006), with larger improvements found in work that accounts for nonrandom student selection into full-day schools (Berthelon, Kruger, and Vienne 2016).
Given the findings of the previous literature, we take the possibility of nonrandom selection seriously and estimate an intent-to-treat effect of having access to full-day schools. Our “access” measure is determined by factors that pre-date the reform announcement—namely, municipality of birth—and is not affected by subsequent migration patterns. This approach is similar in spirit to Berthelon, Kruger, and Vienne (2016), who instrument full-day school enrollment for current students using municipality-level full-day adoption in the previous year. Our approach is more general, as we aggregate access to full-day schooling over Grades 1–12 in order to provide an exposure measure covering an individual’s full academic career. In addition, because our labor market data do not include which schools individuals actually attend, we cannot precisely estimate the magnitude of endogenous mobility; rather, our approach is a reduced-form strategy based on a student’s municipality of birth.
B. Time in School and Outcomes in Adolescence and Early Adulthood
Improvements in short-term academic performance are neither a necessary nor a sufficient condition for additional school time to provide long-term benefits. If test score gains fade over time, or if skills that are measured by standardized tests are not correlated with labor market productivity, short-term improvements may not lead to labor market advantages. There is no empirical consensus on this point—some studies find a negligible relationship between short-term academic improvements and longer-term labor market outcomes in the context of full-day school reforms (Pischke 2007; Pires and Urzua 2015), while others find test score improvements are associated with higher earnings in adulthood (Rose 2006; Chetty, Friedman, and Rockoff 2011; Hansen, Heckman, and Mullen 2004; Murnane, Willett, and Levy 1995; Murnane et al. 2000). More broadly, a growing literature documents that educational investments that exhibit short-term gains followed by a medium-term fade out, such as classroom sizes and early childhood education, can still improve long-term economic outcomes (Chetty et al. 2011; Garces, Thomas, and Currie 2002).
Even if full-day schooling does not affect academic performance, there are several factors that may shape students’ economic opportunities by changing families’ and students’ time use patterns. Extended school days provide families with a childcare subsidy and reduce children’s leisure time. On this point, Contreras, Sepúlveda, and Cabrera (2010) and Berthelon, Kruger, and Vienne (2016) find that Chilean full-day schools increased female labor participation and employment, while Berthelon and Kruger (2011) find greater access to full-day secondary schools lowers adolescent crime and reduces teenage pregnancy rates among lower-income girls in urban areas. In the U.S. context, Cascio (2009) finds that the introduction of public kindergarten increased high school graduation and reduced involvement with the criminal justice system. As teen parenthood and a criminal history are associated with lower earnings later in life, while higher maternal employment potentially increases family resources, all of these findings suggest channels by which additional time in school may lead to longer-term benefits.
The existing work on how extended time in school affects labor market outcomes is relatively limited because many reforms occurred recently, and students who attended a full-day school are still completing schooling or are new labor market entrants. Because Chile was one of the first countries to lengthen its school day, much of the work on labor market outcomes focuses on the JEC reform. In a paper closely related to ours, Pires and Urzua (2015) examine the medium-term effects of the Chilean reform by comparing students who attended a full-day school starting at ages 14–15 (and were surveyed at ages 25–26) to older cohorts who completed school before the reform (surveyed at ages 29–30). They find attending full-day school improved academic performance but increased wages only among students who attended school in the afternoon before the reform. We build upon Pires and Urzua (2015) in several ways. First, their treatment cohorts attended full-day school for up to three to four years, less than half of the full treatment, while our sample covers the full implementation period and includes individuals with up to 12 years of full-day schooling. Second, we extend the analysis a decade by examining employment outcomes into students’ late 30s. This longer age range is important if greater school time delays labor market entry by increasing educational attainment. Finally, Pires and Urzua (2015) leverage variation in the schools that students actually attended and control for observable factors that may affect school choice. We take a complementary approach by using both cross-municipality and cross-cohort variation that is exogenous to the choice set students face.
III. Full-Day School Reform: Jornada Escolar Completa
Until the late 1990s, Chilean elementary and secondary students attended school four to five hours a day. Under this model, many schools operated a two-shift system in which some students attended school in the morning (8 a.m.–1 p.m.), and others attended in the same building during the afternoon (2–7 p.m.). Beginning in 1997, Chile implemented Jornada Escolar Completa (JEC), a large-scale reform that lengthened the school day in publicly funded schools by an average of 1.4 hours, while keeping the total number of school days fixed.5
JEC gradually moved all schools to a single full-day shift, with all students attending in the morning through mid-afternoon (8 a.m.–3 p.m.). This reform represents a substantial increase in schooling time. On average, instructional time increased 30 percent, and the total length of the school day increased 22 percent (Berthelon and Kruger 2011). This additional time could be used for either instructional or extracurricular activities; the stated goal was to improve school quality (Alfaro and Holland 2012). While we do not observe school-level changes in instructional time and therefore cannot speak to the relative productivity of additional instruction versus extra lunch or recreation time, most teachers, parents, and students reported that at least some of the additional time was used for language and math instruction, and only 2 percent reported spending more time studying for standardized tests (DESUC 2005).
While schools could choose when to begin offering an extended school day, JEC required a substantial infrastructure investment, as building and staffing resource needs nearly doubled in areas where schools previously operated in two shifts. Because of these practical constraints, schools operating under capacity tended to be among the first schools adopting JEC (Bellei 2009). Among schools without excess capacity, the Ministry of Education prioritized funding schools in disadvantaged areas and subsidized operational costs with a 20–50 percent increase in central government funding.6 The original legislation required that all schools receiving public funding operate on a full-day schedule by 2007 (public schools) or 2010 (publicly funded voucher schools), resulting in a 14-year rollout period.7
The school day was lengthened at the beginning of the academic year, but not all grades within a school implemented JEC in the same year. The youngest students typically gained access relatively early in the reform and continued a full-day schedule as they progressed through school, resulting in variation across cohorts and municipalities. Since JEC was a change specific to grades within a school that was generally first introduced for younger students, the vast majority—more than 90 percent—of students experienced increased access over their educational careers, and only 8 percent experienced a reduction in the probability of attending a full-day school when moving from one grade to the next. These reductions typically occurred between primary and secondary school. Accordingly, we confront the same issue faced by other work on full-day reforms: we are unable to disentangle dynamic complementarities—by which schooling at particular grades has especially pronounced effects—from treatment “dosage” in the amount of full-day schooling.
We calculate exposure to JEC, 
, as the expected number of years an individual born in cohort c in municipality m would attend a full-day school in Grades 1–12:
1
where 
is an indicator function equal to one if school s in municipality m implemented JEC for grade g when cohort c was in grade g. Nsgcm is the number of students enrolled in grade g in school s in municipality m, obtained from Ministry of Education administrative data. We use administrative enrollment data by school and grade from the 2013 school year as it is sufficiently late in the implementation process to provide a measure of capacity in schools that were built because of JEC. Moreover, since 2013 follows the formal implementation period, enrollment in this year is less prone to intra-municipality sorting between schools that offer JEC and those that do not.8
Because this measure of access is based on students’ locations before the policy was announced and does not depend on the school a student actually attends, it is not affected by students selecting in to full-day schools or moving to cities with greater JEC availability. This approach is similar to that used in several papers that examine the effect of JEC on short-term outcomes (Berthelon and Kruger 2011; Berthelon, Kruger, and Vienne 2016), but it builds upon the point-in-time estimates in previous work by summing full-day exposure across Grades 1–12 in order to obtain the total number of years a student would be expected to attend a full-day school throughout their career.9
provides a continuous measure of years of full-day school access ranging from zero to 12, consistent with the nature of the reform. Since multiple schools serve a single grade in 97 percent of municipalities, the probability a student attends a full-day school in any given year is not exactly equal to zero or one. Moreover, students with access to full-day schooling in an early grade may lose access later in their academic careers. This pattern is most common in areas where a large share of elementary schools adopted full-day schooling relatively early, but secondary schools adopted it relatively late.10
By this measure of full-day access, there is substantial variation both within and across birth cohorts, shown in Figure 1. Panel A shows the fifth, 50th, and 95th percentile of JEC access by birth cohort, indicating that an individual expected to attend four years of full-day school was in the 95th percentile of the 1986 birth cohort, the median of the 1989 cohort, and the fifth percentile of the 1993 birth cohort. Panel B plots the share of students attending a JEC school each year by municipality and illustrates substantial cross-municipality variation in both the introduction and expansion of full-day schools, with the thick line denoting the national average. This plot shows that some areas made considerable initial progress in implementing JEC but lagged in expanding to all grades, while others started slowly but quickly accelerated coverage. While less-populated areas tended to expand more quickly, other area characteristics do not significantly predict the pace of implementation.
JEC Timing Varied across Municipalities
Source: Ministry of Education (2016); CASEN (2006–2017).
Notes: Figures shows the expected years of attending a full-day school across cities of birth by birth year. Panel A: The bottom line with dotes denotes the expected years of full-day schooling for students at the fifth percentile of their cohort-specific distribution. The top line with dots denotes the expected years of full-day schooling for students in the 95th percentile, and the middle line shows the median number of years. Panel B: Figure shows the fraction of students in Grades 1–12 attending a JEC school in a given year with each line showing the pace of implementation for a single municipality. The bold black line is the national average.
Figure 2 summarizes how this varying exposure translates into the JEC exposure distribution for our main sample and shows considerable variation in full-day access. About 11 percent of our sample had no access to full-day schools; we exclude this large spike from the figure. Among those with some exposure to the reform, one-quarter of those are expected to attend a full-day school for at least four years, and 9 percent are expected to attend full-day schools for at least six years.
Expected JEC Exposure
Source: Ministry of Education (2016); CASEN (2006–2017).
Notes: Figure shows the distribution of expected years of JEC attendance between Grades 1 and 12 for individuals born between 1979 and 1992 outside the Santiago metropolitan region who were 23–38 at the time of the CASEN survey. The mass point at exactly zero years (11 percent of the sample) is omitted for visualization purposes.
IV. Empirical Approach
A. Data
We map expected years of full-day schooling to data on economic outcomes in adulthood from the 2006–2017 waves of Chile’s biennial demographic survey, the National Socioeconomic Characterization Survey (CASEN), using information on an individual’s year and municipality of birth (Dominguez and Ruffini 2021). Similar to other household surveys, such as the Current Population Survey (CPS) in the United States, the CASEN is a large, regionally and nationally representative survey that provides individual-level information on labor market participation, household structure, educational attainment, family background, and income.11 Importantly for our purposes, starting with the 2006 survey, each individual was asked where their mother was living when they were born, whether in the current municipality of residence or a different municipality (and if the latter, which). We use this information to identify the municipality of birth, linking approximately 98 percent of respondents to a birthplace.12
We limit the sample to individuals born between 1979–1992 who were school-aged (5–18) the first year of the reform and were thus exposed to between zero and 12 years of full-day schooling. Our main sample limits the data to respondents ages 23–38 at the time of the CASEN survey in order to focus on economic outcomes after individuals have completed schooling.13
B. Exposure to JEC
JEC is typical of full-day schooling reforms in Latin American countries. Longer school days require a substantial increase in facilities and instructional resources. For example, if a single building operated two school “shifts” at capacity before the reform, a transition to full-day schooling would require doubling building space and teaching staff. Since new facilities must be built and additional teachers and staff recruited, these reforms are typically implemented over multiple years.
One approach to estimating the effects of a longer school day assumes that the timing of longer school days is randomly assigned. It then estimates the effect on outcome y of attending a full-day school for JECicmt years for individual i living in municipality m in birth cohort c and surveyed at t as:
2
There are several reasons why JEC may be correlated with students’ potential earnings and employment, even after accounting for cohort- and municipality-specific factors. First, Chile adopted full-day schooling during a period of economic growth; therefore, comparing outcomes of younger and older cohorts will conflate the effect of additional schooling with aggregate wage growth and other improvements in economic opportunities.
Second, relying on geographic variation in full-day access for a single cohort is also potentially problematic. Given the funding requirements of a large-scale schooling expansion, policymakers might prioritize funding to undersubscribed schools with excess capacity or, alternatively, direct limited resources to the neediest areas first. If disadvantaged areas pilot the program, a naive ordinary least squares approach comparing early- and late-adopting areas understates any benefits. On the other hand, if early-adopting schools are more able to support a large-scale expansion, the framework in Equation 2 will overstate any benefits. We explore these patterns in three analyses discussed below and find some evidence that less-populated areas and those with lower levels of educational attainment tended to implement JEC earlier and more quickly, but these patterns largely disappear after controlling for time-invariant municipality conditions. Moreover, in order to account for these patterns, we follow the existing literature and consider only variation across cohorts born in the same municipality and include a vector of controls for both contemporaneous regional economic conditions, as well as survey year trends in baseline (1996) educational attainment, population, poverty, and employment rates by individuals’ municipality of birth.14
Third, even if JEC implementation was randomly allocated across schools over time, the within-cohort approach would not fully account for student selection into full-day schools arising from families moving across municipalities or attending a school in a different municipality from where they live. About 20 percent of school-aged children live in a municipality other than the one where they were born, and lower socioeconomic (SES) populations are significantly less likely to move than students from more educated families. Most of these moves are local, with less than 10 percent of children moving across regions. In our sample, families that move tend to migrate to areas with nearly identical JEC access.15 Regarding school selection, although families can choose which school to attend, most students attend a nearby school; 95 percent travel less than 6 kilometers to school, and most elementary students attend a school within two kilometers (Gallego and Hernando 2010; Chumacero, Gómez, and Paredes 2011). It may still be the case, however, that those who enroll in full-time schools are likely those who benefit the most from the additional school time (Berthelon, Kruger, and Vienne 2016). As our data do not include where adult CASEN respondents lived and attended school during childhood, we define full-day school access, “exposure” to JEC, as the expected number of years a student would attend a full-day school based on their birth municipality and cohort (Equation 1). Our main specifications omit the Santiago metropolitan region, where municipalities are more geographically proximate and travel time to nonneighborhood schools is shorter (Chumacero, Gómez, and Paredes 2011).16
The gradual rollout of JEC provides two sources of variation. First, children born in the same municipality are exposed to different amounts of full-day schooling that varies based on birth year. Second, children born in the same year are exposed to different amounts of schooling depending on their municipality of birth.17 We leverage both this within- and across-cohort variation. A causal interpretation of our results therefore involves the identifying assumption that the pace of JEC implementation is uncorrelated with factors related to changes in educational attainment and labor market outcomes among students born in the same municipality in different years. Using the measure of JEC access from Equation 1, 
, we estimate the effect of full-day schools on outcome yicmt as:
3
for individual i born in cohort c and municipality m and surveyed in year t. In order to improve precision, we include Xicmt, a vector of demographic characteristics including age, gender, indigenous identity, and maternal education. For outcomes other than educational attainment and childbearing, Xicmt, also includes controls for marital status and number and presence of children interacted with gender and household size. Zicmt is a vector of municipality characteristics for a respondent’s current municipality, including employment and poverty rates and average earned income. αtcZ1996r is a separate survey year linear trend in baseline (1996) educational attainment, population, employment, and poverty rates in an individual’s municipality of birth.18
The JEC implementation period coincides with improved economic conditions as real GDP increased about 50 percent between when the oldest and youngest individuals in our sample were born (World Bank 2017). In addition, secondary school became guaranteed through age 21 for cohorts graduating in 2003 or later (Ley 19876 2003).19 Survey year fixed effects, ϕt, account for level differences in economic performance at the time of the survey, and municipality fixed effects, ψm, control for local time-invariant characteristics. We finally include region-specific cohort fixed effects, δcr, in order to limit our comparisons to students born in the same year within a geographic region and capture relatively local economic conditions that may affect each cohort’s access to JEC and subsequent labor market outcomes.20 As δcr varies by cohort, it accounts for both national and regional-level changes in schooling requirements or education policy.21
The empirical approach in Equation 3 assumes a linear treatment effect—that is, that marginal benefits are constant for each additional year of JEC exposure. In order to explore the presence of increasing or decreasing marginal returns, we also present findings from a less parametric approach that replaces the continuous treatment measure with nine one-year exposure bins, pooling all observations with at least eight years of exposure:
4
When interpreting these results and reconciling with the difference-in-difference estimates, we note that access to full-day schooling is skewed: 30 percent of our sample has access to one year or less of JEC, and one percent is estimated to receive more than eight years of full-day instruction (Figure 2). With this distribution in mind, the more flexible strategy suggests diminishing marginal benefits to each additional year of full-day schooling.
To explore whether JEC implementation is correlated with economic and demographic characteristics during the rollout period, Table 1 regresses the fraction of students attending a full-day school on municipal characteristics each year the CASEN was administered during JEC implementation (1996, 1998, 2000, 2003, and 2006). In models without municipality fixed effects, Columns 1–2 show that areas with low levels of educational attainment provided greater JEC access, and some evidence pointing to greater access in low-populated municipalities. Columns 3–6 include municipality fixed effects in order to examine whether increases in JEC coverage relate to local economic changes. While Columns 3–4 suggest greater JEC participation is associated with increased poverty, we do not see a similar relationship when examining per capita income (Columns 5–6).22 In light of these patterns, we include survey year trends in the 1996 levels of educational attainment, population, poverty, and employment rates for each individual’s municipality of birth. Online Appendix tables omit these trends and yield similar results.
Predictive Characteristics of JEC Adoption: Economic Characteristics during the Rollout Period
Online Appendix Table A1 performs a complementary analysis by examining whether JEC’s introduction and the pace of its rollout is correlated with municipality characteristics in the pre-JEC period. Here, we regress the first and last year of JEC implementation, as well as the number of years it took for a municipality to go from 0 to 100 percent coverage, on municipality characteristics from the 1990–1996 period, an approach similar to Hoynes and Schanzenbach (2009).23 While age structure and agricultural share do not significantly predict the timing or speed of JEC adoption, there is some evidence that the most populated and poorer municipalities within a region adopted JEC relatively early (Column 4), but both less-poor and less-educated areas had longer implementation periods (Columns 7–8). Even with regional fixed effects, more than 70 percent of variation in implementation is unexplained by baseline area characteristics.
Finally, in order to illustrate graphically the plausibility of the differences-indifferences parallel trends assumption, Online Appendix Figure A1 examines pre-trends in employment (top panel) and earnings (bottom panel) among 23–38-year-olds between 1990 and 1996 by whether a municipality implemented JEC relatively early (left panel) or quickly (right panel). This figure residualizes each outcome on a vector of region-by-cohort and municipality fixed effects and plots trends by approximate terciles of JEC implementation, defined as the share of students at a JEC school in 1998 (the second year of JEC adoption, left panel), or the first CASEN year in which at least 75 percent of students attended a JEC school (right panel). Both outcomes exhibit noisy patterns and do not suggest that early- or fast-adopting municipalities experienced exceptionally strong economic growth before JEC adoption. If anything, young adult earnings in areas that implemented earlier and more quickly were slightly declining immediately before JEC adoption.
Leveraging this municipality and cohort variation in access to full-day schools, we examine outcomes for all students and subpopulations defined by gender and family socioeconomic status (SES). We proxy for socioeconomic status using maternal educational attainment, defining “high-SES” as individuals whose mother graduated high school and “low-SES” as those whose mother did not complete high school.24 Heterogeneity by gender and family background is of interest for several reasons. First, maternal education is predictive of child outcomes (Andrabi, Das, and Khwaja 2013; Carneiro, Heckman, and Vytlacil 2011; Currie 2009), suggesting a channel for any intergenerational benefits. Second, essentially all children from lower-income families attend government-subsidized schools, whereas most students from the highest-income families attend private schools (CASEN 2016). Moreover, fewer children from disadvantaged backgrounds move across municipalities between birth and school age (18 vs. 28 percent). Therefore, estimated access to full-day schooling is more likely to reflect actual access among this population, and we may interpret lower-income students as a “high-complier” population for whom the reported ITT estimates are likely similar to the TOT effect. In addition, evidence from other educational interventions suggests the returns to educational inputs may be larger for lower-income students (Cunha et al. 2006; Havnes and Mogstad 2011).
Table 2 displays summary statistics for the full sample and each subpopulation. The average respondent is about 28 years old and expected to have attended full-day school for two years. These characteristics are similar by gender and family socioeconomic status. Overall, about 80 percent graduated high school, and 20 percent have at least a four-year university degree, but students from disadvantaged backgrounds have lower levels of educational attainment. About two-thirds of the full sample worked in the previous month, and women have substantially lower participation rates than men.
Summary Statistics: Main Adult Sample
Online Appendix Table A2 builds on the locality analyses in Online Appendix Table A1 to explore whether exposure to JEC is associated with individual characteristics by regressing student characteristics defined at birth on JEC access, controlling for survey year, municipality of birth, and region-by-cohort fixed effects. While there are no significant differences in access by maternal educational attainment, race, or gender, in order to improve precision, all of our results control for these characteristics.
V. Findings
A. Educational Attainment
Changes in educational attainment are one potential mechanism for any patterns in earnings or labor force attachment, but a priori, the effect of JEC on high school and college graduation is ambiguous. On the one hand, less leisure time during high school restricts teenagers from holding part-time jobs and increases the opportunity costs of attending school, which may increase dropout rates. On the other hand, if more time in elementary or secondary school prepares students for higher education or instills nonpecuniary benefits of schooling (a stronger “taste” for education), longer school days may increase educational attainment.
Tables 3–5 estimate the effect of longer school days on high school graduation and university enrollment and graduation under the framework in Equation 3. The first row presents the effect an additional year of full-day schooling access, so that the effect for the average individual in our sample is obtained by multiplying this row by the average JEC access in each sample, 
, about two years for college graduation and three years for the slightly younger high school graduate sample. The implied effect of an additional year-equivalent of education then scales the main estimate by the average increase in instructional time (β/0.3).
Longer School Days and High School Graduation
Longer School Days and College Enrollment
Longer School Days and College Graduation
For the full population, access to an additional year of full-day schooling increases the probability of high school graduation by 2.1 percentage points (Table 3, Column 1) and also increases graduation among all subgroups, particularly women (Column 2) and students from disadvantaged backgrounds (“low-SES,” Column 4).25 These qualitative patterns are robust to alternative samples and more parsimonious sets of local economic controls (Online Appendix Table A3).
Tables 4 and 5 reveal different patterns for college enrollment and graduation. Increases in college attendance are similar across groups, but increases in college graduation are larger for men (1.9 percentage points, or 11 percent) and higher-SES populations (1.9 percentage points, or 6 percent). Therefore, whereas additional time in elementary and secondary school prompted women and low-SES students to complete high school, men and higher-SES students respond by completing college at higher rates.
Other changes to the education system coincided with JEC implementation. In 2003, secondary education became guaranteed up to age 21, which may have mechanically increased high school graduation. Importantly, as guaranteed schooling affected all cohorts born in 1982 and later, regardless of place of birth, region-by-cohort fixed effects account for this national reform. While high school graduation rates for cohorts born after 1982 are 12 percentage points higher than those among the 1979–81 birth cohorts (83 vs. 71 percent), secondary school completion was not universal after 2003, suggesting other interventions could induce schooling completion. We obtain slightly larger results for high school graduation when limiting the sample to cohorts born 1982 or later, suggesting that longer school days increased educational attainment beyond the provisions in the 2003 reform (Online Appendix Table A3, Columns 1–2).
To explore whether additional years of full-day schooling exhibit diminishing marginal benefits or if there is a threshold after which longer school days provide especially large benefits, Figure 3 plots the βy coefficients from the less parametric approach in Equation 4. While any full-day schooling increases high school graduation rates by approximately three percentage points, there are relatively small marginal increases for the next two years of full-day school. Figure 4 shows the likelihood of college graduation among the full population is generally increasing in exposure to JEC, with each additional year of exposure conferring a smaller marginal gain. This aggregate pattern is clearer among men. Consistent with Table 5, additional time in school has less effect on college graduation rates among disadvantaged students, while those from more highly educated families incur a one-time increase that slightly increases with additional exposure.
High School Graduation by Expected Years of Full-Day Schooling
Notes: Figure shows the results from Equation 4, where each coefficient is an indicator for 0, (0,1), [1,2)… [4,5), [8,12] years of 
access. Dependent variable equals one if the respondent completed high school at the time of the CASEN survey. All specifications include municipality of birth, survey year, and birth year-by-region fixed effects; current municipality of residence employment and poverty rates, gender, a quadratic in age, indigenous identity, and maternal education, as well as linear survey year trends in baseline educational attainment, population, poverty, and employment rates by municipality of birth from the 1996 CASEN. Vertical lines denote 95 percent confidence intervals clustered by municipality of birth. Sample is individuals born between 1979 and 1992 outside the Santiago region who were 19–38 years old at the time of survey. See text and Online Appendix for details.
College Graduation by Expected Years of Full Day Schooling
Notes: The figure shows the results from Equation 4, where each coefficient is an indicator for 0, (0,1), [1,2)… [4,5), [8, 12] years of 
access. Dependent variable equals one if the respondent had received a university degree. All specifications include municipality of birth, survey year, and birth year-by-region fixed effects; current municipality of residence employment and poverty rates, gender, a quadratic in age, indigenous identity, and maternal education, as well as linear survey year trends in baseline educational attainment, population, poverty, and employment rates by municipality of birth from the 1996 CASEN. Vertical lines denote 95 percent confidence intervals clustered by municipality of birth. Sample is individuals born between 1979 and 1992 outside the Santiago region who were 23–38 years old at the time of survey. See text and Online Appendix for details.
These figures also illustrate the distributional effects of JEC. About 30 percent of our sample is exposed to less than one year of JEC, while about 5 percent have at least six years of access. The vertical distance from one year of JEC to the [6,7) point then roughly corresponds to changes going from the 30th percentile to the 95th percentile of JEC access (about ten log points for high school graduation and six log points for college graduation).
B. Labor Market Outcomes
The return to secondary schooling was large during JEC implementation, with estimates of the high school wage premium ranging from about 34 percent relative to those with an eighth grade education (8 percent per year of secondary education) to 64 percent relative to those with a sixth grade education (11 percent a year) (OECD 2013; Manacorda, Sánchez-Páramo, and Schady 2010). The estimated earnings premium for postsecondary education is even higher. Manacorda, Sánchez-Páramo, and Schady (2010) find Chilean men with a university degree have higher labor force participation rates and earn 90 percent more than those with a secondary education. These results, combined with increases in educational attainment documented in Section V.A, suggest we might expect higher earnings and greater labor force attachment in adulthood.
1. Employment
As the probability of working increases with educational attainment, Table 6 examines whether JEC changed employment rates, defined as earning at least 30,000 pesos in work income the previous month (approximately $50 in 2017 dollars).26 Except for high-SES students, longer school days increase employment about 0.7–1.3 percentage points from a base of 54–78 percent. These results are larger than the null employment response found in Pires and Urzua (2015). One potential explanation for these differences is that Section V.A documents increases in educational attainment, while Pires and Urzua (2015) focus on students who are in their mid-20s and may not have completed their educations. These different findings emphasize the importance of considering a period after respondents have completed schooling in order to capture the full long-term economic effect.
Longer School Days and Employment in the Previous Month
Figure 5 takes a less parametric approach and shows the probability of employment is generally increasing in access to full-day schooling, with significant employment gains among women and disadvantaged students emerging with approximately two years of JEC access. In contrast, there is no evidence that access to more full-day schooling changes employment among students from more educated families. In additional results we do not find a significant change in the probability that young adults are currently in school, suggesting that the lack of an employment response among students from the highest-SES backgrounds is not driven by selection out of the labor force and into postsecondary schooling.
Employment in Previous Month by Expected Years of Full-Day Schooling
Notes: Figure shows the results from Equation 4, where each coefficient is an indicator for 0, (0,1), [1,2)… [4,5), [8, 12] years of 
access. Dependent variable equals one if the respondent earned at least 30,000 pesos in the past month (about $50). All specifications include municipality of birth, survey year, and birth year-by-region fixed effects; municipality of residence employment and poverty rates, gender, a quadratic in age, indigenous identity, household size, maternal education, marital status, and number and presence of children, interacted with gender, as well as linear survey year trends in baseline educational attainment, population, poverty, and employment rates by municipality of birth from the 1996 CASEN. Vertical lines denote 95 percent confidence intervals clustered by municipality of birth. Sample is individuals born between 1979 and 1992 outside the Santiago region who were 23–38 years old at the time of survey. See text and Online Appendix for details.
2. Earnings
Increases in educational attainment and employment suggest that access to longer school days could increase earnings in early adulthood. Yet to our knowledge, this study provides one of the first analyses of the relationship between earnings and full-day schooling in the context of a full, large-scale national reform.27
Panel A of Table 7 reports the semi-elasticity of earnings with respect to an additional year of JEC, where earnings are measured as the log of earnings in the previous month plus one in order to include individuals with no earnings.28 Consistent with the finding that longer school days improve labor market outcomes, Table 7 shows each additional year of access to full-day schooling increases earnings by 4–5 percent for all groups except those from the most advantaged backgrounds (Columns 1–4). To put these numbers in context, as JEC increased instructional time by 30 percent, the results in Column 1 suggest a return of about 16 percent (0.049/0.3) to each year-equivalent of education. These magnitudes are on the higher end of the returns to education found in higher-income countries (Card 1999) and consistent with the ranges found for Chile during the 1990s.
Longer School Days and Log Monthly Earnings
To examine the full distribution of earnings responses—whether gains are concentrated in the low- or high-wage tails of the earnings distribution—Figure 6 displays results from a series of regressions in which the outcome of interest is a binary variable whether an individual has annual earnings of at least x pesos, following Carrell, Hoekstra, and Kuka (2018). These results incorporate both labor force participation and earnings responses and indicate that access to longer school days had particularly pronounced effects on the low end of the labor market but did not meaningfully increase the probability of earning more than 1.5 million pesos a month ($2,500, the 97th percentile). For high-SES individuals, the pattern is less monotonic, with the most pronounced earnings gains between 0.6 and 1.2 million pesos, consistent with the fact that these individuals have relatively high earnings regardless of JEC access.
Effects of JEC across the Earnings Distribution
Notes: Figure shows the results from a series of regressions as in Equation 3, where the dependent variable equals one if monthly earnings exceeded each earnings threshold (in 2017 pesos). All specifications include municipality of birth, survey year, and birth year-by-region fixed effects; municipality of residence employment and poverty rates, gender, a quadratic in age, indigenous identity, household size, maternal education, marital status, and number and presence of children interacted with gender, as well as linear survey year trends in baseline educational attainment, poverty, population, and employment rates by municipality of birth from the 1996 CASEN. 
defined as the expected years of full-day school attendance based on an individual’s municipality and year of birth, calculated as described in Equation 1. Vertical lines denote 95 percent confidence intervals clustered by municipality of birth. Sample is individuals born between 1979 and 1992 outside the Santiago region who were 23–38 years old at the time of survey.
Figure 7 investigates nonlinearities in access to longer school days and earnings. For the overall, male, and disadvantaged populations, log earnings increase approximately linearly between about two to eight years, with little evidence of diminishing marginal returns. Regardless of how many years more advantaged students are likely to have access to JEC, there is no significant change in earnings.29
Log Monthly Earnings by Expected Years of Full-Day Schooling
Notes: Figure shows the results from Equation 4 where each coefficient is an indicator for 0, (0,1), [1,2)…[4,5), [8, 12] years of 
access. Dependent variable is log(monthly earnings +1) in 2017 pesos. All specifications include municipality of birth, survey year, and birth year-by-region fixed effects; current municipality employment and poverty rates, gender, a quadratic in age, indigenous identity, household size, maternal education, marital status, and number and presence of children interacted with gender, as well as linear survey year trends in baseline educational attainment, population, poverty, and employment rates by municipality of birth from the 1996 CASEN. Vertical lines denote 95 percent confidence intervals clustered by municipality of birth. Sample is individuals born between 1979 and 1992 outside the Santiago metropolitan region who were 23–38 years old at the time of survey. See text and Online Appendix for details.
Finally, Panel B of Table 7 limits the sample to workers (defined as in Table 6) in order to examine whether the overall earnings response is driven by more individuals entering the workforce or higher earnings among the employed. Between half and 80 percent of the overall increase is due to higher earnings for employed women, men, and low-SES groups, and earnings also increase for employed respondents from high-SES backgrounds.
C. Mechanisms
1. Migration
In addition to greater educational attainment, there are several intermediate, nonmutually exclusive channels through which longer school days could increase adult earnings. One possible explanation for increased employment and earnings is that greater educational attainment enables individuals to move to Santiago and other high-wage areas.
Table 8 investigates the relationship between JEC exposure and subsequent moves to a municipality outside the municipality of birth. No subpopulation displays a significant change in migration patterns from greater exposure to longer school days. Online Appendix Table A8 Column 1 shows that migration to Santiago, the largest metropolitan area, likewise did not change. Column 2 takes a more general approach by examining migration to relatively prosperous areas, defined as a standardized index from the leave-out-mean individual income in each respondent’s current municipality.30 Here, we find that individuals with greater access to full-day schooling, particularly those from more highly educated families, tend to move to more prosperous areas.31 In Section V.D, we consider general equilibrium effects in order to analyze how JEC shaped the economic opportunities in an area.
Longer School Days and Cross-Municipality Migration
2. Fertility patterns
Motherhood is associated with labor force withdrawal and lower earnings upon reentry (Waldfogel 1998; Kleven, Landais, and Søgaard 2018; Bertrand, Goldin, and Katz 2010; Kuziemko et al. 2018). Previous work has documented that JEC lowered teen pregnancy rates among disadvantaged women in urban areas (Berthelon and Kruger 2011), and we also find small reductions in teen pregnancy (Online Appendix Table A9, Column 1). More generally, among women who gave birth to at least one child, each additional year of full-day schooling delayed birth by about two months (Table 9 and Figure 8). These delays in childbearing are slightly larger for lower-SES women and women born in urban areas (Table 9, Columns 2 and 4), consistent with earlier work on teenage pregnancy. As not all individuals in our sample have reached their prime childbearing years, this estimate likely understates the full effect of JEC on family formation patterns.
Longer School Days and Age at First Birth (Women)
Age at First Birth
Notes: Figure shows the results from Equation 4 where each coefficient is an indicator for 0, (0,1), [1,2)…[4,5), [8,12] years of 
access. Dependent variable is the age in years a woman gave birth to her first child. All specifications include municipality of birth, survey year, and birth year-by-region fixed effects; current municipality of residence employment and poverty rates, gender, a quadratic in age, indigenous identity, and maternal education, as well as linear survey year trends in baseline educational attainment, population, poverty, and employment rates by municipality of birth from the 1996 CASEN. Vertical lines denote 95 percent confidence intervals clustered by municipality of birth. Sample is women born between 1979 and 1992 outside the Santiago metropolitan region who had given birth to at least one child at the time of the survey. See text and Online Appendix for details.
3. Occupation choice
Another mechanism by which longer school days could increase earnings is through occupational choice. As the majority of additional school time went towards reading and math instruction, students attending full-day schools are expected to have entered the labor force with greater skills, even absent a formal credential. Table 10 shows that for all individuals from nondisadvantaged backgrounds, longer school days increase by about one percentage point (about 3–4 percent) the likelihood of having a managerial, professional, or technical occupation, but they do not affect the share of low-SES individuals in these roles.32
Longer School Days and Working in a Skilled Occupation
Most “high-skill” occupations in Table 10 require a university degree. In order to capture upskilling across the entire skill distribution, Column 1 of Online Appendix Table A10 explores an alternative measure of occupational prestige, defined as the leave-out-mean of log earnings of other workers j in the same four-digit occupation as worker 
. As with the findings for skilled occupations, increased access to full-day schooling increases occupational prestige for all groups other than lower-SES individuals. Therefore, occupational decisions cannot explain the increased earnings among disadvantaged students.
4. Family resources
In addition to increasing students’ human capital accumulation, longer school days provide childcare for families. This implicit childcare subsidy increases family resources and potentially allows parents to enter the labor force or work longer hours rather than provide home-based care. Although the CASEN does not ask whether a respondent’s parents were employed during childhood, we provide evidence suggestive of the extent to which increased parental employment drives our findings by examining changes within a “high maternal LFP increase” sample of municipalities. Areas in this sample experienced a change in labor force participation among mothers with school-aged children between 1996 and 2006 greater than the median municipality (6.3 percentage points from a base of about 38 percent), with an average increase of about 12 percentage points.
Table 11 examines the main outcomes for the high maternal labor force participation sample. For most outcomes, we cannot rule out results of the magnitude found in the main findings, and we do find that benefits arise exclusively in areas with large increases in maternal labor supply. While we are unable to account directly for changes in family resources at an individual level, these results suggest that our main findings are not exclusively driven through changes in family resources during childhood.
Long-Term Economic Well-Being in Municipalities with Largest Increases in Maternal LFP
D. General Equilibrium Considerations and Robustness
The 1990s and early 2000s were a period of political stability and economic growth in both urban and rural Chile. Policymakers across the political spectrum advocated policies to alleviate poverty and open the country’s economy to trade (Foxley 2004). Similar economic reforms occurred in much of Latin America and eastern Europe during this period and continue in many emerging economies today. Therefore, our results may generalize to other settings.
A separate question is the extent to which our findings have internal validity, or whether JEC access captures other local economic changes that affect labor market outcomes. Columns 3–6 in Table 1 indicate that after conditioning on year and municipality fixed effects, increases in municipality-level poverty are associated with slightly greater JEC access, although this relationship is sensitive to the measure of disadvantage and is small in magnitude. Other labor market and demographic characteristics at the time of JEC implementation are not significantly associated with the pace of JEC adoption, and Online Appendix Figure A1 illustrates that before JEC implementation, there are no discernible patterns in employment or average earnings—measures of economic growth—between areas that adopted early and late or between those that adopted more quickly or slowly. Here, we further explore this issue by examining the relationship between full-day schooling and the entire local economy in the long run.
JEC was a large-scale reform increasing classroom time up to 30 percent for all students attending publicly funded schools. Given the nature of the program, the partial equilibrium effects on the treated cohorts—the internal rate of return—may understate the full return to an additional year of schooling. Specifically, Table 3 showed JEC increased educational attainment, thereby increasing the size of the skilled labor force. In standard economic models, this increase in skilled labor is expected to reduce the earnings of skilled workers relative to those with less education (Goldin and Katz 2009). To the extent that younger and older workers are imperfect substitutes, examining spillover effects to skilled and unskilled older workers can provide a sense of the magnitude of any general equilibrium effects (Khanna 2015).
To estimate the presence of general equilibrium and spillover effects of additional schooling, we augment Equation 3 by adding the average years of JEC exposure among the full adult population (ages 18 and older) and labor force:
5
where ml denotes the municipality in which individual i currently lives, and mb is the municipality of birth. Here, β1 captures the private returns to an additional year of fullday schooling. 
is the average exposure among adults living in municipality ml in survey year t, and β2 captures spillover effects of the aggregate increase in educational attainment. As spillover effects are based on individuals’ current municipality of residence, this framework incorporates all migration decisions. We include region of birth by cohort, municipality of birth, municipality of residence, and survey year fixed effects, as well as the standard set of individual and municipality controls, X′ and Z′, from Equation 3.33
We estimate Equation 5 separately by skill level (those with less than a high school diploma and two measures of “high-skill,” those who graduated high school and those who completed college) and by age (individuals who were school-aged when the reform was introduced, “young” birth cohorts 1979–1993, and those who had already entered the labor market, “old” birth cohorts 1954–1978). Since we have labor market information from multiple CASEN waves, and ml is not perfectly collinear with mb, β1, and β2 are separately identified for young cohorts. For the older cohorts who graduated high school before JEC, only the parameter associated with the spillover effects, β2, is identified.
Table 12 shows how exposing an entire population (Panel A) or workforce (Panel B) to longer school days affects each skill category and generation. For young high school graduates, the internal return to education is positive and significant, on the order of about 3–4 percent a year (Column 2), or slightly more than half of the earnings gains estimated in Table 7. In contrast, the internal returns are small in magnitude and insignificant for both the lowest- and highest-skilled groups (Columns 1 and 3). Spillover effects to older cohorts are less precisely estimated for all education groups and differ slightly depending on whether aggregate access to full-day schools is measured across the population or the workforce.
General Equilibrium. Effects of Longer School Days on Log Earnings
Overall, these results are consistent with higher levels of education in the labor force facilitating sectoral shifts and agglomeration economies that stimulate the demand for relatively skilled workers. These patterns across age groups also suggest old and young workers are imperfect substitutes and that any negative externalities from a larger young, relatively skilled workforce are small in this context for at least the first 20 years of the reform. As students exposed to additional years of JEC enter the labor force and progress in their careers, these dynamics may change.
As a complementary exercise to examine spillover effects to cohorts that did not attend full-day schooling, we conduct a placebo analysis on cohorts born between 1959 and 1973 (20 years before the main sample) or 1964 and 1978 (15 years before). Both groups completed secondary schooling before 1997 and therefore did not have access to JEC. Our placebo treatment is arbitrarily set at the expected number of JEC years received by the cohort born z ε{15, 20} years later in the same municipality:
6
If the main results were simply capturing economic growth, we should see improvements in labor market outcomes for these older individuals. Online Appendix Figure A2 does not show economically or statistically significant changes in earnings or skilled occupations for any subgroup or placebo year. We do find small increases in high school graduation for some groups (male 20-year placebo and women 15-year placebo), but these results are between one-third and two-thirds the size of the main estimates and do not translate into increases in college graduation or earnings. Overall, the lack of a consistent pattern across outcomes and placebo groups provides very suggestive evidence that our findings do not solely reflect local economic conditions that affect the entire workforce.34
VI. Conclusion
We find that a large-scale national reform that lengthened the school day for elementary and secondary students improved long-term economic well-being by increasing educational attainment, prompting more women and students from disadvantaged backgrounds to enter the labor force, and generating earnings gains. The magnitude of these earnings gains—about 4–5 percent for an additional year of full-day schooling—is consistent with other work examining the returns to education in Chile during this time. The margins of adjustment vary by subgroup: students from lower-SES families are more likely to enroll in college and enter the workforce, whereas those from more advantaged backgrounds complete college, work in high-skill occupations, and live in wealthy areas at higher rates.
These results are consistent with longer school days promoting greater human capital accumulation, as general equilibrium effects are imprecisely estimated and relatively modest in magnitude for most groups. Moreover, we do not observe especially large improvements in areas that experienced the largest increases in maternal employment during the JEC rollout period, suggesting that our findings are not due solely to increases in family resources or parental employment.
While access to additional time in school benefits students, a broader question is whether such large-scale investments are beneficial from a social welfare perspective. Extending the school day on a national level for all students requires substantial resources. In the Chilean case, full-day schooling increased per-pupil expenditures by at least 20 percent, approximately 18,000 pesos (or 31 USD) per student each month. Extrapolating the estimated earnings increase in Table 7, Panel A, the additional earnings for students attending school in the first 20 years of the reform are between 60 and 120 percent as large as the increase in per-pupil spending over this period. In the steady state, the estimated direct government cost of providing 12 years of full-day school is about 10 percent of the discounted increase in earnings between ages 23–65.35 This back-of-the-envelope calculation is not a full cost—benefit analysis—it does not include infrastructure, maintenance, or teacher hiring costs, or benefits accruing from delayed childbearing or reduced crime (Berthelon and Kruger 2011)—but it does illustrate that many benefits of educational investments are realized only in the long run, while costs are primarily incurred in the short term. Altogether, the broad-based nature of our results shows that large-scale investments in public education can generate long-term and meaningful improvements in economic well-being.
Footnotes
The authors thank Hilary Hoynes, Enrico Moretti, Steve Raphael, Jesse Rothstein, Chris Walters, and participants in the Association for Education Policy and Finance 2018 meetings and UC Berkeley Public Policy research seminar, Development Lunch, and Political Economy working group for helpful comments and suggestions. Dominguez acknowledges funding provided by ANID Fondecyt de Iniciacion N. 11220261 and the MIGRA Millneium Nucleus, ANID-MILENIO-NCS2022-051. The data used in this article are available online at ICPSR (https://doi.org/10.3886/E139181V1).
Supplementary materials are available online at: https://jhr.uwpress.org/.
↵1. See, for example, President Obama’s March 10, 2009 speech to the Hispanic Chamber of Commerce. http://www.nytimes.com/2009/03/10/us/politics/10text-obama.html (accessed January 6, 2023).
↵2. A municipality is a geographic measure similar to a city.
↵3. There is no information on the municipality in which students attended school, which could differ from the municipality of residence, over this time period.
↵4. Comprehensive information on school resources, including funding and other staff metrics is not available for the full implementation period, limiting our ability to determine the extent to which our findings are driven by increased financial or staffing resources.
↵5. About 93 percent of students in the K–12 system enroll in publicly funded schools. These schools include public schools that are managed by local municipalities and private-voucher schools that are managed by private entities but subject to central government legislation. Originally, schools were mandated to expand the school day in Grades 3–12 by 2007, but in practice, most implemented the reform in Grades 1 and 2 as well. As young students also had access to longer days, we measure JEC exposure across Grades 1–12.
↵6. JEC subsidy amounts were school–grade specific, based on the grade served, switch costs, school characteristics, and preexisting infrastructure deficits. Relatively small and rural schools tended to receive the largest subsidies (Berthelon and Kruger 2011; Contreras, Sepúlveda, and Cabrera 2010).
↵7. Ministry of Education data indicate that in 2016, approximately 36 percent of students attended a public school and 55 percent attended a voucher school. Private nonvoucher schools (covering about 6 percent of students) were not subject to JEC and private nonvoucher enrollment remained relatively unchanged when our sample was in school. In practice, the implementation period was delayed, and by 2010, only about 75 percent of schools had a full-day schedule.
↵8. 2013 is also the earliest year grade-level enrollment data are available from the Ministry of Education. If schools that adopted JEC relatively early experienced increases in enrollment relative to those that adopted later within the same municipality, using a later year will cause the estimated
to be larger than the true E(JECcm), and therefore our results will represent a lower bound on the returns to full-day access. Results, available upon request, show larger improvements under specifications that weigh JEC access by the fraction of schools (rather than students). Our main specifications exclude all schools with a single student in grade g in 2013. Results are robust to including all schools.↵9. The household survey data include respondents’ year of birth, but not their birth month. The Chilean school year begins in March, and children who turn five through June are eligible to enroll (McEwan and Shapiro 2008) provide a full description of Chilean enrollment cutoffs). We define age in first grade as a child’s year of birth plus six; accordingly, for children born in January through June, our approach assigns them the JEC exposure of a younger cohort (that is, weakly more years of full-day schooling than they actually had access to), thereby underestimating JEC exposure and providing a lower bound of the actual exposure effect.
↵10. As cumulative access depends on both when a cohort first gained access and how quickly JEC expanded, a typical event study framework is not feasible in this setting. In an event study spirit, however, Figures 3–5 and 7–8 illustrate the extent to which there are constant returns to an additional year of JEC access.
↵11. Our main specifications use regionally representative weights in order to provide the most comprehensive coverage of the population; results are qualitatively unchanged when using municipal (“comuna”)–level weights or without weighting.
↵12. Unmatched observations report mothers’ residences at birth at a larger unit of geographic measurement than a municipality (for example, the region or the province).
↵13. Results examining high school graduation extend the sample to include individuals ages 19–22 who were born in the 1979–1992 window. Results are robust to excluding these individuals.
↵14. Results are robust to alternative measures of poverty and employment, such as per capita income, as well as municipality of birth linear trends.
↵15. The typical mover moves to an area with about 0.06 percent greater access.
↵16. In 2015 enrollment data, about 15 percent of elementary school students in Santiago live and attend school in different municipalities, compared with less than 10 percent in other regions. The final two columns of Online Appendix Tables A3–A6 and A8–A10 include Santiago and show results that are smaller in magnitude and less precise. With the available data, we are unable to determine whether this pattern is due to a weak first stage with estimated JEC access being a noisy measure of actual access or heterogeneous benefits between urban and rural locations.
↵17. All individuals in our sample were born in 1992 or earlier, before the reform was announced in 1997. As we rely on location decisions before the policy was announced (that is, those at birth), our estimated access to full-day schooling is not affected by any migration decisions occurring after the policy announcement.
↵18. The tables in the Online Appendix show results are nearly identical when omitting baseline trends.
↵19. Columns 1 and 2 in Online Appendix Table A3 show similar increases in educational attainment when limiting the sample to cohorts graduating either before or after the compulsory schooling reform, and sample sizes indicate that most individuals in our sample graduated before this policy became effective.
↵20. In the 14 regions outside Santiago, there are between four and 54 municipalities with at least one respondent in our sample. These municipalities tend to be much smaller than regions, with a population between about 200 and 300,000 compared with a regional population of 100,000 to 1.8 million. As approximately one-quarter of the cross-municipality variation in the employment rate and two-thirds of the cross-municipality variation in the poverty rate is between regions, region-by-cohort fixed effects capture a meaningful share of the national variation in economic well-being.
↵21. JEC exposure is defined by municipality and birth cohort, so municipality-by-cohort fixed effects are not separately identified from the treatment variable.
↵22. The relationship between poverty and JEC expansion is also small in magnitude. Moving from zero to 100 percent poverty implies the share of students attending full-day schools would increase by 14 percentage points. Between 1996 and 2006, the poverty rate in the median municipality fell by 12.5 percentage points; therefore, scaling the estimated coefficient by the (absolute) changes in poverty implies a one percentage point change in the share of students attending a full day school over the decade.
↵23. We pool CASEN years 1990, 1992, 1994, and 1996 to maximize the sample size, but municipality-level information is not available for the smallest municipalities.
↵24. Respondents with unknown maternal educational attainment are not included in either subgroup (about 20 percent of the sample from CASEN waves 2006–2015 and 40 percent for 2017). There is no significant correlation between years of JEC access and maternal education nonresponse.
↵25. The low-SES sample consists of individuals whose mothers have no more than a basic education, as reported by individuals and linked by family structure. See Online Appendix for details. We pool men and women from disadvantaged households; there are no substantial gender differences by SES group.
↵26. Results are similar to defining work as employment in the week before the reference period. We measure employment over the previous month based on a nominal income amount as it represents a minimum amount of labor force attachment under a broader employment definition.
↵27. Pires and Urzua (2015) examine labor market outcomes for students who were in high school when JEC was introduced and measure earnings when these students are in their mid-20s. Our study broadens our understanding of this relationship by examining labor market outcomes for students who had up to 12 years of access to longer schooling and by tracking earnings through individuals’ 20s and 30s.
↵28. Online Appendix Table A6 shows larger earnings gains using levels or earnings or an inverse hyperbolic sine transformation as the dependent variable.
↵29. Although we find a marginally significant 0.2 hour increase in the number of monthly hours worked for high-SES individuals, reported usual hours are likely particularly prone to measurement error: half of workers report working exactly 45 hours a week, and there is no information on part-year or seasonal work.
↵30. The index is
where is per capita income in municipality m surveyed at year t, ΣmΣtymt is the grand mean across all municipality–years, and is the corresponding standard deviation.↵31. As emigrants are not surveyed in CASEN, findings do not reflect international migration.
↵32. “Skilled” occupations are defined by the ILO as managerial, professional, and technical occupations (major codes 1, 2, and 3).
↵33. Applying the model outlined in Khanna (2015) is not feasible in this setting, as all of our “young” cohorts receive some exposure to longer school days; that is, there are no purely “untreated” municipalities.
↵34. Mother’s residence at birth was introduced in the CASEN in 2006. Therefore, we lack the statistical power to conduct this analysis with individuals of the same age at the time of the survey, and note that the placebo cohort are older than the main sample. All analyses, including the placebo specifications, control for a quadratic in age.
↵35. We obtain a similar range when calculating the net costs of the first 20 years of implementation. Each of these estimates assume a 3 percent social discount rate.
- Received April 2019.
- Accepted March 2021.
