Children s time in school and female labor force participation in Chile

Transcription

1 Children s time in school and female labor force participation in Chile Matias Berthelon, Universidad Adolfo Ibañez* Diana Kruger, Universidad Adolfo Ibañez** Melanie Oyarzún, Universidad Catolica de Valparaiso*** Abstract: Lack of adequate childcare is one of the main reasons Chilean women cite for not participating in the labor force, partially explaining the country s low female labor force participation rate. We investigate the effect of a reform that lengthened school schedules from half to full days essentially providing zero-cost childcare on women s decision to participate in the labor force. We identify the effect of the policy from its gradual implementation across municipalities over time and rule out alternative explanations, finding evidence of positive and important effects on mothers labor force participation and employment stability. JEL Classification: H41, H52, I25, I28, J13, J16, J18, J22, O15 Keywords: Full day schooling, primary education, female labor participation, education reform, Chile * Corresponding Author: addresses: matias.berthelon@uai.cl. Mailing address: Av. Padre Hurtado 750, Viña del Mar, Chile. Tel.: ; fax: ** address: diana.kruger@uai.cl Av. Padre Hurtado 750, Viña del Mar, Chile. *** address: melanie.oyarzun@gmail.com. Av. Brasil 2830, Piso 7, Valparaíso, Chile The authors would like to thank seminar participants at the University of Chile, the Catholic University of Chile, Oregon State University, University of Maryland, and the Inter American Development Bank. Berthelon and Kruger received financial support from Chile s National Committee of Scientific and Technological Research (Comisión Nacional de Investigación Científica y Tecnológica,CONICYT), through FONDECYT Project No The authors thank the Sub-Secretariat of Social Provision for granting permission for the use of Chile s Social Protection Surveys for the years 2002, 2004, 2006, and All results, errors and omissions are sole responsibility of the authors.

2 1. Introduction Although women s labor supply in Chile has increased substantially, from 35 percent in 2002 to 49 percent in 2012, it remains among the lowest in Latin America and among OECD countries (Figure 1). 1 In addition, the gap between men s and women s participation rates is approximately 30 percentage points, and inequalities also exist among women along several dimensions: labor market participation rates are lower among low education and low income women, possibly reflecting different human capital levels, employment opportunities, and constraints in job search and retention. Increasing female employment, especially among lowerincome women, would increase family income and foster economic development. Since women s participation in the work force plays an import role in households socioeconomic conditions, it is important to understand the reasons why they don t work. One of the main determinants of inactivity among women is motherhood, and the negative relationship between the number of children of school age and their mothers activity in the labor force has been verified in empirical studies across a range of countries. 2 Among inactive women with school-age children worldwide, one of the main reasons for not working and not seeking employment is lack of childcare. For instance, lack of family-friendly employment policies in the United States, such as parental leave and part-time work entitlements, explains almost 30 percent of the decrease in U.S. women s labor force participation in the past 25 years (Blau and Kahn, 2013). 1 Multiple surveys are available to estimate labor force participation in Chile. Although it is measured using different questions and periods of reference, all sources reveal consistent stylized facts: female labor force participation has increased during the last decades, yet is among the lowest in Latin America. 2 For theoretical discussions on the relationship between labor supply and childcare, see Blau and Robins (1988), Blau (2003), and Blau and Currie (2006), whichalso summarizes the empirical literature. Studies for developing countries include Cruces and Galiani (2004) in Argentina, Peña-Parga and Glassman (2004) in Colombia, and Gong and van Soest (2002) and Wong and Levine (1992) in Mexico. 2

3 In a context where women face difficulties entering the labor force, public policies that improve access to childcare should thus have a positive impact on female participation and employment. The empirical evidence is vast and conclusive. Studies that have analyzed the relationship between childcare and female labor supply find that mothers labor supply measured as participation, employment, and hours worked increases when the price of childcare falls and when access is improved. 3 The economics literature has focused on the effects of childcare for children aged 5 or younger; less is known about the effects of childcare for older children on mothers labor decisions. This paper seeks to improve what we know about childcare for school-aged children, by analyzing an education reform that increased the length of school schedules for primary and secondary students in Chile, from half to full-day shifts. The reform increased by almost 35 percent the number of hours that children aged 6 to 18 spend in school and under adult care, thereby providing zero cost childcare that may potentially impact the employment decisions of their mothers. The implementation of the Full-Day School reform, which we refer to as FDS in this paper, 4 was gradual across Chilean municipalities, due mostly to budget and infrastructure constraints. We exploit this exogenous source of variation in access to FDS schools to identify the effect of childcare on women s employment decisions. Hernando (2009) and Contreras et al. (2010) analyze the impact of the FDS reform on women s labor force participation using repeated rounds of the 3 Blau and Currie (2006) provide a survey of empirical studies of childcare price elasticities in the U.S. Studies for other countries also find a negative response of female labor supply to the price of childcare: for instance, Loskshin (2004) for Russia, Powell (2002) for Canada, Fong and Lokshin (2000) for Rumania, and Lokshin et al (1999) for Kenya. Papers analyzing the effects of childcare expansion include Gelbach (2002) and Cascio (2009) for the U.S., Baker et al. (2008) and Lefebvre and Merrigan (2008) for Canada, Berlinski and Galiani (2007) for Argentina, Calderon (2012) for Mexico, and Aguirre (2012), Martínez and Medrano (2009) and Martínez and Encina (2009) for Chile. 4 The program is referred to as JEC for the Spanish acronym of the reform s name, Jornada Escolar Completa. 3

4 national Chilean household survey, CASEN, which has been fielded every two to three years since 1990, and both find that the increased time children spend in school led to a small increase in labor force participation of women. Our paper differs from these studies and contributes to the existing literature in several ways. First, we analyze the role of childcare of older, school-aged children in the labor decisions of their mothers. As discussed earlier, the economics literature has extensively studied the effects of childcare for children aged 5 and younger, but not so for the impact of childcare of older children, despite the important role it may have in affecting mothers decision to work. Second, we analyze not only women s participation decision, but also whether they were able to obtain more stable employment as a result of having access to full-day schools for their children. We use richer measures of labor force status that cover a longer period of reference, including measures of whether a woman has participated in the labor market any time during the last 12 months, whether she participated in the labor force for 6 or more months during the year, and the percentage of the year they worked or sought employment. These variables allow for a deeper understanding of labor force attachment, employment stability, and their response to childcare availability. Third, unlike previous studies our estimates control for individuals unobserved heterogeneity that includes differences in unobservable attachment to the labor force and/or preferences regarding childcare arrangements, thus addressing omitted factors in prior research. We achieve this by taking advantage of the first longitudinal data set constructed in Chile, the Social Protection Survey (or EPS for its Spanish acronym), which allows us to follow the labor market decisions of approximately 7,000 women over a period of 7 years, from 2002 to The survey s name in Spanish is Encuesta de Protección Social. 4

5 We find that the expansion of the FDS program significantly increased women s labor force participation. Our estimates indicate that if FDS coverage increases by 40 percentage points which is about the change required to move to full coverage of the policy in our sample average female labor force participation increases by 4 percentage points. This number represents a large increase considering other economic events that have also affected women s participation in the work force. For instance, between 2006 and 2009, average labor force participation in our sample declined by 4 percentage points, mostly due to worsening labor market outcomes related to the economic crisis. Our estimated impact of moving to full FDS coverage would have been large enough to offset the effect of the crisis. In addition, we find that the program had a larger impact on mothers with children in primary school, and among that group, mothers without preschool children benefitted most. Our results indicate that full FDS availability would increase the likelihood that mothers of primary school-aged children engage in the labor market by 9 percent. We rule out several possible explanations, such as endogeneity of the FDS program and female labor force participation, preexisting trends, and overlap with a major childcare policy implemented during part of the period analyzed. We are confident that our estimated effects of the FDS policy on labor force participation are not driven by alternative explanations. Our findings have direct policy implications regarding the design of policies to promote women s insertion into the labor force. Several countries in the region including Chile have implemented or are considering after-school programs for children in primary school age. Furthermore, although most countries in Latin America continue to have short school schedules operating as multiple shifts, a small number of countries are considering extending school days to longer schedules, such as the Chilean reform we analyzed. Our findings suggest that such 5

6 measures would increase labor participation of women, as well as facilitate longer-term, more stable employment. Our findings also suggest that the lack of family-friendly policies in Chile could explain at least partly the country s low female participation rates, similar to recent findings for other OECD countries (Blau and Kahn, 2013). Extending the school day in Chile, so that it became more synchronized with mother s employment schedules, had the indirect effect of promoting mothers employment and access to more stable jobs. Thus, if promoting women s participation in the labor force is a relevant public policy objective, then policies need to address the provision of childcare not only for young but also for older, school aged children, who are not old enough to care for themselves. 2. Background: Female labor force participation and full-day school reform in Chile Female labor force participation in Chile The female labor force participation rate in Chile is lower than other OECD countries, and lower than the Latin American average (Figure 1). The explanations offered range from rigidities in the labor market regarding part-time work, to cultural factors and conservative values. 6 Nonetheless, according to Chilean household surveys, average labor force participation among women aged 25 to 55 has increased from approximately 42 percent in 1990 to 59 percent in 2009 (Figure 2). The increase in women s participation has occurred in all income and education levels, but as in other countries, income and education are positively correlated with female participation in the labor force: over 70 percent of women from the highest income quintile participated, compared to just 36 percent of women from the lowest quintile, and while 81 percent of women 6 Chilean empirical studies include Contreras et. al (2010), Contreras and Plaza (2010), Perticara (2006), Acosta et al. (2005), and Mizala et. al (1999). 6

7 with university degrees were economically active, compared to 38 percent of women without a primary education (Figure 2). Motherhood is also a factor: being a mother of children aged 13 years or younger is correlated with lower labor force participation, but the gap between mothers and non-mothers has been closing since 1990 (Figure 3). This is due to a relatively faster increase in the participation of mothers with children is in primary school (aged 6 to 13 years), who by 2009 had the same labor force participation rate as women without children in this age group. This gap begins to narrow rapidly after 1998, which coincides with the year that the FDS reform began to be implemented. The participation rate of mothers with pre-school aged children (5 or younger) also increased during the period, but a gap of 8 percentage points still existed in 2009 relative to women without pre-school aged children. The full-day school reform A large-scale, national education reform that increased the amount of time students spend in the classroom has been gradually implemented in Chile since The reform augmented instruction time by about 35 percent in primary schools without increasing the number of days in the academic calendar, which amounted to an additional 1.5 to 2 hours of daily classroom time. The reform presented operational challenges that restructured how public schools function in Chile at both primary and secondary levels: schools went from providing two half-day shifts to one fullday shift, which gave the reform its name, Full Day Schooling. 7 The FDS program mandated that all primary and secondary schools that receive public funds municipal or private subsidized must offer a full-day program by 2007 and 2010, 7 In Chile the reform is referred to as JEC, which is the Spanish acronym for full-day schooling ( Jornada Escolar Completa ), approved in law No.19,532. 7

8 respectively. 8 Additionally, the FDS law mandates that all publicly funded schools created after 1997 must initiate operations as full-day schools. 9 We describe the evolution of primary school enrollment by FDS adscription in Figure 4; by 2009, coverage of the FDS reform had reached 66 percent of total primary school enrollment. Although childcare was not an explicit goal of the policy, by increasing the time that children spend in school and the number of hours they receive adult care, the FDS policy reduced the cost of childcare for school-aged children, which may potentially impact the employment decisions of their mothers. Since mothers with children in primary school report lack of childcare as an important determinant of inactivity (see Table 1), the FDS policy should lead to an increase in women s participation. Although the reform was implemented in both primary and secondary schools, we focus only on the impact of primary full-day school availability on mothers labor force participation decisions, because lack of childcare is not a concern to mothers with children in secondary school (Table 1). The implementation of the reform was gradual due to infrastructure and financial constraints. The first schools to switch to the FDS regime were rural and smaller schools with excess capacity; these were the first to change because the reform required little or no new infrastructure. Schools that wish to change their operations to FDS compete for public funds through an application process with the Ministry of Education, where they submit their academic plans and request the required funds to operate under the full day regime. The FDS program has operational and infrastructure costs. The operational component includes variable costs that increase as a result of lengthening the amount of time children spend in 8 Private schools which represent 8% of the total enrollment are not obligated to ascribe to the FDS program. 9 The impact of the FDS reform in several outcomes has been studied. It has been found that it improved results of students standardized tests (Bellei, 2009), reduced teen pregnancy and juvenile crime (Berthelon and Kruger, 2011), and improved cognitive test scores, reduced high-school dropout, and lowered adolescent motherhood (Pires and Urzua, 2014). 8

9 school; for example, teachers salaries, administrative costs, and the provision of school lunches. To cover operational costs, the per-student subsidy regularly paid to all public and voucher schools increased by 40%. The most important expense (and constraint) associated with a full-day school is the expansion of schools infrastructure to accommodate, in many cases, twice the number of students at any given time. In the funds application process, schools competed for limited capital infrastructure funds to finance their infrastructure requirements. The Ministry of Education does not allocate FDS funds randomly; rather, they select schools by balancing two possibly opposing criteria: (i) schools with relatively low switch costs (i.e., rural schools, as described earlier) and (ii) priority schools with pre-existing deficits in infrastructure and located in areas of socio-economic vulnerability (in section 5 we discuss how the non-random funds allocation rules may affect our empirical strategy). Identification in our empirical estimations relies on the fact that the FDS program was taken up at different rates across Chilean regions and municipalities. Table 2 presents the pace of implementation of the program in primary schools across Chile s thirteen administrative Regions, from the initial year of the reform until 2009, indicating the average share of primary schools that were under a full-day regime; the last column of Table 2 is the relative size of the primary student population in each region. An inverse relationship exists between enrollment and FDS implementation. The Santiago Metropolitan Region (RM) was home to 37 percent of Chilean primary school students in 2009, yet there the reform was slowest, mainly because schools in this predominantly urban area have physical space constraints or higher costs of expansion of their infrastructure. The fastest implementation occurred in the sparsely populated IX, X and XI Regions, comprised mostly of rural areas and small cities and where a total of approximately 15 percent of 9

10 primary school students reside. In each of these regions, about 83 percent of primary schools were full day by In the Santiago metropolitan region, home to almost 40 percent of primary school students, the FDS reform had reached only 60 percent coverage by Table 2 also reveals that the reform had significant temporal and regional variation in the depth of its implementation. Our identification strategy is based on this quasi-experimental nature of the reform s implementation at the municipal level. 3. Model, data and variables Model We estimate a reduced-form model of female labor force participation. As we are interested in the effect of full-day primary schooling on mother s labor force attachment, the model can be described as follows: FDS (1) where the dependent variable is a measure of the labor force status of woman i living in municipality m in year t. We use several measures of labor force status, which are described below. As parents are able to choose the school for their children, this choice can introduce a source of endogeneity. 10 In order to minimize this effect, we measure the total availability of FDS schools at the local (municipal) level. Additionally, our data does not identify the school that children attend. The policy variable of interest, FDS, measures the share of full-day primary schools in municipality m in year t. As mentioned earlier, our identification strategy relies on the gradual and heterogeneous implementation of the policy over time and across municipalities. 10 Gallego and Hernando (2009) found that the FDS reform did not affect demand for schools, so it is reasonable to propose that the policy is exogenous to mothers labor force participation decisions. 10

11 Our estimations include time-varying individual characteristics in vector, and municipality-level characteristics in vector. Given that we are using a panel data set we include an individual-level fixed effect, i, which allows us to control for individual unobserved heterogeneity in the labor participation decisions. Additionally, we include a full set of municipality fixed effects,, to control for geographic time-invariant characteristics, and regiontime fixed effects,, to control for macro-level trends. Finally, imt is an idiosyncratic error term. It s important to include municipality fixed effects to control for omitted factors that affect women s labor supply that are specific to municipalities, such as local labor market conditions, proximity to labor markets that demand women s labor, and women s relative wages, among others. In doing so, the impact of policy is identified through within-municipality changes in coverage of FDS schools over the period we analyze. Since our first year of data is 2002, and the policy had been in place for five years, it may be difficult to detect an effect from the policy if implementation was almost complete. However, Table 2 reveals that by 2002 only 60 percent of schools had adopted full-day schooling, leaving scope for identification in our data set. Data and variables Individuals information, i.e., labor force participation outcomes and socioeconomic characteristics in, comes from Chile s Social Protection Survey, EPS. The EPS is the first long-term longitudinal survey implemented in Chile, with several rounds available (years 2002, 2004, 2006, and 2009), whose aim is to collect information for the design of social policies in particular, for the assessment and reform of the Chilean retirement system. For this purpose, it collects detailed information of respondent s current labor status and labor history, and also information on education, health, household characteristics, and family demographics. In order to avoid the correlation between women s schooling, retirement, and labor decisions, we 11

12 include in our sample only women aged 25 to 55, leaving us with an unbalanced panel of more than 6,900 women. The panel nature of the data allows us to estimate a model with individual fixed effects to identify the effect of the reform controlling for women s unobservable characteristics, such as attachment to the labor force and/or preferences regarding childcare arrangements. We construct three measures of women s labor force status. First we use a discrete variable that indicates if the woman worked at any time during the year of the survey. 11 The rich nature of the EPS data regarding individuals employment history allows us to construct two additional variables that measure longer-term attachment to the labor force: a discrete 0-1 variable that indicates whether the woman participated in the labor force more than 50% of time during the year of the survey, and a continuous variable that measures the fraction of the year spent in the labor force. 12 The EPS data also allows us to construct measures of women s years of education, whether she is the head of the household, and her age (and age squared) as proxies for experience. The EPS does not report the school that children attend, so in order to construct the policy variable of interest, FDS, we obtained administrative school data from the Ministry of Education that contains detailed yearly information on full day enrollment within a school. We are interested in constructing a measure of the share of the schools offering FDS within a municipality. Before 2007, schools that adopted the program were not required to implement full days for all their grade-levels, they were just required to offer it to all classrooms of the same grade. For instance, a school offering first to sixth grades, and with two classrooms per grade, could offer FDS for fifth and sixth grade only, provided that both classrooms in each grade were included. Therefore in 11 This measure of labor force status is different from other surveys in Chile whose reference periods for labor force participation can be the week or the month prior to the interview. 12 The fraction of the year is obtained by first generating a monthly binary variable that indicates whether the woman worked during that month. Then we add the total months worked during the year, and divide by either 12 or the number of months covered by the survey (the 2009 survey was carried out at different times of the year for some of the participants). 12

13 order to determine whether a school is participating in the program it is relevant to consider the depth of the implementation within the school in any given year. We define a school as an FDS school when more than half the grade levels are FDS. Thus our policy variable is the share of schools in a municipality with more than 50% of their grade levels under the full-day regime. 13 For the municipal-level variables in vector, we use data from Chile s CASEN Household Surveys to construct measures of average adult educational attainment, male and female labor force participation rates, poverty rate, unemployment rate, and average income per capita in each municipality. Summary statistics of all variables are found in Table 3. In our sample of women aged 25 to 55 years, 74 percent of women participated in the labor force at some point during the year of the survey. At first glance, the level of women s labor force participation reported in the EPS is at odds with other Chilean surveys, which report lower levels (particularly the CASEN survey). The difference is due to the way in which the different surveys inquire about status in the labor force. In the CASENs, respondents are asked about labor force status during the previous 4 weeks, whereas the EPS inquires about respondents complete labor history, so that our discrete measure of women s participation in the labor force captures activity in the labor force for a longer period of reference relative to the CASEN surveys (as our summary statistics reveal). 14 Longer-term attachment is somewhat lower: 70 percent of women participated for 6 or more months during any given year, and on average women worked 69 percent of survey year. 13 As a refinement we constructed a second measure of FDS availability, which includes the fraction of schools that offer all their grade levels as FDS and share of total enrollment under FDS. Results were not qualitatively different from the ones reported here and they are available upon request. 14 Another difference between the EPS and CASEN surveys is the person who responds questions. In CASEN surveys, the head of the household answers questions pertaining for all members of the household; since 70% of households are headed by males, they respond regarding the activities of their spouse/mate, and they potentially under-report female participation depending on whether they consider their spouse s activities as work. We thank Jaime Ruiz-Tagle for pointing this out to us. 13

14 Regarding FDS implementation in our sample, 59 percent of schools offered FDS in at least half of their grade levels. 4. Results In order to assess the impact of full-day schools availability on women s labor force participation, we first start with a simple model in which we only include region-year fixed effects, to control for heterogeneous time-trends in female LFP across the thirteen regions of the country, and a set of municipality fixed effects (see Table 4, column 1). We find no significant effect of the FDS program, which indicates that, controlling for permanent differences across municipalities, the FDS policy is not correlated to average trends in female labor force participation. In the next estimate, we include the full set of individual and municipal characteristics (Table 4, column 2), finding that even though the point estimate of the FDS increased six times, it is still not significantly different from zero. This specification which does not control for individual fixed effects is closest to previous estimates in the literature (with pooled cross section data), and our results are similar to the ones reported by Contreras et al. (2012) and Hernando (2009). Our next specification introduces an individual fixed effect without the individual and municipality controls, to exploit the panel structure of our data (Table 4, column 3). The point estimate increases ten fold and the standard error decreases (as compared to the corresponding estimates of pooled data in column 1), and we find that once we control for women s unobservable characteristics, the likelihood of participating in the labor force is positively affected by the availability of FDS primary schools in the municipality. Finally, when the full model is estimated including individual and municipal characteristics, as well as the individual fixed effect, the previous result remains relatively 14

15 unchanged (Table 4, column 4). We observe that individual characteristics hold the expected signs and that municipal-level variables are not statistically significant. One possible explanation of their lack of significance is that most of the variation of these variables comes form differences between municipalities, since they change little across time, so that once municipality fixed effects are included, this set of variables does not significantly affect labor force participation. Results reported in Table 4 highlight one of the main contributions of our paper, namely, that unobservable women s characteristics play an important role in the decision to participate in the labor force, and our estimates are able to account for them given the panel structure of our data. The decision to engage in the labor market is determined not only by observed characteristics such as women s education level, training in the workplace, or local employment opportunities but also in great part by unobservable individual factors. Accounting for this unobserved heterogeneity yields more accurate estimates of the effect of full-day schooling. In terms of the magnitude of the impact of the FDS program, the point estimates in column 4 indicate that if the share of FDS schools in a municipality increases by 1 (or 100 percentage points), then the probability that a woman participates in the labor force would increase by 10 percentage points. To put this result in relevant context, during the period we analyzed, about 60 percent of primary schools were FDS (see Table 3), so that an increase of 40 percentage points, i.e., increasing to full FDS coverage, would lead to a predicted increase in female LFP of 4 percentage points. If we consider that in our sample female labor force participation declined by 4 percentage points between 2006 and 2009 due to worsening economic conditions related to the 2008 international crisis a movement towards universal coverage of the FDS program would have an effect similar in magnitude to the negative effect of the crisis. In the next set of estimates we test whether FDS availability has an impact on the measures 15

16 of labor force participation that include longer term attachment to the labor market; for ease of exposition, Table 5 reports only the estimates of the effect of FDS access. To compare outcomes, column 1 of Table 5 repeats the estimate of the full model reported previously, while column 2 reports the effect of FDS access on female participation in the labor force 50% of the year or more; we don t find a statistically significant effect of FDS on this outcome. In the last column of Table 5, we report effects on the share of the year worked, and find that if the share of FDS schools increases by 40 percentage points, women would increase the share of the year they work by 3.7 percentage points, or about 5 percent. 15 We find that access to schools with longer schedules increases the likelihood that women participate in the labor force, revealing that childcare for school-aged children is an important determinant of mothers employment decisions. Furthermore, our results suggest a significantly large effect not only on the probability that women engage in the labor force, but also on working for longer periods (Table 5, column 3). The full-day school policy appears to have long term effects in the sense that it provides women with additional time to devote to longer term, and more stable work. Results in Table 5 reveal the average effect of the policy on all women. We also analyzed whether the full-day school policy had a positive impact on women most affected by the reform i.e., mothers of children of primary school age. We would expect that women with children in this age range would be the ones benefiting most from the policy, and we should not see an effect on mothers with children of other age groups: younger children did not benefit from the policy, and mothers of high school children did not report childcare as a relevant factor in their employment decision (see Table 1). 15 The marginal effect reported is relative to the rate of labor force participation, and is obtained by dividing the point estimate by average female LFP. 16

17 We analyze the heterogenous impact of the reform with two approaches: first, by estimating a model interacting the FDS variable with a dummy for the presence of a son or daughter in the household that is of primary school age (6 to 13 years). Thus our first exercise explores whether mothers of primary-school aged children responded differently to FDS access than other women. Results are reported in the first three columns of Table 6 (columns 1 through 3); the effect of the FDS variable on mothers of primary school-aged children is the sum of the two coefficients, whereas the effect on other women (without children in primary schools) is the FDS coefficient. Table 6 includes the results of a test of the joint significance of both coefficients. As expected, we find that FDS access had no significant impact on labor force participation all three measures of women without children in primary school: in columns 1 through 3 of Table 6, the coefficient of the FDS variable without an interaction term is never statistically significant from zero. In contrast, we observe that FDS access had a large positive effect on mothers of primary school aged children: the interaction term of FDS with a dummy for motherhood of a primary school-aged child is positive and significant (Table 6, columns 1 to 3), and a test of joint significance indicates that both coefficients are jointly significant (results of this test are reported at the bottom of the table). The point estimates reveal as expected a stronger effect for this subgroup. Since mothers have lower labor force attachment, the marginal effect of FDS access is also stronger. For instance, the unconditional probability that a mother of a child in primary school works at any time during the year is 71.3 percent (compared to 76 percent for other women). For these mothers, therefore, an increase in 40 percentage points in FDS coverage would increase both the probability of working at any time during the year and the probability of working at least half 17

18 of the year by 9 percent (columns 1 and 2). 16 In column 3 we observe that a similar increase of FDS coverage would increase time spent in the labor force by 10 percent, which is two times the magnitude of the effect obtained considering the whole sample. Presence of pre-school children limits mothers possibility to work, especially if they do not have access to childcare. We would expect the effect of the FDS policy to be smaller on mothers of primary school aged children with siblings of pre-school age. Even if these mothers have access to full-day primary schools, without childcare during working days for all their children, their employment possibilities are not amplified by the FDS policy. 17 Thus, our second exercise consists in analyzing differentiated effects of the policy between two groups of mothers: mothers with at least one child of primary school age, with and without pre-school aged children. Within the sample of mothers affected by the reform, i.e., mothers of children aged 6 to 13, we construct a dummy variable that takes the value of one if the mother also has a child in childcare or preschool age (5 years or younger), and we interact this variable with our FDS policy measure. We report these results in columns 4 through 6 of Table 6. In this specification, the effect of FDS access on mothers of primary school children without younger siblings is indicated by the coefficients of the FDS variable, while the effect on mothers of both primary and pre-school children is captured by the sum of the FDS variable and the interaction term. Results from the test of joint significance of the two variables are also reported. Our estimates indicate that larger FDS availability increases the labor force participation of mothers without children in pre-school, in two of our measures of LFP: ever participating and 16 We obtain the marginal effect for mothers by adding both coefficients, then multiplying this sum by 0.4 (the increase in FDS coverage) and dividing by the mean of the dependent variable of the relevant subgroup. 17 In the last three years of our data ( ), there was an increase in the supply of public childcare and preschools through different programs that were independent from (and not coordinated with) the FDS program. Nonetheless, in our section on robustness checks, we perform an exercise to see whether our results are affected by the expansion of childcare and preschool centers during this period and find that they are not (see section 5 and Table 8). 18

19 share of the year worked (Table 6, columns 4 and 6). Based on results of the FDS variable in column 4, an increase of 40 percentage points in FDS availability would increase the probability of working at any time by 9.4 percent, which is more than two times the magnitude of the baseline estimates. In comparison, among the group of mothers who also have children in preschool age, the increase in probability is only 1.7 percent. 18 These results are in line with expectations about how access to childcare affects women s labor decisions, as increased care facilitates their chance of engaging in labor market related activities. In our results, women with less access to childcare those with children of preschool age benefit less from the longer school schedules provided by the FDS program. 5. Robustness checks In this section we perform a series of exercises to ensure that our estimates capture the true causal effect of the FDS reform on women s labor outcomes. We first discuss how the nonexperimental design of the implementation of the FDS reform may bias our results, followed by a discussion on whether the results are driven by pre-existing trends in women s labor force participation. Finally, we perform a series of robustness checks to eliminate alternative explanations for the observed results. Identification of FDS effects One possible concern with our estimates is that since the policy was not randomly implemented, there could exist endogeneity between women s labor decision and the the FDS program. To the extent that the gradual implementation of the reform is exogenous to women s labor decisions, we can be confident that our parameter of interest in equation (1) is correctly identified. This implies that the process through which public funds were allocated to schools is 18 We obtain this effect by adding the FDS coefficient and the interaction term with presence of preschool aged children. 19

20 critical to estimate the true effect of the policy. The funds allocation process was not random; as was discussed earlier, the Ministry of Education prioritized vulnerable schools with relatively poor students so that schools in poorer municipalities received FDS funds first. 19 If present, the direction of a bias depends on the correlation between municipal socioeconomic conditions and female labor force participation. In such a scenario, the estimated effect of the policy, which we denote as, can be expressed as:, (2) where is the true effect, and and are as defined above. Since access to schooling reduces childcare costs, it should facilitate mothers participation in the labor force so that the FDS reform is expected to have a positive effect (i.e., we expect > 0). If labor force participation of women is lower in poorer municipalities that received funds early during the period, the second term of equation (2) is negative and the estimated coefficient for the impact of FDS ( is biased towards zero, so that underestimates the true effect of the reform. If women s LFP is higher in poor municipalities that received funds earlier, then the second term of equation (2) is positive and over-estimate the true effect. To assess the direction of the possible bias in due to the program s non-experimental implementation, we estimate the following regression with municipality-level data: FDS (3) 19 Although parents can choose any public school for their child, evidence shows that families choose schools that are closest to where they live, so that the socio-economic profile of a school reflects the average socio-economic conditions of the area where it is located (see Chumacero and Paredes, 2012). Thus, schools located in municipalities with poor socio-economic conditions were given priority to FDS funds. 20

21 The dependent variable FDS measures the ratio of full-day primary schools in a municipality m in year t. 20 Vector includes municipal-level poverty rate, income per capita, unemployment rate, and level of primary school enrollment. To assess the direction of the bias discussed above, in a second specification we include municipal-level female (and male) labor force participation rates (vector ). The estimations also include year dummy-variables,, to control for the program s implementation over time as public investments grew, and an error term. We estimate equation (3) with a linear model using data from CASEN household surveys for years 1998, 2000, 2003, 2006 and Results are reported in Table 7. Columns (1) and (2) report results without and with controls for the female and male labor force participation rates, respectively. The only control variable that appears to be correlated with female and male labor force participation rates is the municipality unemployment rate, as could be expected. Results also confirm that FDS funds were allocated according to administrative guidelines, directed towards schools in municipalities with higher poverty rates and lower income per capita. Additionally, funds were first allocated to schools in municipalities with lower enrollment, and controlling for poverty and income, to municipalities with lower unemployment. Regarding the possible bias in our estimated effect due to the non-random implementation design, Column (2) of Table 7 reveals that FDS funds were assigned to municipalities where the female labor force participation rates were lower. Thus, the second term in equation (3) is negative and our estimate is most likely a to be lower-bound effect. Mean reversion in labor force participation 20 The FDS variable is constructed in the same way as described in previous sections. 21 We use the CASEN household surveys to construct a panel of municipal-level data because (i) it is representative at the municipal level, and (ii) it s available for the entire period of the FDS reform, whereas the EPS data is not representative at the municipality level and is only available since

22 It s possible that municipalities with lower female LFP rates witnessed a relatively faster increase in women s participation during the period because participation was initially at a lower level. Since the FDS funds were allocated to schools in areas with lower LFP rates first, then the observed positive effect of FDS on women s participation could be a spurious result of preexisting trends in women s participation. To assess whether this reversion to the mean is occurring, we follow both descriptive and analytical approaches. First, we denote municipalities that received FDS funds early in the implementation as early adopters and those that received funds later in the period as late adopters 22 and compare the evolution of female LFP rates between these municipalities. Descriptively, Figure 5 reveals that female labor force participation rates followed almost identical trends throughout the period, independent of whether the municipality was an early or late adopter of the FDS reform. This is suggestive evidence that the effect of FDS coverage on female participation is not due to mean reversion. In a second verification, we control for pre-existing trends in our regressions following the approach in Duflo (2001). We classify a municipality as Low LFP if its female participation rate in 1996 the year prior to the reform was below the median in that year s distribution, and then include inter-action terms of Low and year fixed effects in the baseline regressions. 23 This clears the FDS effect of any differences in female LFP trends across municipalities that may have been in place prior to the reform. We report results of these estimates in the first line of Table 8. We find no significant changes from previous estimates, thus, given our findings in the descriptive and 22 Early or late adoption is defined based on whether FDS coverage in the municipality in 2004 was less than 1 standard deviation below average implementation rate; 2004 marks the half-way point of the program s implementation. We used different definitions of early/late adopters and results were consistent; results not shown, available upon request. 23 We also used a definition of Low LFP if the municipality s female LFP in 1996 was in the lowest 25th percentile, and results were similar. Results are available upon request. 22

23 regression exercises, we are confident that the estimated effects of the policy are not driven by preexisting trends. Target Population As discussed earlier, our results indicate that the FDS policy had a strong impact on the group of women who would most likely benefit from the extended school schedule: mothers whose youngest child was of primary school age. Similarly, women without children should not respond to the FDS policy, and neither would men, since they have a more permanent attachment to the labor force. In Table 8 we report estimates of the effect of FDS availability on the three LFP variables for these two groups women without children and men and we find that neither of them are not affected by the availability of longer school days. Pre-school Policy As described in the Introduction, in 2006 Chile began to implement a large expansion of childcare and pre-school centers, affecting children between 3 months and 5 years of age. The childcare and pre-school expansion and FDS reforms were independent of each other, implemented by separate public institutions, and were not coordinated. When the pre-school expansion began, the FDS reform had been in place for 9 years. Nonetheless, to rule out the possibility that our estimated results are partially capturing the expansion in pre-school centers, we re-estimated our baseline regressions omitting the last year (2009) from our panel data, so that no overlap of the two policies is present in the estimations. These results are reported in the last line of Table 8, and we observe almost identical point estimates as the baseline regressions. 6. Conclusions This paper analyzed a national school reform in Chile that lengthened the school day from 23

24 half to full day schedules, and the effect it had on women s labor force participation decisions. We identify the effect of the policy from its quasi-experimental implementation. After ruling out several alternative explanations, we find that availability of FDS primary schools in the municipality significantly increases the likelihood that women participate in the labor force. If the supply of primary schools were to increase by 40 percentage points similar to moving to full FDS coverage female LFP would increase by almost 4% in Chile. This is a large increase, considering the country s low levels of women s participation in the labor force. The increased access to longer school days also had a positive impact on longer term labor force attachment, suggesting that improved childcare arrangements for older children can facilitate more stable employment for women. We also find that the effect of the policy is concentrated among groups of women with children in the age level to which the reform is targeted, relative to women with children of other age groups. The effects are strongest among mothers whose youngest child is in primary school: among this subgroup, increasing FDS availability to full coverage would increase the probability of working by 9.4 percent. We also find that longer-term attachment to the labor force measured as participating for at least half of the year in the labor force was positively affected among women with children in primary school. Our results are consistent with others in the literature that find positive effects of childcare expansion for pre-school children on women s labor decisions. Prior research for Chile finds that women have not responded to increased access to pre-school childcare centers which we corroborate in this paper however, in we find evidence that women respond to increased childcare provided by longer school days for primary school children. Additionally, our findings suggests that lack of childcare not only limits participation but also the quality of jobs that mothers 24