The World Bank Economic Review, 38(1), 2024, 74–94 https://doi.org10.1093/wber/lhad024 Article Public Pre-Primary and Maternal Employment in Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 Algeria: Evidence from a Natural Experiment Caroline Krafft and Moundir Lassassi Abstract Globally, employment rates of women remain substantially below those of men. Since women disproportion- ately care for children, policies that offer care alternatives or lower the cost of care, should, theoretically, increase women’s employment. This paper tests whether public pre-primary education can increase women’s employ- ment, using a natural experiment in Algeria. Education reforms in Algeria substantially expanded public pre- primary, targeting children aged five. The paper uses data from 2006 (early in the expansion), 2012, and 2018 (after pre-primary had substantially expanded). The analyses use a discontinuity in whether children are eligi- ble for pre-primary, based on their birthdates, to identify the effect of pre-primary on women’s employment. Increased pre-primary education did not increase and may have actually decreased women’s employment. One potential explanation for these findings is the half-day schedule of pre-primary may be difficult to reconcile with employment. Negative effects are concentrated among women living in nuclear families, who lack alternative caregivers. JEL classification: H52, I28, J21, J22 Keywords: employment, gender, pre-primary, education, Algeria 1. Introduction Globally, female employment rates are substantially below male employment rates. As of 2022, only 44 percent of women were employed compared to 68 percent of men (ILO 2022). Women’s employment rates are particularly low (below 20 percent) in the Middle East and North Africa (MENA) region (ILO 2022). Children and particularly the (opportunity) costs of working while raising children are potential drivers of low rates of female employment (Attanasio, Low, and Sánchez-Marcos 2008; Gathmann and Sass 2018). Since women disproportionately care for children, policies that offer care alternatives or lower the cost of care, should, theoretically, increase women’s employment. Expanding access to early childhood Caroline Krafft (corresponding author) is an associate professor at St. Catherine University, St. Paul, MN, USA; her email address is cgkrafft@stkate.edu. Moundir Lassassi is Research Director at the Center for Research in Applied Economics for Development (CREAD), Algiers, Algeria; his email address is lassassim@gmail.com. The authors appreciate the comments of participants in the Economic Research Forum’s 2020 annual conference, particularly discussant Rana Hendy, as well as partic- ipants in the Doha Institute for Graduate Studies School of Public Affairs Development Economics seminar. A supplementary online appendix is available with this article at the World Bank Economic Review website. C The Author(s) 2023. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com The World Bank Economic Review 75 care and education (ECCE) is one such policy. This paper tests the impact of expanding pre-primary,1 a form of ECCE, on women’s employment in Algeria. Expanding ECCE access, primarily through greater public provision or lower costs, can increase women’s employment, but its impacts are not guaranteed. Moreover, most of the evidence comes from high-income countries. Child care costs fully explain the lower participation of mothers of preschoolers in the United States (Connelly 1992). Estimates from the United States indicate that child care subsi- Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 dies increase the employment of (low-income) mothers between 12 and 33 percentage points (Berger and Black 1992; Crawford 2006; Blau and Tekin 2007; Davis et al. 2018). Public provision of pre-primary in high-income countries has more mixed results. Some studies have found no or little effect from expand- ing pre-primary, or only impacts on specific subgroups, such as single mothers (Cascio 2009; Fitzpatrick, 2010, 2012; Havnes and Mogstad 2011). Other studies, however, do find significant effects of pre-primary on women’s employment (Gelbach 2002; Baker, Gruber, and Milligan 2008). In low- and middle-income countries (LMICs), the evidence on the impact of ECCE to date is over- whelmingly positive. A review of 22 studies with plausibly causal identification strategies, examining the impact of institutional childcare on women’s labor market outcomes, found positive effects in 21 of the 22 studies (Halim, Perova, and Reynolds 2023). However, none of these studies were in settings with low rates of women’s employment.2 The mean baseline/control group rate of maternal labor-force par- ticipation across studies was 54 percent. The lowest baseline/control group rate of maternal labor force participation across the studies was 28 percent (Halim, Perova, and Reynolds 2023). Can pre-primary increase women’s employment in a context where their employment rates are very low? The paper answers this question in Algeria, where the rapid expansion of public pre-primary educa- tion facilitates analyses. The analyses use the cut-off age for pre-primary enrollment (age as of December 31 of that school year) to identify the effect of pre-primary on women’s employment using a regression discontinuity design (RDD), comparing 2006, 2012, and 2018. The results demonstrate that the expansion of pre-primary did not increase women’s employment. If anything, it actually decreased their employment. To the best of the authors’ knowledge, this paper is the first to show that pre-primary schooling may actually reduce women’s employment. This finding is an important contribution and a caution for assuming that provision of pre-primary has unidirectional and positive effects on women’s employment outcomes. One potential explanation for these findings is that the half-day schedule of expanded pre-primary may be difficult to reconcile with employment. The one study of 22 in the review of institutional childcare on women’s labor market outcomes (Halim, Perova, and Reynolds 2023) that did not find a positive effect noted that care was only part-day and did not align well with work schedules (Medrano 2009). Studies elsewhere show full-day kindergarten or afterschool care can increase women’s employment (Cannon, Jacknowitz, and Painter 2006; Berthelon, Oyarzún, and Kruger 2015; Martinez A. and Perticará 2017; Dhuey, Lamontagne, and Zhang 2019). These results may also be due to the challenging and low female employment-rate context in Algeria. The same study of the 22-paper review that did not find positive effects of childcare was also in the lowest- participation context of all the studies (28 percent) (Medrano 2009; Halim, Perova, and Reynolds 2023). When reconciling employment and care responsibilities is already especially difficult, or in contexts with weak female labor demand, pre-primary may not be sufficient to change women’s employment outcomes. 1 Pre-primary is defined, per the International Standard Classification of Education (ISCED), as programs targeting ages three to five with educational objectives (UNESCO 2021). 2 Schlosser (2011) examines the impact of Israel’s expansion of preschool on Arab mothers (who have low rates of participation) within Israel and finds an increase in participation. 76 Krafft and Lassassi Careful policy design around hours of care may also be particularly important when attempting to increase women’s employment in such lower-employment contexts.3 The paper is organized as follows. Section 2 provides background on pre-primary and the labor mar- ket in Algeria. The paper then describes the survey data in section 3. Section 4 describes the identification strategy of RDD. Section 5 presents the results and a number of mechanisms, specifications, and robust- ness checks. Section 6 concludes with a discussion of implications for policy and increasing women’s Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 employment globally. 2. Background 2.1 Algeria’s Labor Market A rich literature examines the determinants of women’s employment in MENA countries. The deter- minants can generally be organized into “needs,” the economic and care requirements of the household, “values” including gender norms, and “opportunities,” whether jobs are accessible and suitable (Spierings, Smits, and Verloo 2010). Empirically, women’s participation and employment depends on household com- position, especially the presence of young children (Keo, Krafft, and Fedi 2022). However, little is known about what programs or policies might increase women’s employment in the MENA region. In particular, there is limited empirical evidence on the effects of child care costs, an important component of women’s opportunity costs, for women in the MENA region. Algeria is one of the many MENA countries with low female employment rates, estimated at 13.8 percent as of 2019 (Algeria National Statistics Office 2019). This compares to an employment rate of 60.7 percent for men. Algeria had the fourth-largest (of 146 countries) increase in years of schooling from 1980 to 2010 (Campante and Chor 2012). Recent cohorts have achieved gender parity in education (Assaad et al. 2020). Despite increases in women’s education, their employment has only increased very slightly over time, from 11.4 percent in 2009 to 13.8 percent in 2019 (Algeria National Statistics Office 2019). Participation is higher for unmarried than for married women (Assaad et al. 2020). This suggests that domestic responsibilities, such as child care, that come with marriage and family formation, are a key constraint on women’s employment. 2.2 Education System in Algeria Algeria’s official education system begins with pre-primary education. Children then proceed to primary education (starting at age six). There are two types of pre-primary education in Algeria (Mahdjoub 2017). Preparatory education is one year in duration and starts at age five. Preschool education is officially three years and starts at age three but may be two years starting at age four. Both preparatory and preschool education are forms of pre-primary, and this paper refers to them collectively as pre-primary. When the paper describes features related to one type of pre-primary specifically it names that type as, dis- tinctly, preparatory or preschool. Pre-primary education is voluntary (Mahdjoub 2017). Public prepara- tory classes are physically and administratively attached to primary schools, overseen by the Ministry of Education (Secretary General of the Republic of Algeria 2008). Preschool is implemented by a variety of actors, including other ministries, such as the Ministry of Religious Affairs, and employers (Mahdjoub 2017). When children can enroll in school is determined by their birthdate. Children can enroll in preparatory starting the calendar year in which they will turn five. For example, a child born between January 1 and December 31, 2016, would enroll in preparatory for the 2021–2022 school year (Chalal 2018; H. 2021). The analyses exploit the discontinuity this policy creates for the RDD identification strategy. Children 3 For example, Schlosser (2011) shows that a full-day preschool expansion increased the participation of Arab mothers in Israel. The World Bank Economic Review 77 born between January 1 and March 31 can, in some cases, petition to be accepted a year early (Chalal 2018). Primary and pre-primary school days are short and may be difficult to reconcile with employment. For instance, starting in 2011, preparatory students and those in the first two years of primary had only 21 hours of school per week. They finished school earlier, at 2:30 p.m., as well as having a lunch break between 11:15–1:00 p.m. and no class on Tuesday afternoons (Nawel 2011; UNESCO-IBE 2012). Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 2.3 Pre-primary Expansion in Algeria After its independence, in 1962, Algeria did away with public pre-primary education to focus on achiev- ing universal compulsory (primary and lower secondary school) education (Bouzoubaa and Benghabrit- Remaoun 2004). With its 2003 education reform, Algeria planned to expand pre-primary once more starting in the 2004–2005 school year. In January of 2008, the National Education Guideline Law No. 08–04 further emphasized pre-primary education (Mahdjoub 2017). Algeria massively expanded pre-primary in a period of less than five years (fig. S1.1, in the supple- mentary online appendix, available with this article at The World Bank Economic Review website). The expansion focused on children aged five (the preparatory year) (UNICEF Algeria 2014). Starting in 1993, gross enrollment in pre-primary (for five-year-olds) was only 1 percent. As recently as 2005, the gross enrollment rate in pre-primary was only 6 percent, but starting in 2006, enrollments rapidly increased, to 29 percent in 2006, 36 percent by 2008, and 80 percent by 2009 (then 84 percent in 2010 and 79 percent in 2011).4 Of primary school entrants at the beginning of the 2015 school year, 70 percent had attended pre-primary (Mahdjoub 2017). Among those who attended pre-primary, 71 percent had attended a preparatory class, 25 percent a Ministry of Religious Affairs preschool, and the rest other private or public programs (Mahdjoub 2017). This paper uses data from 2006 (early in the expansion), 2012, and 2018 (after pre-primary had substantially expanded). 3. Data The identification strategy is a RDD, based on the cut-off age for pre-primary enrollment (age as of December 31 of that school year). This section discusses the three rounds of survey data, overall sample, and analysis sample. The analysis sample focuses on children whose age is within six months above or below the cut-off for those who would be age four or age five on December 31 of that school year. This section also describes the outcome (women’s employment) and the measure of pre-primary. The next section provides the exact RDD estimation methods. 3.1 Surveys This paper uses data from the Algeria Multiple Indicator Cluster Surveys (MICS), which were conducted in 2006,5 2012/13 (referred to as 2012), and 2018/19 (referred to as 2018) by UNICEF in collabora- tion with the Algerian government (Ministry of Health Population and Hospital Reform and National Office of Statistics 2008; Ministry of Health Population and Hospital Reform 2015; Ministry of Health Population and Hospital Reform and United Nations Children’s Fund (UNICEF) 2020).6 The MICS is a nationally representative survey, representative of both households and individuals (including subgroups 4 The 2011 data is the most recent reported as of January 2023. 5 There were rounds of the MICS in 1995 and 2000, but they only asked about education in the household roster for ages 6+ (Ministry of Health and Population Republic of Algeria - National Institute of Public Health 2001; Ministry of Health and Population Republic of Algeria 1996). 6 The 2006 data were received in correspondence with the Ministry of Health Population and Hospital Reform, 2012 and 2018 data are publicly available from https://mics.unicef.org/surveys. 78 Krafft and Lassassi such as women or women with young children).7 The MICS contains four questionnaires: a household questionnaire, a household member questionnaire, a questionnaire for women aged 15–49, and a ques- tionnaire for children under the age of 5 (addressed to the mother or primary caretaker of the child). For the survey conducted in 2006, 29,008 households, 43,642 women, and 15,000 children under the age of five were included (Ministry of Health Population and Hospital Reform and National Office of Statistics 2008). Fielding took place from March 2006 to June 2006. For the survey conducted in 2012, Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 27,198 households, 38,548 women, and 14,701 children under the age of five were included (Ministry of Health Population and Hospital Reform 2015). Fielding took place from October 2012 to March 2013. For the survey conducted in 2018, 29,919 households, 35,111 women, and 14,873 children under the age of five were included (Ministry of Health Population and Hospital Reform and United Nations Children’s Fund (UNICEF) 2020). Fielding took place from December 2018 to April 2019. Thus, while some of the surveys span two calendar years, none span multiple school years, facilitating the identification strategy. Because the identification strategy (described in detail below) focuses on mothers with children near the age cut-offs for pre-primary, the analysis samples are somewhat smaller: 2,499 observations around the age four cut-off and 2,667 around the age five cut-off in 2006; 2,666 around the age four cut-off and 2,678 around the age five cut-off in 2012; and 2,976 around the age four cut-off and 2,996 around the age five cut-off in 2018. 3.2 Outcome Measures The paper investigates the impact of pre-primary on women’s employment. The survey specifically asks for each household member (aged 15+) their situation in the 30 days preceding the interview, classified as: (1) employed (2) studying/training (3) searching for work (unemployed) (4) retired, or (5) other out of labor force. The specific outcome the analyses examine is employment (as a percentage of the population). 3.3 Pre-Primary Measures There is information on pre-primary from both the under-five (for ages three and four) questionnaire and the household member questionnaires (for ages five and six).8 Note that these questionnaires and questions were based at the age at the time of the survey. This age-at-time-of-survey distinction in the questionnaires does not affect the discontinuity used for identification of age on December 31 of that school year.9 For ages three and four, their parent or guardian is asked whether the child attends a preschool education pro- gram such as a public or private center or kindergarten.10 A “yes” response to this question is considered attending pre-primary for ages three to four. For ages five and six, a question on whether or not the child 7 Data are representative after the application of sampling weights, which are used throughout. Sampling weights account for nonresponse. The response rate was 98 percent in 2006, 98 percent in 2012, and 96 percent in 2018 on the household level (outcomes such as employment are from the household roster). The sampling frame for each MICS was the most recent population census. 8 While in 2006 and 2012 education questions in the household module were asked of ages 5–24, in 2018 they were asked of ages 3–24, and a skip included in asking about preschool in the under-five module based on the education status in the household module, so effectively both modules are used for ages three to four in 2018. 9 For example, a child who was born December 1, 2007, who was interviewed on October 31, 2012, would be age five (by 30 days) on December 31, 2012, but age four at the time of interview and asked the age-four questions. A child who was born December 1, 2007, who was interviewed on January 15, 2013, would be age five (by 30 days) on December 31, 2012, but age five at the time of interview and asked the age-five questions. 10 The exact question is slightly different over time. The 2006 question focuses on out of home or preschool programs and includes Koranic programs in the list of examples (2012 does not, but 2018 does), whereas the 2012 question mentions educational learning programs and nurseries (2006 does not, but 2018 does), and 2018 also explicitly mentions the preparatory year (Ministry of Health Population and Hospital Reform 2015; Ministry of Health Population and Hospital Reform and National Office of Statistics 2008; Ministry of Health Population and Hospital Reform and United Nations Children’s Fund (UNICEF) 2020). The World Bank Economic Review 79 is attending school in the current school year is used (2005–2006 for 2006; 2012–2013 for 2012; 2018– 2019 for 2018). If the child is not attending school, he or she is not in pre-primary. If she or he is attending school, if he or she is in the preparatory or preschool level, she or he is classified as attending pre-primary. The pre-primary measure thus includes a mix of preparatory school and preschool types (the data do not allow us to distinguish the type of pre-primary). Some five-year-olds and most six-year-olds are already in primary school, and analyses count them as missing (since they are in primary, they are ineligible for Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 pre-primary) for the pre-primary measure. Moreover, some six-year-olds at the time of the survey might have been five in terms of eligibility, and therefore eligible and are included in the pre-primary measure. Essentially, analyses measure pre-primary for children under age six as of December 31 in the relevant school year, who would have been eligible for pre-primary. 3.4 Covariates Analyses test for differences in covariates as a check on RDD assumptions and also include controls in a sensitivity analysis for the RDD estimation. In terms of controls, analyses include data on the location of the household, both in terms of urban vs. rural residence and region (seven regions are used). Since household composition will affect childcare options and labor force decisions, analyses also control for whether the household is extended compared to nuclear (nuclear includes only the head, spouse, and their children). For women, analyses include information on their education level, age group, and marital status. 4. Methods 4.1 Fuzzy Regression Discontinuity Design As with past studies (e.g., Fitzpatrick 2010; Berlinski, Galiani, and McEwan 2011; Dang, Masako, and Nguyen 2019), this paper uses a fuzzy RDD (FRDD) strategy and exploits policies that set a cut-off date for pre-primary eligibility. The essential idea of RDD is that an individual’s “treatment” status (pre- primary) is determined by some “assignment variable” (Imbens and Lemieux 2008; Lee and Lemieux 2010). In this case, children are eligible for a particular grade of pre-primary only if they are of an age for that pre-primary grade as of December 31 of that school year11 (the cut-off) (Chalal 2018; H. 2021). Therefore, weeks from the age cut-off at December 31 of that round’s school year is the assignment vari- able. The data have the full date of birth (day, month, year) for all children to facilitate this identification strategy. If the birth date is December 31 or earlier, this assignment variable is zero or positive, and the child can enter that grade of pre-primary sooner. If the birth date is later, this value is negative, and the child has to wait until the next school year to enter that grade of pre-primary. Since children can enter pre-primary at various ages, analyses divide the sample into discontinuity groups, based on the age on December 31, for the December 31 date included in that round’s school year. Specifically, analyses use six months older or younger than would turn age four on December 31 of that round’s school year, and likewise for age five.12 Analyses thus have two groups, what are referred to as the age four group (3.5–4.5 years old on December 31) and the age five group (4.5–5.5 years old on December 31). Children thus fall into one (and only one) discontinuity group and are near equally distributed on each side of the discontinuity.13 Children in the age five group whose age was above the cut-off would be eligible to enter the preparatory grade of pre-primary, which substantially expanded over time. Children 11 This date is therefore age December 31 of 2006 for the 2006 round, December 31 of 2012 for the 2012/13 round, and December 31 of 2018 for the 2018/19 round. 12 This is the same as in Berlinski, Galiani, and McEwan (2011), except in their case the cut-off was June 30 rather than December 31. 13 In the rare cases where a mother had more than one child in the same discontinuity group, the information for the youngest child was used. 80 Krafft and Lassassi below the cut-off may, however, still access other types of pre-primary. Children in the age four group whose age was above the cut-off would be able to access preschool grades for age four. This level did not, however, experience much of an expansion, and thus the age four group acts in some ways as a placebo relative to age five. 4.2 Polynomial and Bandwidth Selection Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 Visually inspecting the relationships between the treatment and assignment variable around the cut-off, as well as the outcome and assignment variable is critically important. Analyses initially undertake a visual inspection and “bin” the data, estimating bin means locally throughout the distribution on each side of the cut-off, the standard non-parametric approach.14 This approach allows for visual inspection of the data for relationships, in order to confirm whether there is a discontinuity, to see potential functional forms, and to check for any violations of key assumptions. Analyses use the typical FRDD estimator of the treatment effect, namely the ratio of the jump in the outcome (employment) to the jump in the probability of treatment (pre-primary) (Imbens and Lemieux 2008). In the first stage, it is critically important for there to be a discontinuity, that is, for being above the cut-off to effectively predict pre-primary.15 Only if this assumption holds can analyses possibly identify the effect of pre-primary on women’s employment.16 The main results are based on local non-parametric polynomial kernel estimators estimated on either side of the cut-off. Analyses estimate both conventional, bias-corrected, and bias-corrected robust (robust for short) treatment effects.17 The bias-correction method uses local linear regression to address leading bias when using kernel regressions (Calonico, Cattaneo, and Titiunik 2014a). This approach is more ro- bust to large bandwidth choices, improving power. The robust standard error accounts for the additional variability generated by the bias (Calonico, Cattaneo, and Titiunik 2014a). Analyses use a rectangular ker- nel, as recommended (Imbens and Lemieux 2008; Lee and Lemieux 2010). The conventional treatment effect estimates with this local non-parametric approach thus are the same as parametric estimates with an equivalent specification using two-stage least squares (2SLS).18 Analyses additionally present those 2SLS estimates, with their robust standard errors. Given the well-known problems with higher-order polynomials (Gelman and Imbens 2019) and lacking any reason to expect a higher-order polynomial is applicable in the case at hand, analyses follow the recommendations of Gelman and Imbens (2019) and estimate at most a quadratic, as well as linear and differences in means estimates. Analyses empirically test which model fits best, based on the reduced form model using being above the cut-off rather than treatment itself (Lee and Lemieux 2010).19 The preferred model is the one with the smallest Akaike information criterion (AIC), the recommended approach to choosing the polynomial’s order (Lee and Lemieux 2010). The paper presents sensitivity analyses with the alternative polynomials as well. Analyses prioritize the functional form for the outcome and use the 14 Analyses use the Stata package rdplot (Calonico, Cattaneo, and Titiunik 2014b). 15 Denote treatment (pre-primary) as Ti , the running variable (weeks of age relative to December 31 cut-off) as Xi , c as the cut-off, and Di as an indicator variable for being above the cut-off (Di = 1[Xi ≥ c]). The effect of the assignment rule on treatment (attending pre-primary) is then: Ti = α + f (Xi − c) + β Di + ε i , where f( ) is some polynomial functional form and ε i an independent error term. 16 Analyses estimate the effect of pre-primary on women’s employment (Yi ) based on: Yi = γ + g(Xi − c) + τ Ti + vi , where g() is some polynomial functional form and vi an independent error term. 17 Analyses use the Stata package rdrobust (Calonico, Cattaneo, and Titiunik 2014b). 18 The instrument for treatment (pre-primary) is the indicator for being above the cut-off (Imbens and Lemieux 2008). If the model assumptions hold, being above the cut-off will predict treatment (pre-primary) but be otherwise unrelated to outcomes (employment). 19 Analyses specifically estimate Yi = γ + g (Xi − c) + τ Di + vi . Here, τ is the intent-to-treat estimator (Lee and Lemieux 2010). The World Bank Economic Review 81 same functional form (polynomial) for the treatment equation as well (Imbens and Lemieux 2008; Lee and Lemieux 2010). In addition to questions of polynomial selection and estimation, the bandwidth from the cut-off within which results are estimated is an important decision for RDD. Analyses take as their starting point a bandwidth that includes observations within six months above and below the December 31 cut-off for each group (group above/below age four cut-off, group above/below age five cut-off). The paper tests the Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 sensitivity of the results to smaller bandwidths (e.g., within three months above/below the cut-off). In general, bandwidth selection is a trade-off between precision and bias, where additional curvature in the underlying relationship will mean more bias, and a smaller optimal bandwidth (Lee and Lemieux 2010). As recommended, analyses use the same bandwidth for estimates of both the treatment and outcome (Imbens and Lemieux 2008). If covariates are balanced and the assignment variable is as-good-as-random, covariates are irrelevant to estimation (Imbens and Lemieux 2008; Lee and Lemieux 2010). Their inclusion can, however, increase precision, so sensitivity analyses include estimations with covariates. 4.3 Assumptions and Threats to Identification There are a number of assumptions that need to hold for FRDD to identify a causal parameter. If there is a change in outcome (employment), at the treatment threshold, it can be attributed to the treatment (pre-primary) so long as one can expect a smooth (continuous) relationship between the outcome and assignment variable in the absence of treatment. In this case, aside from the ability to enroll children in pre-primary, the effect of a child’s birth date moving from a few weeks before December 31 to a few weeks after should be no different from an equivalent shift at any other date. The FRDD also assumes monotonicity (Imbens and Lemieux 2008; Lee and Lemieux 2010). In this case, that means that crossing the cut-off does not cause some children to take up pre-primary and others to reject pre-primary. This is at least conceptually plausible; it cannot be directly tested, but analyses confirm below there is the expected jump in pre-primary at the cut-off. One concern with RDD identification strategies is potential for precise control of the assignment vari- able (Imbens and Lemieux 2008; Lee and Lemieux 2010). Conceptually, intentional manipulation is less likely to be the case with a child’s birthdate occurring several years in the past relative to pre-primary than with some other assignment variables (e.g., income is much more manipulable). Parents have some control over when children are born, but much less so the exact date, precluding precise control. Age/birthdate is a very common assignment variable in the literature for this reason (Lee and Lemieux 2010). Nonetheless, analyses check the distribution of the assignment variable as a credibility check (Imbens and Lemieux 2008; Lee and Lemieux 2010). Analyses also check for balance of other covariates above and below the cut-off, as if the birthdate is as-good-as-random, other covariates should be orthogonal to whether a child is above or below the cut-off (Imbens and Lemieux 2008; Lee and Lemieux 2010).20 This test is similar to excludability of the cut-off; a child being a few weeks older cannot be related to labor market outcomes except through their pre-primary enrollment. Two other issues that can threaten identification are anticipatory behavior and general equilibrium ef- fects. If women anticipate that they will soon receive child care, when their children reach the pre-primary threshold, they may change their behaviors, for instance, begin to work or seek work. The jump in the outcome may thus be somewhat attenuated. In terms of general equilibrium effects, if women increase their labor supply in response to pre-primary, this could shift wages and opportunities for women with children below the threshold. For both issues, these effects are arguably second-order concerns (anticipa- tory effects of child care are going to be much smaller than actual child care effects; general equilibrium 20 Additionally, to the best of the authors’ knowledge, there are not any other programs or services for young children in Algeria that use this same cut-off. 82 Krafft and Lassassi effects will be small given that a fraction of women have children of this age, some fraction of whom might change their labor market behavior). Second, both these issues will attenuate estimates, making the results conservative. If these assumptions hold, then the FRDD estimates the treatment effect based on compliers (children who attend pre-primary because their birthdate is older than the cut-off, but who would not have oth- erwise attended) (Imbens and Lemieux 2008; Lee and Lemieux 2010). The resulting estimate is a local Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 average treatment effect (LATE) (Lee and Lemieux 2010). The results thus do not necessarily generalize to what would happen if there were different cut-offs from those currently implemented. 5. Results This section begins the presentation of results with an investigation of pre-primary enrollment, investi- gating the potential discontinuity in pre-primary enrollment based on birthdates. Analyses then test two key assumptions of FRDD, checking the distribution of birthdates and balance of covariates around the cut-off. The FRDD results for women’s employment are then presented both visually and for a number of specifications and sensitivity analyses. The final results subsection explores potential mechanisms. 5.1 Pre-Primary Results Figure 1 presents pre-primary attendance by weeks from the cut-off, age group (+/- six months of being age four or five on December 31), and year. The points are the bin means (six bins on each side of the cut-off, thus roughly months).21 The bins are used for descriptive visualization, but the full underlying distribution is retained for subsequent analyses. The figure also presents confidence intervals around the bins and a linear fit. Notably, in 2006 (early in the expansion), for age four, there is no discontinuity. There is a visible discontinuity at age five in 2006. In 2012 and 2018, there is evidence of a small/slight discontinuity for age four. The discontinuity for age five in 2012 is modest, but larger in 2018. There is also a clear increase in pre-primary attendance with age (weeks older than Dec. 31), not just at the cut-off. The relationship is roughly linear. In the supplementary online appendix, fig. S1.2 and fig. S1.3 present intercept (mean only) and quadratic fits; visually, the intercept (mean only) fit is fairly similar or slightly worse than the linear specification. The quadratic fit is fine in some cases, but in others is very ill fitting and does not show a consistent functional form for the quadratic across years and age groups. 5.2 Checks of FRDD Assumptions A critical assumption of the FRDD approach is the absence of assignment variable manipulation. Figure 2 presents a histogram of the distribution of children’s age in weeks from the cut-off, by age group and year, as a check of this assumption. The distribution appears random relative to the cut-off; there is not, for example, a much higher share of births just a few weeks older than the cut-off. No other point shows unusual or systematic density. Manipulation test procedures based on local density also generally support lack of manipulation.22 Across the two age groups and three time periods (six tests), age four cut-offs are never significant. The age five cut-off is significant in 2006 using both the robust and conventional estimators, in 2012 with the conventional but not robust estimator, and in 2018 with the robust but not 21 As well as the intuitive appeal of months, the integrated mean squared error optimal bins (calculated with the rdplot package (Calonico, Cattaneo, and Titiunik 2014b)) tend to be around six on each side: six on the left and nine on the right for age five in 2018, six on the left and four on the right for age five in 2012, three on the left and six on the right for age five in 2006, 11 on the left and seven on the right for age four in 2018, five on the left and right for age four in 2012, five on the left and eight on the right for age four in 2006. It is preferable to consistently present the same number of bins for comparability in the visual. 22 Tests are implemented using rddensity version 2.3 (Cattaneo, Jansson, and Ma 2018). The World Bank Economic Review 83 Figure 1. Pre-Primary Attendance (Percentage) by Weeks from Cut-off, Age Group, and Year Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 Source: Authors’ calculations based on Algeria MICS 2006, 2012, 2018. Note: “Age [X] cut-off” refers to the group of children within (+/-) six months of being age [X] on December 31 of that school year (the cut-off). Points are bin means for six bins on each side of the cut-off (roughly months), which were the most common integrated mean squared error optimal bins. Shading represents 95 percent confidence intervals. The line is a linear fit. conventional estimator. Although in some cases tests are significant, differences in density are small and also have variable directionality across the samples, corroborating lack of manipulation. If the assumptions of RDD hold, there will not be any differences in covariates around the cut-off. Table S2.1, in the supplementary online appendix, tests that assumption, presenting the means of various mother characteristics23 above and below the threshold for each age group cut-off, their difference, and the t-test. Note that one would expect some significant differences by chance with this many tests. There are few significant differences, and they are not consistent ones; for example in 2012 for the age four cut-off there is a four percentage point difference (28.4 percent below; 32.4 percent above) in middle education, but in other years this difference is insignificant or the sign reversed. F-tests for overall balance are significant, except for age four in 2006, but with small F-statistics (1.099–2.456). The results suggest the cut-off is as-good-as-random or nearly so, supporting the identification strategy. 5.3 Employment of Mothers: FRDD Estimates The presentation of the FRDD results starts with the “reduced form” model (using being above or below the cut-off as the key covariate) to identify the correct functional form, based on the AIC. Analyses test 23 Table S2.1 in the supplementary online appendix also allows an assessment of sample characteristics generally. For instance, fewer than 1 percent of mothers are single; parenthood is almost exclusively reserved for within marriage in Algeria. 84 Krafft and Lassassi Figure 2. Histogram (Proportion of Observations) of Age in Weeks from Cut-off, by Age Group and Year Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 Source: Authors’ calculations based on Algeria MICS 2006, 2012, 2018. Note: Showing 95 percent confidence intervals for linear polynomial, robust standard errors. This figure is a test for assignment variable manipulation. simple intercept (mean only), linear, and quadratic functional forms for the employment outcome. The results for age four are in table 1 and for age five in table 2. First, in terms of fit, the AIC (smallest is best) indicates that an intercept-only functional form (polynomial order zero) tends to be the best fit, although in 2012 quadratic forms fit better. In order to be consistent, analyses use the intercept-only functional form as the main model but present sensitivity analyses with the other models. The reduced form models make it possible to assess the statistical significance of the results as well. In the preferred, intercept-only model, being above the cut-off does not have a statistically significant relationship with employment for the age-four group but does for age five. The contrast between age four and age five also bolsters the results, since the expansion targeted age five. In 2006, having a child above the cut-off caused significantly higher employment, by around 3 percentage points (relative to a base employment rate of 9 percentage points). In 2012, employment decreased a significant 3 percentage points if the child was above the cut-off (qualified for pre-primary). In 2018, employment was also a significant 3 percentage points lower. The expansion of pre-primary in Algeria thus appears to have worsened, rather than improved, women’s employment outcomes for women with eligible children. Next are the visualizations of these employment outcomes around the discontinuity. Figure 3 presents the intercept (mean only) FRDD fits for the outcome of employment with binned data (the points are the bin means (six bins on each side of the cut-off, thus roughly months)) above and below the cut-off. The bins are used for descriptive visualization, but the full underlying distribution is retained for subsequent analyses. In the supplementary online appendix, fig. S1.4 presents a linear fit while fig. S1.5 presents a quadratic fit. The intercept-only functional form appears, visually, to be an appropriate fit. The linear and Table 1. Reduced-Form FRDD Intercept, Linear, and Quadratic Models of Employment (Percentage Points) by Year, Age Four 2006 2012 2018 The World Bank Economic Review Above cut-off − 0.501 − 3.471 − 3.375 0.161 − 0.105 − 1.161 0.015 − 0.502 3.057 (1.216) (2.384) (3.621) (1.051) (2.190) (3.336) (1.166) (2.379) (3.562) Age in weeks from cut-off (age 4) 0.129 0.556 0.026 1.285∗∗ 0.041 − 0.145 (0.118) (0.504) (0.103) (0.433) (0.115) (0.489) Above cut-off # Age in weeks from cut-off − 0.028 − 0.917 − 0.032 − 2.365∗∗∗ − 0.044 − 0.509 (age 4) (0.159) (0.650) (0.142) (0.578) (0.151) (0.624) Age in weeks from cut-off (age 4) # Age in 0.016 0.048∗∗ − 0.007 weeks from cut-off (age 4) (0.019) (0.016) (0.018) Above cut-off # Age in weeks from cut-off 0.001 − 0.007 0.032 (age 4) # Age in weeks from cut-off (age 4) (0.024) (0.021) (0.023) Constant 10.521∗∗∗ 12.148∗∗∗ 14.018∗∗∗ 7.914∗∗∗ 8.259∗∗∗ 14.026∗∗∗ 11.351∗∗∗ 11.908∗∗∗ 11.068∗∗∗ (0.885) (1.735) (2.756) (0.751) (1.570) (2.482) (0.848) (1.763) (2.773) N (obs.) 2499 2499 2499 2666 2666 2666 2976 2976 2976 R-sq. 0.000 0.001 0.002 0.000 0.000 0.007 0.000 0.000 0.001 Adj. R-sq. − 0.000 − 0.000 − 0.000 − 0.000 − 0.001 0.005 − 0.000 − 0.001 − 0.001 AIC 24151 24153 24155 25167 25171 25158 29027 29031 29032 Source: Authors’ calculations based on Algeria MICS 2006, 2012, 2018. Note: ∗ p < 0.05; ∗∗ p < 0.01; ∗∗∗ p < 0.001. OLS models. Standard errors in parentheses. Outcome is employment (as a percentage). 85 Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 86 Table 2. Reduced-Form FRDD Intercept, Linear, and Quadratic Models of Employment (Percentage Points) by Year, Age Five 2006 2012 2018 ∗ ∗ ∗ ∗∗ Above cut-off 2.686 1.865 4.271 − 2.639 − 5.118 − 6.128 − 3.328 − 4.221 − 4.270 (1.160) (2.272) (3.413) (1.141) (2.309) (3.458) (1.142) (2.273) (3.431) Age in weeks from cut-off (age 5) 0.006 − 0.970∗ 0.284∗∗ − 0.168 0.167 − 0.054 (0.114) (0.466) (0.105) (0.445) (0.110) (0.462) Above cut-off # Age in weeks from cut-off (age 0.051 1.491∗ − 0.386∗ 0.739 − 0.260 0.210 5) (0.153) (0.612) (0.150) (0.606) (0.149) (0.606) Age in weeks from cut-off (age 5) # Age in − 0.038∗ − 0.017 − 0.008 weeks from cut-off (age 5) (0.018) (0.017) (0.017) Above cut-off # Age in weeks from cut-off (age 0.020 − 0.008 − 0.001 5) # Age in weeks from cut-off (age 5) (0.023) (0.022) (0.022) Constant 8.630∗∗∗ 8.703∗∗∗ 4.388 10.930∗∗∗ 14.760∗∗∗ 12.794∗∗∗ 12.684∗∗∗ 14.841∗∗∗ 13.846∗∗∗ (0.824) (1.647) (2.590) (0.794) (1.624) (2.482) (0.814) (1.636) (2.603) N (obs.) 2667 2667 2667 2678 2678 2678 2996 2996 2996 R-sq. 0.002 0.002 0.004 0.002 0.005 0.007 0.003 0.004 0.004 Adj. R-sq. 0.002 0.001 0.003 0.002 0.004 0.005 0.002 0.003 0.002 AIC 25706 25710 25707 25731 25727 25727 29128 29129 29132 Source: Authors’ calculations based on Algeria MICS 2006, 2012, 2018. Note: ∗ p < 0.05; ∗∗ p < 0.01; ∗∗∗ p < 0.001. OLS models. Standard errors in parentheses. Outcome is employment (as a percentage). Krafft and Lassassi Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 The World Bank Economic Review 87 Figure 3. Employment (Percentage) by Weeks from Cut-off, Age Group, and Year Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 Source: Authors’ calculations based on Algeria MICS 2006, 2012, 2018. Note: “Age [X] cut-off” refers to the group of children within (+/−) six months of being age [X] on December 31 of that school year (the cut-off). Points are bin means for six bins on each side of the cut-off (roughly months). Shading represents 95 percent confidence intervals. The line is an intercept (mean only) fit. quadratic fits lack a clear pattern across ages and years in terms of the slopes, slope on each side of the cut-off, and concavity or convexity (for the quadratic form). The results of the FRDD estimator for this specification are presented in table 3 (for age four) and table 4 (for age five). The tables present the first-stage estimates (the percentage point jump in pre-primary at the discontinuity), and the treatment effects (the percentage point change in employment for a 1 percentage point increase in pre-primary, based on the discontinuity). Conventional and bias-corrected estimates are presented, as well as robust standard errors for the bias-corrected estimates. There are no significant treatment effect estimates (although the first stage under the conventional estimates is strong) for age four except a negative effect with the bias-corrected estimate in 2006 (but this is not significant with the robust standard error). For five-year-olds, in 2006, pre-primary has a significant and positive effect on employment with the conventional estimates (does not remain significant in the bias-corrected or robust models). In 2012, pre- primary has a negative effect on employment, which becomes larger and significant in the bias-corrected models (remains significant with the robust standard errors). In 2018, pre-primary has a significant nega- tive effect on employment in the conventional estimators, which increases in the bias-corrected estimators, remaining significant (but not with the robust standard errors). In sum, although there are a number of estimates and the significance of results is sensitive to the estimator and the standard error used, the bal- ance of evidence indicates pre-primary’s expansion in 2012 and 2018 worsened women’s employment outcomes. 88 Krafft and Lassassi Table 3. FRDD Intercept Models of Employment (Percentage Points) by Year, Age Four 2006 2012 2018 Conventional − 0.081 0.016 0.001 (0.236) (0.133) (0.099) Bias-corrected − 0.655∗∗ − 0.001 − 0.031 Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 (0.236) (0.133) (0.099) Robust − 0.655 − 0.001 − 0.031 (0.482) (0.298) (0.211) First conventional 6.158∗∗∗ 10.227∗∗∗ 15.863∗∗∗ (1.454) (1.903) (1.772) First bias-corrected − 0.758 4.295∗ 12.222∗∗∗ (1.454) (1.903) (1.772) First robust − 0.758 4.295 12.222∗∗ (2.885) (3.939) (3.737) N (Obs.) 2499 2666 2976 N right 1321 1391 1564 N left 1178 1275 1412 Source: Authors’ calculations based on Algeria MICS 2006, 2012, 2018. Note: ∗ p < 0.05; ∗∗ p < 0.01; ∗ ∗ ∗ p < 0.001. Models based on rdrobust. Standard errors in parentheses. Outcome is employ- ment (as a percentage). “First” denotes the first-stage discontinuity in pre-primary. N right denotes the number of observations to the right of the cut-off; N left denotes the number of observations to the left of the cut-off, and N (Obs.) is the total number of observations (sum of N left and N right). Table 4. FRDD Intercept Models of Employment (Percentage Points) by Year, Age Five 2006 2012 2018 Conventional 0.180∗ − 0.113 − 0.090∗ (0.089) (0.069) (0.044) Bias-corrected 0.116 − 0.293∗∗∗ − 0.157∗∗∗ (0.089) (0.069) (0.044) Robust 0.116 − 0.293∗ − 0.157 (0.167) (0.148) (0.091) First conventional 14.899∗∗∗ 23.407∗∗∗ 36.906∗∗∗ (1.971) (2.454) (2.213) First bias-corrected 15.645∗∗∗ 7.880∗∗ 19.392∗∗∗ (1.971) (2.454) (2.213) First robust 15.645∗∗∗ 7.880 19.392∗∗∗ (3.785) (5.150) (4.567) N (Obs.) 2667 2678 2996 N right 1294 1413 1553 N left 1373 1265 1443 Source: Authors’ calculations based on Algeria MICS 2006, 2012, 2018. Note: ∗ p < 0.05; ∗∗ p < 0.01; ∗ ∗ ∗ p < 0.001. Models based on rdrobust. Standard errors in parentheses. Outcome is employ- ment (as a percentage). “First” denotes the first-stage discontinuity in pre-primary. N right denotes the number of observations to the right of the cut-off; N left denotes the number of observations to the left of the cut-off, and N (Obs.) is the total number of observations (sum of N left and N right). 5.4 Sensitivity Analyses There are a series of further sensitivity analyses for the estimates of the impact of pre-primary on women’s employment in the appendix, which are summarized here. In table S2.2 (age four) and table S2.3 (age five), analyses halve the bandwidth used, from six months to roughly three months (13 weeks). Re- sults are substantively similar to the main results, although, as expected, standard errors increase, and therefore statistical significance does not always persist. In table S2.4 (age four) and table S2.5 (age The World Bank Economic Review 89 five), analyses use a linear functional form (polynomial degree one). In table S2.6 (age four) and table S2.7 (age five), analyses use a quadratic functional form (polynomial degree two). Results for treat- ment effects are similar, but most of the significant results become insignificant as standard errors are large. In table S2.8 and table S2.9 analyses include covariates in the FRDD estimations. As expected, since covariates are nearly random relative to the cut-off, these results are quite similar to the main results. In table S2.10, analyses include in the age five sample children who are in primary school as Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 not in pre-primary rather than excluding them, and the results are generally similar, although some lose significance. An additional set of estimates use 2SLS directly to calculate the FRDD treatment effect. Table S2.11 (age four) and table S2.12 (age five) present the first stages and table S2.13 (age four) and table S2.14 (age five) present the second stage results. The intercept only and linear models are presented (these are the same point estimates as conventional FRDD estimates, albeit different standard errors), along with an intercept (for employment) and linear (for pre-primary enrollment) combination that is not estimable with rdrobust. The effect of pre-primary on age four is insignificant in all the models. The age-five results are, unsurprisingly, similar to the main results, with some differences in statistical significance. 5.5 Potential Mechanisms One reason that the expansion of pre-primary, and particularly public preparatory (age five), may have been ineffective in increasing women’s employment is that the hours of care were difficult to reconcile with employment. For instance, public preparatory provided only 21 hours of care per week starting in 2011 (Nawel 2011; UNESCO-IBE 2012). Figure 4 presents the distribution of hours of pre-primary for the Figure 4. Distribution of Pre-Primary Hours per Week by Mother’s Employment, Age Four Source: Authors’ calculations based on Algeria MICS 2006, 2012. Note: Pooling 2006 and 2012 data. Not available in 2018. Kernel density (Epanechnikov), bandwidth 3. Using children in the age-four group, within (+/-) six months of being age four on December 31 of that school year (the cut-off). Question was only asked of those ages three to four at the time of the survey. Dashed vertical line marks 21 hours of public preparatory. 90 Krafft and Lassassi age-four group (ages 3.5–4.5) by women’s employment.24 Women who were employed tended to have children with longer hours of pre-primary at age four, a median of 20 hours per week, compared to a median of 12 for women who were not employed. Moreover, a third (33 percent) of women who were employed had more than 21 hours of care. The following year, if their children attended public preparatory, these women might find it difficult to reconcile employment and preparatory’s shorter hours, leading to the drop in employment in 2012 and 2018, after the expansion of the preparatory grade (age Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 five). The preparatory expansion in 2012 and 2018 may have induced mothers who would have previously kept their children in other care arrangements if they worked to send their five-year-olds to preparatory instead. Table S2.15, in the supplementary online appendix, demonstrates for the age-five group that pre-primary enrollment not only increased over time but became less selected. This interpretation is corroborated by undertaking heterogeneity analysis, examining the results for nuclear households (the woman, spouse, and children only) versus extended households in table S2.16 and table S2.17, in the supplementary online appendix. Women in nuclear households will lack alternative caregivers for after the short pre-primary day, whereas women in extended households are likely to have other caregivers available. The first stage (discontinuity in pre-primary around the cut-off) is significant and similar in magnitude for nuclear and extended households in each year and for each age group. The FRDD estimates of the impact of pre-primary on employment show that the negative effects at age five for 2012 and 2018 are concentrated among nuclear households only. Short days of pre-primary without other caregiving assistance are difficult for women to reconcile with employment. In a similar vein, table S2.18 and table S2.19, in the supplementary online appendix, explore FRDD es- timates for being in wage employment (versus non-wage or non-employment) and non-wage employment (versus wage or non-employment). Women in MENA tend to leave wage work at, or in anticipation of, marriage, given the difficulties of reconciling such work with caregiving, but at marriage they maintain or increase their levels of non-wage work (e.g., self-employment), which is easier to reconcile with caregiving (Selwaness and Krafft 2021; Assaad, Krafft, and Selwaness 2022). The results demonstrate that decreases in employment at age five in 2012 and 2018 are primarily for wage employment, with much smaller and usually insignificant effects for non-wage work. This result is consistent with challenges in reconciling wage work, which typically has a more rigid schedule than non-wage work, with preparatory schedules. 6. Discussion and Conclusions Pre-primary education is frequently cited as an important intervention to raise women’s employment in LMICs generally and MENA in particular. Conceptually, increasing access to pre-primary would address a key constraint on women’s employment, namely women’s domestic responsibilities and opportunity cost of time. Empirically, evidence from other countries shows reducing costs or increasing access to child care generally has the potential to increase women’s employment (Berger and Black 1992; Crawford 2006; Berlinski and Galiani 2007; Blau and Tekin 2007; Baker, Gruber, and Milligan 2008; Martinez, Naudeau, and Pereira 2012; Davis et al. 2018; Clark et al. 2019; Dang, Masako, and Nguyen 2019). The literature to date from LMICs has likewise almost universally demonstrated positive effects of child care access on women’s labor-market outcomes (Halim, Perova, and Reynolds 2023). This previous literature has, however, come from contexts with comparatively high rates of women’s participation (Halim, Perova, and Reynolds 2023). This paper investigated whether pre-primary could increase employment in Algeria, a context with very low employment rates (14 percent) for women. The paper specifically tested the effect of a substantial pre- primary expansion in Algeria on women’s employment. Pre-primary enrollment rates for children aged 24 The question was only asked for those children ages three to four at the time of the survey. The question was not asked in 2018; figure pools 2006 and 2012 data. The World Bank Economic Review 91 five were 5 percent in 2005 but rose to 79 percent by 2011. The analyses identified the effect of pre- primary using FRDD, based on the December 31 age cut-off to be eligible for pre-primary. The research compared results from 2006 (early in the expansion), 2012, and 2018, and for ages four and five. Analyses demonstrated that covariates were close to as-good-as-random and there was no manipulation around the cut-off, to ensure the validity of the identification strategy. The analyses demonstrated a clear jump in pre-primary enrollment at the cut-off in recent years and particularly for age five. Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 What the paper found was that the pre-primary expansion did not increase women’s employment. If anything, it appears to have actually decreased employment for women with children aged five in 2012 and 2018. Age five was the target of the expansion, and the significant effects at age five (but not age four) support the identification strategy and findings. The significance of results is somewhat sensitive to the specification, particularly functional form (the study used the functional form most often recommended based on the AIC). Why would pre-primary have a negative effect on women’s employment in Algeria? The part-day (and lunch break) structure of public preparatory may make it harder for women to work. The length of pre-primary care for employed women with children in the age-four group is often longer than public preparatory (age five) hours. Moreover, negative effects are concentrated among women living in nuclear families, who lack alternative caregivers. Other types of care than public preparatory may be better able to meet women’s need for care during work hours. The fact that Algeria’s pre-primary expansion appears to have not increased, and likely instead have decreased, employment underlines the importance of policy design. In order for pre-primary to increase women’s employment, it may need to provide work-day length care. The importance of all-day school or afterschool to raising women’s employment rates has been demonstrated in other contexts (Cannon, Jacknowitz, and Painter 2006; Berthelon, Oyarzún, and Kruger 2015; Martínez A. and Perticará 2017; Dhuey, Lamontagne, and Zhang 2019). The present study is the first work to show that part-day care can reduce women’s employment, but is congruent with other evidence that part-day care may not be effective in raising women’s employment (Medrano 2009). Careful policy design may be particularly important in contexts where women’s baseline employment is low. A study of Israeli Arab mothers, whose baseline labor force participation was only 17 percent, found that the implementation of full-day pre-primary for ages 3–4 led to a 7 percentage point increase in women’s participation (Schlosser 2011). Ensuring that pre-primary expansions align with women’s work schedules will be a critical area for future investigation; if Algeria switched to all-day pre-primary, would this then increase women’s employment? Declaration of interest statement None. Data availability This study uses the 2006, 2012, and 2018 Multiple Indicator Cluster Survey data. The 2006 data were received in correspondence with the Ministry of Health Population and Hospital Reform; 2012 and 2018 data are publicly available from https://mics.unicef.org/surveys. Stata .do files for replication will be made available on the corresponding author’s website. References Algeria National Statistics Office. 2019. “Activity, Employment, and Unemployment in May 2019 (French).” No. 879.https://www.ons.dz/IMG/pdf/emploi_chom_mai_2019.pdf. 92 Krafft and Lassassi Assaad, R., R. Hendy, M. Lassassi, and S. Yassin. 2020. “Explaining the MENA Paradox: Rising Educational Attain- ment, Yet Stagnant Female Labor Force Participation.” Demographic Research 43(28): 817–50. Assaad, R., C. Krafft, and I. Selwaness. 2022. “The Impact of Marriage on Women’s Employment in the Middle East and North Africa.” Feminist Economics 28(2): 247–79. Attanasio, O., H. Low, and V. Sánchez-Marcos. 2008. “Explaining Changes in Female Labor Supply in a Life-Cycle Model.” American Economic Review 98(4): 1517–52. Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 Baker, M., J. Gruber, and K. Milligan. 2008. “Universal Child Care, Maternal Labor Supply, and Family Well-Being.” Journal of Political Economy 116(4): 709–45. Berger, M.C., and D.A. Black. 1992. “Child Care Subsidies, Quality of Care, and the Labor Supply of Low-Income, Single Mothers.” Review of Economics and Statistics 74(4): 635–42. Berlinski, S., and S. Galiani. 2007. “The Effect of a Large Expansion of Pre-Primary School Facilities on Preschool Attendance and Maternal Employment.” Labour Economics 14(3): 665–80. Berlinski, S., S. Galiani, and P.J. McEwan. 2011. “Preschool and Maternal Labor Market Outcomes: Evidence from a Regression Discontinuity Design.” Economic Development and Cultural Change 59(2): 313–44. Berthelon, M.E., M. Oyarzún, and D.I. Kruger. 2015. “The Effects of Longer School Days on Mothers’ Labor Force Participation.” IZA Discussion Paper No. 9212. Institute for the Study of Labor. Bonn, Germany. Blau, D., and E. Tekin. 2007. “The Determinants and Consequences of Child Care Subsidies for Single Mothers in the USA.” Journal of Population Economics 20: 719–41. Bouzoubaa, K., and N. Benghabrit-Remaoun. 2004. “Preschool Education in Morocco and in Algeria (French).” Per- spectives. Revue Trimestrielle d’éducation Comparée 34(4): 471–80. Calonico, S., M.D. Cattaneo, and R. Titiunik. 2014a. “Robust Nonparametric Confidence Intervals for Regression- Discontinuity Designs.” Econometrica 82(6): 2295–326. Calonico, S., M.D. Cattaneo, and R. Titiunik. 2014b. “Robust Data-Driven Inference in the Regression-Discontinuity Design.” Stata Journal 14(4): 909–46. Campante, F.R., and D. Chor. 2012. “Why Was the Arab World Poised for Revolution? Schooling, Economic Oppor- tunities, and the Arab Spring.” Journal of Economic Perspectives 26(2): 167–88. Cannon, J.S., A. Jacknowitz, and G. Painter. 2006. “Is Full Better than Half? Examining the Longitudinal Effects of Full-Day Kindergarten Attendance.” Journal of Policy Analysis and Management 25(2): 299–321. Cascio, E.U. 2009. “Maternal Labor Supply and the Introduction of Kindergartens into American Public Schools.” Journal of Human Resources 44(1): 140–70. Cattaneo, M.D., M. Jansson, and X. Ma. 2018. “Manipulation Testing Based on Density Discontinuity.” Stata Journal 18(1): 234–61. Chalal, A.N. 2018. “They Can Be Enrolled in the 1st Year of Primary School at Age 5: Exemption for Children Born from January to March 31, 2013.” Courrier D’Algérie. Clark, S., C.W. Kabiru, S. Laszlo, and S. Muthuri. 2019. “The Impact of Childcare on Poor Urban Women’s Economic Empowerment in Africa.” Demography 56: 1247–72. Connelly, R. 1992. “The Effect of Child Care Costs on Married Women’s Labor Force Participation.” Review of Economics and Statistics 74(1): 83–90. Crawford, A. 2006. “The Impact of Child Care Subsidies on Single Mothers’ Work Effort.” Review of Policy Research 23(3): 699–711. Dang, H-A.H., H. Masako, and C.V. Nguyen. 2019. “Childcare and Maternal Employment: Evidence from Vietnam.” GLO Discussion Paper No. 349. Global Labor Organization. Essen, Germany. Davis, E.E., C. Carlin, C. Krafft, and N.D. Forry. 2018. “Do Child Care Subsidies Increase Employment Among Low-Income Parents?” Journal of Family and Economic Issues 39(4): 662–82. Dhuey, E., J. Lamontagne, and T. Zhang. 2019. “The Impact of Full-Day Kindergarten on Maternal Labour Supply.” IZA Discussion Paper Series No. 12507. Institute of Labor Economics. Bonn, Germany. Fitzpatrick, M.D. 2010. “Preschoolers Enrolled and Mothers at Work? The Effects of Universal Prekindergarten.” Journal of Labor Economics 28(1): 51–85. Fitzpatrick, M.D. 2012. “Revising Our Thinking About the Relationship Between Maternal Labor Supply and Preschool.” Journal of Human Resources 47(3): 583–612. Gathmann, C., and B. Sass. 2018. “Taxing Childcare: Effects on Childcare Choices, Family Labor Supply, and Chil- dren.” Journal of Labor Economics 36(3): 665–709. The World Bank Economic Review 93 Gelbach, J. 2002. “Public Schooling for Young Children and Maternal Labor Supply.” American Economic Review 92(1): 307–22. Gelman, A., and G. Imbens. 2019. “Why High-Order Polynomials Should Not Be Used in Regression Discontinuity Designs.” Journal of Business & Economic Statistics 37(3): 447–56. H., K. 2021. “Algeria: Start Date of Enrollment in Primary and Preschool.” Dzair Daily. March 21. Halim, D., E. Perova, and S. Reynolds. 2023. “Childcare and Mothers’ Labor Market Outcomes in Lower- and Middle- Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 Income Countries.” World Bank Research Observer 38(1):73–114 Havnes, T., and M. Mogstad. 2011. “Money for Nothing? Universal Child Care and Maternal Employment.” Journal of Public Economics 95(11–12): 1455–65. ILO. 2022. World Employment and Social Outlook: Trends 2022. Geneva: International Labour Office. Imbens, G.W., and T. Lemieux. 2008. “Regression Discontinuity Designs: A Guide to Practice.” Journal of Economet- rics 142(2): 615–35. Keo, C., C. Krafft, and L. Fedi. 2022. “Rural Women in Egypt: Opportunities and Vulnerabilities.” In The Egyptian Labor Market: A focus on Gender and Vulnerability, edited by Krafft, C, and R Assaad, 225–56. Oxford: Oxford University Press. Lee, D.S., and T. Lemieux. 2010. “Regression Discontinuity Designs in Economics.” Journal of Economic Literature 48: 281–355. Mahdjoub, R. 2017. “Expanding Preschool Coverage: Reflections on the Structural Elements of a New Policy.” In- saniyat: Algerian Journal of Anthropology and Social Sciences 75–76: 37–66. Martínez A., C., and M. Perticará. 2017. “Childcare Effects on Maternal Employment: Evidence from Chile.” Journal of Development Economics 126: 127–37. Martinez, S., S. Naudeau, and V. Pereira. 2012. “The Promise of Preschool in Africa: A Randomized Impact Evaluation of Early Childhood Development in Rural Mozambique.” New Delhi, India: International Initiative for Impact Evaluation. Medrano, P. 2009. “Public Day Care and Female Labor Force Participation: Evidence from Chile.” Serie Documentos de Trabajo (SDT) Departamento de Economica Universidad de Chile No. 306. Santiago, Chile. Ministry of Health and Population Republic of Algeria. 1996. “National Survey of the Objectives of the Mid- Decade ‘MDG Algeria 1995’ (French).”https://mics-surveys-prod.s3.amazonaws.com/MICS1/Middle%20East%2 0and%20North%20Africa/Algeria/1995/Final/Algeria%201995%20MICS_French.pdf. Ministry of Health and Population Republic of Algeria - National Institute of Public Health. 2001. “Na- tional Survey on the Mother and Child Health Objectives of the End of the Decade Algeria 2000 MICS2 (French).” https://mics-surveys-prod.s3.amazonaws.com/MICS2/Middle%20East%20and%20North%20Africa/ Algeria/2000/Final/Algeria%202000%20MICS_French.zip. Ministry of Health Population and Hospital Reform. 2015. “Republic of Algeria Multiple Indicator Cluster Survey Principal Report (French).” https://mics-surveys-prod.s3.amazonaws.com/MICS4/Middle%20East%20and%20N orth%20Africa/Algeria/2012-2013/Final/Algeria%202012-13%20MICS_French.pdf 2008. Ministry of Health Population and Hospital Reform, and National Office of Statistics. 2008. “Republic of Algeria Multiple Indicator Cluster Survey Principal Report (French).” https://mics-surveys-prod.s3.amazonaws.com/MIC S3/Middle%20East%20and%20North%20Africa/Algeria/2006/Final/Algeria%202006%20MICS_French.pdf. Ministry of Health Population and Hospital Reform, and United Nations Children’s Fund (UNICEF). 2020. “Algeria - Multiple Indicator Cluster Survey (MICS) 2018-2019.” https://mics-surveys-prod.s3.amazonaws.com/MICS6/M iddle%20East%20and%20North%20Africa/Algeria/2018-2019/Survey%20findings/Algeria%202018-19%20 MICS_French.pdf. Nawel, D. 2011. “Back to School: The Continuous Schedule Is Not Fully Applied (French).” Algérie 360, September 12. Schlosser, A. 2011. “Public Preschool and the Labor Supply of Arab Mothers: Evidence from a Natural Experiment.” Tel Aviv University. Tel Aviv, Israel. Secretary General of the Republic of Algeria. 2008. “January 23 2008 Law of Orientation on National Education (French).” Official Journal of the Republic of Algeria, no. 4. Selwaness, I., and C. Krafft. 2021. “The Dynamics of Family Formation and Women’s Work: What Facilitates and Hinders Female Employment in the Middle East and North Africa?” Population Research and Policy Review 40(3): 533–87. 94 Krafft and Lassassi Spierings, N., J. Smits, and M. Verloo. 2010. “Micro- and Macrolevel Determinants of Women’s Employment in Six Arab Countries.” Journal of Marriage and Family 72(5): 1391–407. UNESCO-IBE. 2012. “World Data on Education VII Ed. 2010/11 Algérie (French).” UNESCO. 2021. “Right to Pre-Primary Education: A Global Study.” UNICEF Algeria. 2014. “All in School: Middle East and North Africa Out-of-School Children Initiative: Summary: Algeria: Country Report on Out-of-School Children.” Downloaded from https://academic.oup.com/wber/article/38/1/74/7242192 by Joint Bank/Fund Library user on 02 February 2024 World Bank. 2019. “World Bank EdStats.” World Bank Databank. Retrieved November 24, 2019. www.databank.w orldbank.org.