Policy Research Working Paper 10305 Maternal Work and Children’s Development Examining 20 Years of Evidence Maria C. Lo Bue Elizaveta Perova Sarah Reynolds East Asia and the Pacific Region Office of the Chief Economist February 2023 Policy Research Working Paper 10305 Abstract Maternal work may affect children positively through for time of equal quality by other caregivers, children’s increased household income, higher control of mothers over development may be penalized. Stress associated with work available income, and expansion of maternal information may also decrease the quality of parenting. This review networks through work contacts and greater decision-mak- summarizes causal evidence on the relationship between ing power of mothers as they become more economically maternal work and children’s development. The majority empowered. However, maternal work may reduce maternal causal studies find positive or null impacts of maternal work time spent with children. If maternal time is not substituted on children’s development. This paper is a product of the Office of the Chief Economist, East Asia and the Pacific Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at eperova@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Maternal Work and Children’s Development: Examining 20 Years of Evidence Maria C. Lo Bue, Elizaveta Perova and Sarah Reynolds J13, J21, J16, E24 Keywords: child development, maternal labor force participation, gender equality, review of evidence Complex links between maternal work and children’s development Maternal work is linked to children’s development through several channels. On the one hand, maternal work may affect children’s development positively. First, children benefit from living in households with higher income, and mothers—in comparison to fathers—spend a higher proportion of their resources on children (Lundberg 1997; Ermish and Francesconi 2013; Dunbar, Dunbar, Lewbel, and Pendakur, 2013). There are also intangible benefits of maternal work, such as mothers’ exposure to a larger social network through work contacts, greater access to information, and a potential increase in mothers’ decision- making power, all of which may allow for better choices around monetary and non-monetary investments in children (Currie, 2009; Doss, 2013; Lépine and Strobl, 2013). On the other hand, maternal work may affect children’s development negatively through reduction in maternal time spent with children and, consequently, fewer opportunities for interactions and consistent engagement. Child development would be penalized by maternal work if institutional or other non- maternal childcare is of inferior quality than maternal care and the increase in income does not compensate with nutritional and safety expenses, for example. Stress associated with work may reduce the quality of parenting (Baker, Gruber, Milligan, 2008). The combination of these factors may lead to less secure attachment with the mother - an important factor for children’s socio-emotional development and subsequent academic success (Alto and Petrenko, 2017; Allen, 2008). Maternal work may also increase the likelihood that older children take on more domestic responsibilities and spend less time on homework or enrichment activities (Afridi et al., 2016). With both positive and negative theoretical impacts of mothers’ work on child development, it is important to examine whether up-to-date empirical evidence suggests overall positive, negative, or null impacts. Our study summarizes the evidence of maternal work on child and adolescent health and intellectual development from the past 20 years. We note strengths and limitations of the studies and highlight common threads relevant to different facets of policy design, including considerations for emerging economies. Broad coverage of contexts and measures Using a variety of search terms, we retreived 1,181 studies from databases that cover medical, psychological, economic and other social science literature. We identified a total of 613 relevantly-titled articles published between 2002 and 2021, and further limited them to 312 which examined the links between maternal work and development of children ages 0 to 18 years. We further narrowed our selection to 80 papers, which utilized statistical methods to plausibly establish causal impact. Within these studies, we limited our analysis to a set of 26 articles, which used the most rigourous methods: instrumental variables, sibling fixed effects, and multiple analytic approaches (most commonly individual fixed effects and lagged dependent variables combined). Twenty-six causal papers cover a range of contexts: 21 papers are from Western countries (Australia, Denmark, Italy, Norway, the United Kingdom, and the United States) and 5 are from non-Western: three lower-middle income (the Arab Republic of Egypt, India, and Indonesia) and one high income (Chile). Among papers on high-income countries (all Western countries and Chile), 5 focus on low-income populations (such as mothers of welfare in the United States), and 22 rely on nationally representative surveys or censuses. 2 Overall, the 26 causal studies examined 20 different measures of child development. To reduce multiple dimensions of this body of evidence, we combined them into 6 main outcomes: schooling, cognitive, behavioral and 3 types of health measures. Schooling outcomes included years of schooling, current enrollment, grade attaintment and high school grades. Cognitive outcomes included test scores from a range of psycometrically validated exams. Behavioral outcomes are largely comprised of self-reported risky beahviors for adolescents and parent-reported standardized indexes for behavioral problems for younger children. We grouped health outcomes into three categories: (1) height outcomes, which capture long-term nutrition, (2) weight outcomes, which consider short-term nutrition and physical activity,1 and (3) other health measures, most of which were indicators of medical concerns such as lung capacity, hemoglobin levels, asthma episodes, ear infections and overnight hospitalization. Several studies examined more than one group of outcomes. To facilitate the discussion of results, we refer to separate groups of outcomes within a study as a substudy. A study can have one substudy or multilple substudies. Twenty-six causal studies yielded 40 substudies with evidence on 6 outcomes (height, weight, health, schooling, behavior and cognition). For each substudy, we categorized whether it detected null, positive or negative impacts of maternal work on child development for each group of outcomes it examined. If at least one outcome in a group is negative/positive in a statistically significant way, while impacts on other outcomes in that group are not statistically significant, we count it as evidence of negative/positive impact for the substudy.2 If there was no statistically significant impact on any outcomes in the group, we count it as evidence of null effect. There were no papers which detected a statistically significant positive impact on one outcome in a group, and negative on a different one. Main findings: Impacts on children’s development The most abundant evidence is on weight-related outcomes, followed by cognitive and behavioral outcomes (12, 8 and 8 substudies, respectively). The evidence is more scant on schooling, health and height outcomes, with 5, 4 and 3 substudies, respectively. The majority of causal evidence from 10 countries over the last 20 years suggests that maternal work does not hurt children’s development. Of the substudies, 68% (27 out of 40) do not find evidence of negative impact, with 6 finding positive, and 21 null effects (Figure 1). An interesting feature of the studies reviewed is that many of them use a wide range of outcomes, and only 2 corrected for multiple hypotheses testing, or adjusted for the probability of discovering a non-null effect purely by chance. While common practice in medicine, correction for multiple hypotheses testing started making forays into social science only recently.3 Thus the 20-year span of our analysis incorporates many papers that were not yet submitted to this additional check. 1 Weight outcomes include deviations from the norm in both directions: underweight as well as obesity. 2 For example, the study in Indonesia examines 2 indicators for health: hemoglogin and lung capaicty. The authors find statistically signficant impacts on hemoglobin, but not on lung capacity. We categorize this result as evidence of positive impact on health. 3 Some of the earliest papers that raised the question of correcting for multiple hypotheses testing and demonstrating sensitivity of results include Anderson (2008), List, Shaikh and Xu (2019), Kling, Liebman and Katz (2007). 3 Using the values of coefficients and standard errors in the papers, we have carried out two types of checks for multiple hypotheses testing, using both Bonferroni correction and the less conservative Benjamini- Hochberg (1995) correction for False Discovery Rate. Application of these corrections reduces the number of substudies with negative results to 9,4 with 78% of the substudies yielding non-negative results (Figure 2). Understanding when maternal work may have a negative impact If one of the objectives of the policy maker is to do no harm, it is important to focus on the negative sub- studies and understand whether there are specific features of maternal work that may have negative impact on children. A closer examination of papers with negative outcomes reveals several trends. First, the environment in which children of working mothers are while mothers are working matters. Both Rashad and Sharaf (2019) and Shajan and Subbyamoola (2020) find that maternal work increases the likelihood that children are stunted in Egypt and India, respectively. These two countries have low presence of early childhood education facilities, compared to other countries analyzed in the papers reviewed. Thus, the children of working mothers are likely to be with them in the field or in the market, rather than in center-based care. Indeed, Shajan and Subbyamoola (2020) point out that their results are driven by children of women employed in agricultural and manual, rather than professional jobs. Gennetian et al. (2010) examine the impacts of maternal work for low-income populations in the United States, focusing on women eligible for the public welfare program. The earnings from the public welfare program may not be sufficient to cover the costs of high-quality childcare. Second, specific aspects of maternal work matter. Felfe and Hsin (2012) examine the impacts of characteristics of maternal work, comparing working mothers exposed to more or less hazardous environments. Thus, their study suggests the negative impact of work-related stress and hazards, rather than maternal work per se. James-Burdumy (2005) finds that maternal work results in negative cognitive impacts at age 9 only during the first year of life. However, work during the second and third years of life does not have this effect, suggesting that children's vulnerability or sensitivity to mothers' working is in very early childhood. Additionally, the cognitive impact is not all-around: only math scores were impacted, while reading and vocabulary scores had positive, albeit not significant association with maternal work. Third, in developed countries, when viable childcare options are present, the impacts are generally low. James-Burdumy (2005) notes that the size of the "loss" to mathematics skill is relatively small at a 0.03 effect size. Furre Haaland, Rege and Votruba (2013) find that 5 additional years of full-time employment reduce a child’s education by 0.065 year (or 4 percent of the standard deviation in their sample). Lastly, one study pertains to a relatively narrow context, which allows for a potentially different interpretation of outcomes. Morrill (2011) focuses on the impact on hospitalizations in the United States. Notably, her analysis does not control for whether the mother has insurance. In the United States, the availability of insurance may be an important factor in the decision whether to take the child to the doctor or not, conditional on the same level of severity of incident. Mothers who work are more likely to have insurance. Thus, the paper may be capturing a different treatment for the same level of severity of 4 The sub-studies with non-null results which did not withstand correction for multiple hypotheses testing found negative behavioral impact (Felfe and Hsin, 2012 and Aughinbaugh and Gittleman, 2003); negative impact on being underweight (Rashad and Sharaf, 2012), and increase in the likelihood of obesity (Von Hinke Kessler Scholder, 2008). 4 symptoms, and plausibly, higher likelihood to seek medical attention may be considered a positive outcome. Discussion Of the 90 countries where World Values Surveys were administered between 2017 and 2022, in about half of the countries half of the respondents either agree or strongly agree that a preschool child suffers with a working mother. The rate of agreement ranges from 9% in Denmark to 88% in Bangladesh. In three- quarters of the countries, at least 30% of respondents agree that maternal work before primary school is detrimental for the child (Haerpfer et al., 2022). However, the rigorous research over the past 20 years does not support this widely shared opinion. Most of the papers find non-negative impact of maternal work on children. Several studies that do find negative impacts suggest that specific aspects of maternal work, such as stress or timing, or circumstances in which women work (such as lack of availalbity of high-quality childcare) may be driving negative impacts, rather than maternal work per se. Fortunately, these parameters can be addressed by policy: through improving working conditions, supporting provision of childcare and offering parental leave. Maternal work is important for economic growth and welfare. Women’s entry into the labor force has been a cornerstone of economic growth during the last 50 years in the United States (Hsieh et al., 2019). Currently, a number of governments (including the United States and France) are increasing their budgetary allocations to childcare services in order to support maternal work. Our review does not point to inevitable hidden costs of maternal work in the form of human capital losses of children; rather, it points to the necesity to shape the conditions in which women work through parental leave and childcare policies. Such policy measures are likely to ensure that the economies can reap the short-term economic benefits of maternal work without long-term losses in human capital. Figure 1 5 Figure 2 6 References Afridi, F., Mukhopadhyay, A. & Sahoo, S. Female labor force participation and child education in India: evidence from the National Rural Employment Guarantee Scheme. IZA J Labor Develop 5, 7 (2016). https://doi.org/10.1186/s40175-016-0053-y Allen, Joseph P. 2008. “The Attachment System in Adolescence.” In Handbook of Attachment: Theory, Research, and Clinical Applications, 2nd Ed, 419–35. New York, NY, US: The Guilford Press. Alto, Michelle E., and Christie L. M. Petrenko. 2017. “Fostering Secure Attachment in Low- and Middle- Income Countries: Suggestions for Evidence-Based Interventions.” Evaluation and Program Planning 60 (February): 151–65. https://doi.org/10.1016/j.evalprogplan.2016.11.003. Baker, Michael, Jonathan Gruber, and Kevin Milligan. “Universal Child Care, Maternal Labor Supply, and Family Well‐Being.” Journal of Political Economy 116, no. 4 (August 2008): 709–45. https://doi.org/10.1086/591908. Currie, J. (2009). ‘Healthy, Wealthy, and Wise: Socioeconomic Status, Poor Health in Childhood, and Human Capital Development’. Journal of Economic Literature, 47(1): 87–122. https://doi.org/10.1257/jel.47.1.87 Cheryl Doss, Intrahousehold Bargaining and Resource Allocation in Developing Countries, The World Bank Research Observer, Volume 28, Issue 1, February 2013, Pages 52–78, https://doi.org/10.1093/wbro/lkt001 Duflo, Esther. 2003. “Grandmothers and Granddaughters: Old Age Pension and Intra-Household Allocation in South Africa.” World Bank Economic Review 17 (1): 1–25. Dunbar, Geoffrey R, Arthur Lewbel, and Krishna Pendakur. “Children’s Resources in Collective Households: Identification, Estimation, and an Application to Child Poverty in Malawi.” American Economic Review 103, no. 1 (February 1, 2013): 438–71. https://doi.org/10.1257/aer.103.1.438. Haerpfer, C., Inglehart, R., Moreno, A., Welzel, C., Kizilova, K., Diez-Medrano J., M. Lagos, P. Norris, E. Ponarin & B. Puranen (eds.). 2022. World Values Survey: Round Seven - Country-Pooled Datafile Version 5.0. Madrid, Spain & Vienna, Austria: JD Systems Institute & WVSA Secretariat. doi:10.14281/18241.20 Kling, J., Liebman, J., and Katz, L. (2007). Experimental analysis of neighborhood effects. Econometrica, 75(1):83–119. Lépine, A., and E. Strobl (2013). ‘The Effect of Women’s Bargaining Power on Child Nutrition in Rural Senegal’. World Development, 45: 17–30. https://doi.org/10.1016/j.worlddev.2012.12.018 John A. List & Azeem M. Shaikh & Yang Xu, 2019. "Multiple hypothesis testing in experimental economics," Experimental Economics, Springer;Economic Science Association, vol. 22(4), pages 773-793, December. Lundberg, S. J, R. A Pollak, and T. J Wales. 1997. “Do Husbands and Wives Pool Their Resources? Evidence from the United Kingdom Child Benefit.” Journal of Human Resources 32 (3): 463–80. 7 Reviewed Papers Afridi, F., Mukhopadhyay, A., & Sahoo, S. (2016). Female labor force participation and child education in India: Evidence from the National Rural Employment Guarantee Scheme. IZA Journal of Labor & Development, 5(1), 7. https://doi.org/10.1186/s40175-016-0053-y Agiro, B. T., & Huang, W.-C. (2020). Re-Examining the Effect of Maternal Employment on Child Overweight: The Case of School-Age Children. Journal of Family and Economic Issues, 41(1), 140–157. https://doi.org/10.1007/s10834-019-09658-8 Anderson, P. M., Butcher, K. F., & Levine, P. B. (2003). Maternal employment and overweight children. Journal of Health Economics, 22(3), 477–504. https://doi.org/10.1016/S0167-6296(03)00022-5 Aughinbaugh, A., & Gittleman, M. (2004). Maternal employment and adolescent risky behavior. Journal of Health Economics, 23(4), 815–838. https://doi.org/10.1016/j.jhealeco.2003.11.005 Baum, C. L. (2015). Does Early Maternal Employment Harm Child Development? An Analysis of the Potential Benefits of Leave Taking. Journal of Labor Economics. https://doi.org/10.1086/345563 Bishop, J. (2011). The Effect of Maternal Employment on Youth Overweight in Australia. Economic Record, 87(s1), 92–104. https://doi.org/10.1111/j.1475-4932.2011.00747.x Del Boca, D., Pasqua, S., & Suardi, S. (2016). Child Care, Maternal Employment, and Children’s School Outcomes. An Analysis of Italian Data. European Journal of Population, 32(2), 211–229. https://doi.org/10.1007/s10680-015-9370-0 Dervisevic, E., Perova, E., & Lo Bue, M. C. (2021). Maternal employment and children’s outcomes: Evidence from Indonesia. The World Bank. Dunifon, R., Kalil, A., Crosby, D., & Su, J. H. (2013). Mothers’ Night Work and Children’s Behavior Problems. Developmental Psychology, 49(10), 1874–1885. https://doi.org/10.1037/a0031241 Ermisch, J., & Francesconi, M. (2013). The Effect of Parental Employment on Child Schooling. Journal of Applied Econometrics, 28(5), 796–822. https://doi.org/10.1002/jae.2260 Felfe, C., & Hsin, A. (2012). Maternal work conditions and child development. Economics of Education Review, 31(6), 1037–1057. https://doi.org/10.1016/j.econedurev.2012.06.008 Gennetian, L. A., Hill, H. D., London, A. S., & Lopoo, L. M. (2010). Maternal employment and the health of low-income young children. Journal of Health Economics, 29(3), 353–363. https://doi.org/10.1016/j.jhealeco.2010.02.007 Greve, J. (2011). New results on the effect of maternal work hours on children’s overweight status: Does the quality of child care matter? Labour Economics, 18(5), 579–590. https://doi.org/10.1016/j.labeco.2011.03.003 8 Haaland, V. F., Rege, M., & Votruba, M. (2013). Nobody Home: The Effect of Maternal Labor Force Participation on Long-Term Child Outcomes (SSRN Scholarly Paper ID 2363912). Social Science Research Network. https://doi.org/10.2139/ssrn.2363912 Hubbard, M. N. (2009). Effects of maternal employment and child care on the health of young children [Ph.D., The University of North Carolina at Chapel Hill]. https://www.proquest.com/docview/304960202/abstract/4C8AF9FF2D534249PQ/1 James‐Burdumy, S. (2005). The Effect of Maternal Labor Force Participation on Child Development. Journal of Labor Economics, 23(1), 177–211. https://doi.org/10.1086/425437 Mendolia, S. (2016). Maternal Working Hours and the Well-Being of Adolescent Children: Evidence from British Data. Journal of Family and Economic Issues, 37(4), 566–580. https://doi.org/10.1007/s10834- 015-9480-1 Morrill, M. S. (2011). The effects of maternal employment on the health of school-age children. Journal of Health Economics, 30(2), 240–257. https://doi.org/10.1016/j.jhealeco.2011.01.001 Pekkurnaz, D. (2014). Dynamic analysis of maternal employment, child care, quality and early childhood development [Ph.D., The University of North Carolina at Chapel Hill]. https://www.proquest.com/docview/1613182648/abstract/D6F7129E26342F5PQ/1 Pilkauskas, N. V., Brooks-Gunn, J., & Waldfogel, J. (2018). Maternal Employment Stability in Early Childhood: Links with Child Behavior and Cognitive Skills. Developmental Psychology, 54(3), 410–427. https://doi.org/10.1037/dev0000438 Rashad, A. S., & Sharaf, M. F. (2019). Does maternal employment affect child nutrition status? New evidence from Egypt. Oxford Development Studies, 47(1), 48–62. https://doi.org/10.1080/13600818.2018.1497589 Reynolds, S. A., Fernald, L. C. H., & Behrman, J. R. (2017). Mothers’ labor market choices and child development outcomes in Chile. Social Science and Medicine - Population Health, 3, 756–766. https://doi.org/10.1016/j.ssmph.2017.08.010 Shajan, J., & Sumalatha, B. S. (2020). Maternal employment and children’s health in India: An exploratory analysis. Journal of Public Affairs, n/a(n/a), e2580. https://doi.org/10.1002/pa.2580 von Hinke Kessler Scholder, S. (2008). Maternal employment and overweight children: Does timing matter? Health Economics, 17(8), 889–906. https://doi.org/10.1002/hec.1357 Waldfogel, J., Han, W.-J., & Brooks-Gunn, J. (2002). The effects of early maternal employment on child cognitive development. Demography (Pre-2011), 39(2), 369–392. Ziol-Guest, K. M., Dunifon, R. E., & Kalil, A. (2013). Parental employment and children’s body weight: Mothers, others, and mechanisms. Social Science & Medicine, 95, 52–59. https://doi.org/10.1016/j.socscimed.2012.09.004 9 Supplementary Materials for Maternal Work and Children’s Development: Examining 20 Years of Evidence By Maria C. Lo Bue, Elisaveta Perova and Sarah Reynolds 1. Methods 1.2 Search To identify previous studies, we first listed search terms determined from an informal scoping review: we considered the main types of outcomes researched and used the authors’ knowledge of common realms of outcomes in the child development. (All authors had published on this topic previously.) The following three sets of search terms emerged: “Maternal” OR “Mother”; “Work” OR “employment” OR “labor”; and “child development” OR “health” OR “nutrition” OR “obesity” OR “stunting” OR “wasting” OR “underweight” OR “overweight” OR “behavior” OR “socio-emotional” OR “cognition” OR “schooling” OR “education” OR “illness” OR “asthma accident”. In October 2021 a research assistant used the 104 search term permutations to collect citations from EconLit and Web of Science.5 From this list, we selected the top three permutations with the most results and searched them in PubMed. This search did not provide additional documents to include in the review, so we did not continue with the full search in PubMed. We did find some additional citations using this approach with PsychInfo, so we also repeated the citation search using this database. We compared the citations from our search to the citations included in the systematic review by Lucas-Thompson, Goldberg, and Prause (2010). As a result, we added the additional search terms “outcomes” OR “achievement” OR “adolescent” OR “adolescence,” which increased our total to 1182 records. Duplicates were removed, reducing our total to 614 records. Titles were screened for topic relevance by a research assistant and one author of this review. We excluded studies focusing on samples of children with disabilities from birth, premature children, or teen mothers. We excluded outcomes on usage of health services (i.e., vaccinations, nutrition related behaviors, doctor visits). Similarly, we excluded outcomes that are considered mediators leading to the final outcomes in our list (e.g., dietary intake, sedentary behaviors, sleep, work-family conflict). We excluded employment relating to migration or sex work. This screening provided us with 313 records. For the final decision regarding inclusion, the abstracts and/or methods sections of papers were examined by an author of this review, with the other authors consulted regarding the empirical quality of the studies. We limited the studies to those published between 2002 and 2021. Studies were included if they had quantitative analysis, a measure of maternal work as a causal variable in the paper, at least one child or adolescent development outcome (cognition, schooling, behavior, anthropometrics, and health). Moreover, we restrict the analysis to studies analyzing natural or quasi-natural experiments and using the following methodologies: individual fixed or random effects, lagged dependent variables,6 sibling fixed 5 The search was restricted to the titles of articles, since when the entire content of the articles was searched, it returned tens of thousands of papers. Several keyword searches (in contrast to title searches) of the permutations that yielded the most results did not provide additional papers for the review, so we did not pursue a keyword search. 6 Even though many papers analyzed longitudinal data on maternal work, we excluded studies that only controlled for observables. One exception is the lagged dependent variable approach, which controls for the baseline value of the outcome variable. Although the lagged outcome variable is an observable, including it accounts for unobserved maternal or child characteristics that affected the earlier outcome, which reduces selection bias due to unmeasured child and family characteristics. However, if the study had controls for child development measures at baseline and these were from the same realm as the outcome variable, even if not exactly the same measure, we 10 effects, instrumental variables, difference in differences, or randomized control trials. We excluded therefore studies based on cross-sectional data (129),7 non-analytical studies, such as review studies (24), macro studies (1), studies with small sample size (2), non-rigorous studies or with problematic data (2), studies not focusing on the selected child development outcomes (9), studies with the wrong intervention (21). Further studies were excluded because they were not written in English (2), not accessible via internet (25), or duplicates (17). This screening provided us with 81 studies. We further limited our review to 26 studies using the most rigorous methods to establish causal impacts: those using instrumental variables, sibling fixed effects, and studies using multiple analytic approaches (most commonly individual fixed effects and lagged dependent variables). We did not find studies with difference in differences, randomized control trials, or studies that combined random effects with another method. Only one study relies on experimental variation generated by randomly assigned welfare-to-work program (Gennetian et al., 2005), however, it uses instrumental variables strategy due to imperfect compliance with program assignment. Three identification strategies used in the remaining causal studies include instrumental variables, fixed effects and lagged dependent variables (or some combination of these techniques). Figure 1 Prisma Diagram 1.3 Scope of the analytical approaches used in the selected studies Instrumental variables The instrumental variables approach allows for identification of causal impacts of an endogenous independent variable (maternal work) on a dependent variable (children’s development) when researchers can identify a variable which is strongly correlated with the former but not the latter (an instrument). The validity of instrumental variables relies on several assumptions, some of which are testable while others are not. First, there should be a strong correlation between the variables used as did include the study in the review. For example, studies with the outcome BMI that controlled for birthweight (though not the dichotomous variable low birthweight) were included; studies on cognition that controlled for birthweight were not included if they did not also have an early measure of cognition. 7 We excluded qualitative work and studies with cross-sectional or correlational analysis, including propensity score matching. Propensity score matching only uses observables to create comparisons and does not control for unobserved heterogeneity. 11 instrument and maternal work: as this assumption is testable, we only include papers with strong first stage in the review.8 The exclusion restriction – or assumption that the instrument only affects children’s development through its impact on maternal employment only – is impossible to test. The range of papers that we use exploits a variety of instruments, from local unemployment rates to eligibility of the youngest child for kindergarten to changes in tariffs for female-dominated industries to child’s health at birth. Exclusion restriction is more likely to be satisfied in some cases than in other. For example, children’s birth weight is likely to affect children’s development outcomes directly, in addition to maternal employment decisions. Several papers construct instrumental variables based on economic variables at the location of residence, such as local unemployment rates, female labor force participation, per capita governmental transfers. Variables reflecting local economic conditions may also affect children’s development directly–for example, through higher quality educational establishments and medical services. Although we note that plausibility that exclusion restriction holds varies in papers we reviewed, we did not exclude any based on these criteria, as it is not testable, and exclusion would rely on our subjective judgement. Fixed effects Fixed effects strategies identify causal impacts through controlling for unobserved but fixed omitted variables. This approach relies on two core identifying assumptions. The first assumption is that conditional on a set of fixed time-invariant characteristics, maternal decision to work is as good as randomly assigned. The second assumption is that the effect of maternal work is additive and constant. There are two variations of this identification strategy in the papers which we reviewed: a subset of papers uses maternal (or family) fixed effects, another set uses child fixed effects. In the former case, identification relies on comparing development outcomes of at least two children born to the same mother, when her work status varies by child. To identify causal impacts, we need to assume that unobservable differences between children do not affect maternal decision to work. In the latter case, researchers estimate impacts from comparison of outcomes of the same child overtime, with mother working in some periods and not working in other. Identification of causal impacts requires assuming that maternal decision to work does not respond to time-variant changes in child’s development. Assumptions in both cases can be easily violated: for example, mothers may choose to stay at home when their less academically able children are at school to help them with homework. Similarly, in the case of child fixed effects, mothers may adjust their work in response to children’s development: for example, they may withdraw from the labor market in critical years, or in response to illness. Of course, the likelihood that the identifying assumptions are violated depends on the set of controls included in the estimating regression: for instance, it is more plausible that maternal decision to work in maternal fixed effects framework is as good as randomly assigned when conditional on her children’s ability. Lastly, fixed effects estimation is susceptible to attenuation bias from measurement error. Instrumenting independent endogenous variable can reduce the measurement error, and indeed several papers in our review rely on a combination of fixed effects and instrumental variables. Several papers use random effects approach, which also relies on identifying causal impacts by purging the influence of time-invariant variables, but, unlike fixed effects, allows residuals for a given child to be correlated across periods. 8 All selected papers either presented the first stage or stated that it was strong. 12 Lagged dependent variables Lagged dependent variables also exploit variation in outcomes of the same child observed over time. However, identifying assumption is different: to establish causal relationship between maternal work and child’s development in the lagged dependent variables framework, one needs to assume that maternal decision to work is as good as random, conditional on child’s past development outcomes (not unobserved time-invariant characteristics). This assumption is likely to be violated if maternal decisions to work also respond to unobserved environmental variables, which may also affect children’s outcomes such as employment of other household members. Notably, fixed effects and lagged dependent variables can be thought of as bounding the causal effect of interest; and several studies included in this review include both FE and LDV. Consequently, we included studies that rely on lagged dependent variables when combined with child FE. 2. Samples Our review covers 10 countries: 7 high income (Australia, Chile, Denmark, Italy, Norway, UK and US), and 3 lower-middle income (the Arab Republic of Egypt, India and Indonesia). Only four countries are non- Western: Chile, Egypt, India and Indonesia. Table A1 presents the distribution of the number of papers by country and method. Table A1