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 ?> Number of Number of papers by method Country studies IV Child FE & LDV Mother FE Mixed methods Australia 1 1 Chile 1 1 Denmark 1 1 Egypt, Arab Rep. 1 1 India 2 1 1 Indonesia 1 1 Italy 1 1 Norway 1 1 United Kingdom 3 1 1 1 United States 14 4 4 3 3 Total 26 9 5 5 7 The papers rely on a wide range of data, including survey and administrative data. A significant share of surveys are nationally representative, some focus on low-income populations. Some studies focus on a specific subsample – due to methodological demands (e.g., children with younger siblings only). Table AX <table 1?> presents the details for each study. 3. Scope of the measures Maternal employment There was a broad spectrum of how maternal employment was operationalized. 6 studies used exclusively a binary variable indicating whether the mother is employed and 3 studies differentiate between part- time and full-time employment with full-time usually defined as more than 35 hours per week. 13 Work intensity was considered in 13 studies and it was generally measured by continuous variables indicating weekly working hours9 or employment duration, defined as the number of years, months, weeks or quarters in which the mother worked. In one instance (Reynolds et al. 2017) employment duration was measured as the number of weekly hours worked between the two surveys as a fraction of weekly working hours under full-time employment. Many studies considered multiple employment variables simultaneously. For example, being employed and work intensity. Thus, the interpretation holds constant intensity while considering when the work occurred. Finally, in 2 studies additional qualitative aspects of maternal employment were considered while controlling for working hours. Felfe and Hsin (2012) analyzed the impact of mothers’ exposure to work- related hazards and stress on child cognitive and behavioral outcomes. Dunifon et al. (2013) considered standard in contrast to non-standard work schedules. These were measured as dummy variables, and could include night work, weekend work, or hours that changed weekly. Child development outcomes The studies reviewed cover a wide range of outcomes. To better organize the results, we grouped them in 6 categories: behavioral, cognitive, schooling, health, height and weight. Behavioral outcomes are the most subjective as they are most often parent-reported often using a set of questions provided in the Child Behavior Checklist (CBCL), a validated checklist consisting of 99 items assessing a range of problem behaviors relating to socio-emotional well-being. In one study (Mendolia, 2016) different behavioral outcomes such as self-reporting on smoking, low self-esteem, high life satisfaction and intention to leave education were examined. Overall, 8 studies reported behavioral outcomes. Cognitive outcomes, reported in 8 studies, vary as different tests for different realms (reading, mathematics, and receptive vocabulary, for example) are used. These were typically applied by a trained psychologist. The scores are generally not comparable across tests or children’s ages (older children are expected to have higher scores than younger children), so results are standardized; sometimes standardization is done by child age and other times child age is a control variable. Schooling outcomes were examined in 5 studies. The type of measures chosen differed according to whether children are observed during school age or not. In the first case, indicators of enrolment and being on track are used together with continuous variables such as the years of schooling (Dervisevic et al. 2022) or the time spent at school (Afridi et al., 2016). In the latter case, school achievement was measured by respondents’ final grades at high school (Del Boca, 2016), achievement (yes/no) of advan ced level qualification for admission in universities (Ermish and Francesconi, 2013) or by the years of education at age 27 (Haaland, Rege, Votruba, 2013). Four studies included additional health outcomes. These cover a broad range from hospitalizations, injury and asthma episodes (Morrill, 2011) to parent reported perception of general health (Gennetian et al. 2010; Pekkurnaz, 2014) to more specific indicators such as levels of hemoglobin or lung capacity 9 This was defined in most papers as average hours per week. Alternative definitions are present in two articles: Pekkurnaz, 2014 used the number of hours worked in the past week¸ Dunifon et al. 2013 recoded the number of hours worked into five <four?> dummies (1-19; 20-34; 35-44; 45+). 14 (Dervisevic et al. 2022). The breadth of sources from administrative data to surveys make this category of outcomes the most disparate. Height outcomes include height for age z-scores (HAZ) and a binary variable for stunting, defined as low height-for-age, which is a proxy for chronic or recurrent malnutrition. Height is usually a direct observation, measured by a trained interviewer. Height measures cumulative nutrition and is particularly valuable prior to age 5, when growth distributions in healthy populations have been found to be similar worldwide (WHO, 2006). Three studies, all of them from non-Western lower middle-income countries use height outcomes (Rashad and Sharif, 2019; Shajan and Sumalatha, 2020; Dervisevic et al., 2022). Weight outcomes are based on weight for age z-scores (WAZ) or body mass index (BMI), and – depending on the context – capture deviation from normal towards malnutrition or obesity. Seven studies (Von Hinke Kessler Scholder 2008; Hubbard 2009; Anderson et al 2003; Bishop 2011; Pekkurnaz 2014; Greve 2011; Agiro and Huang 2020) considered binary indicators for obesity and overweight, which in most cases where constructed based on information on BMI. Two studies (Haaland, Rege, Votruba 201 and Ziol-Guest, Dunifon & Kalil 2013) use BMI. Three studies included other weight measures such as WAZ (Shajan and Sumalatha 2020) or an indicator of “underweight” (Dervisevic et al., 2022; Rashad and Sharif, 2019). Unlike height, weight fluctuates over time and therefore reflects current and acute as well as chronic malnutrition. The studies had a trained professional take this data. 15 4. Correcting for multiple hypotheses testing Table 1: Bonferroni p-values and FDR q-values for papers with multiple outcomes and without multiple hypotheses corrections Standard Bonferroni Paper Outcome Coefficient Naïve p-value FDR q-value error p-value Overnight hospitalization 0.037 0.013 0.005 0.043 0.035 Overnight hospitalization 0.045 0.017 0.009 0.081 0.035 Overnight hospitalization 0.092 0.038 0.015 0.133 0.035 Injury/poisoning 0.056 0.028 0.047 0.423 0.081 Morill (2011) Injury/poisoning 0.050 0.030 0.099 0.890 0.111 Injury/poisoning 0.074 0.060 0.220 1.000 0.220 Asthma episode 0.116 0.060 0.054 0.484 0.081 Asthma episode 0.143 0.059 0.016 0.141 0.035 Asthma episode 0.192 0.110 0.081 0.733 0.105 Stunted 0.186 0.088 0.034 0.069 0.069 Rashad & Sharaf (2019) Underweight 0.073 0.046 0.113 0.225 0.113 Shajan and Sumalatha Height for age -0.208 0.066 0.002 0.003 0.002 (2020) Weight for age -0.173 0.054 0.001 0.003 0.002 Letter word score -0.055 0.024 0.022 0.066 0.059 Felfe & Hsin (2012) Passage completion -0.049 0.026 0.059 0.178 0.059 Applied problem solving -0.048 0.025 0.055 0.165 0.059 Internal behavioral 0.048 0.025 0.055 0.110 0.110 problems Felfe & Hsin (2012) External behavioral 0.007 0.024 0.771 1.000 0.771 problems Von Hinke Kessler (2008) Pre-school 0.004 Above 0.1 Above 0.3 Above 0.1 16 Above 0.03 Above 0.01 and Above 0.03 and Age 7 0.038 and below below 0.05. below 0.15. 0.15. Age 11 0.019 Above 0.1 Above 0.3 Above 0.1 Years of education 0.013 0.005 0.012 0.050 0.012 Completed high school 0.003 0.001 0.003 0.011 0.004 Furre Haaland (2013) Attended colleage 0.004 0.001 0.001 0.003 0.002 Log inc 29 0.005 0.002 0.001 0.003 0.002 Time at school 6.506 1.102 0.000 0.000 0.000 Afridi et al. (2016) Enrollment 0.472 0.138 0.001 0.002 0.001 Grade progression 0.406 0.101 0.000 0.000 0.000 PIAT reading, year 1 -0.00069 0.00064 0.281 1.000 0.422 PIAT reading, year 2 0.00064 0.00059 0.278 1.000 0.422 James-Burmudy, 2005; PIAT reading, year 3 0.00039 0.00047 0.407 1.000 0.488 weeks worked PIAT math, year 1 -0.00117 0.00067 0.081 0.485 0.422 PIAT math, year 2 0.00034 0.00061 0.577 1.000 0.577 PIAT math, year 3 0.0006 0.00051 0.239 1.000 0.422 PPVT, year 1 0.03651 0.5222 0.944 1.000 0.944 PPVT, year 2 -0.05182 0.05137 0.313 1.000 0.580 PPVT, year 3 -0.00949 0.0469 0.840 1.000 0.944 James-Burmudy, 2005; PIAT reading, year 1 -0.0526 0.02506 0.036 0.322 0.129 hours worked PIAT reading, year 2 -0.02109 0.02131 0.322 1.000 0.580 PIAT reading, year 3 0.01945 0.0864 0.822 1.000 0.944 PIAT math, year 1 -0.07385 0.02595 0.004 0.040 0.036 PIAT math, year 2 0.01512 0.02275 0.506 1.000 0.759 PIAT math, year 3 0.04108 0.02026 0.043 0.383 0.129 Any risky behavior 1.018 0.546 0.062 0.685 0.294 Aughinbaugh and Gittelman (2003) Drank alcohol 0.629 0.246 0.011 0.116 0.116 Smoked cigarettes 0.184 0.269 0.494 1.000 0.764 17 Used marijuana 0.315 0.257 0.220 1.000 0.485 Used other drugs 0.22 0.373 0.555 1.000 0.764 Had sexual intercourse 0.058 0.243 0.811 1.000 0.892 Convicted of crime 0.141 0.355 0.691 1.000 0.845 drank alcohol at least 0.039 0.325 0.904 1.000 0.904 several tiems a month smoked cigarettes 0.438 0.315 0.164 1.000 0.452 everyday used marijuana at least 1- -0.511 0.489 0.296 1.000 0.543 2 tiems per week used no birth control at -0.544 0.311 0.080 0.883 0.294 last sex Externalizing behavior -0.004 0.003 0.182 0.730 0.243 Pilkauskas et al. (2018) Internalizing behavior 0.001 0.002 0.617 1.000 0.617 PPVT 0.007 0.002 0.000 0.002 0.001 Woodcock-Johnson 0.008 0.002 0.000 0.000 0.000 18 References WHO (2006) WHO Multicentre Growth Reference Study Group: WHO Child Growth Standards: Length/Height-for-Age,Weight-for-Age,Weight-for-Length,Weight-for-Height and Body Mass Index-for- Age: Methods and Development. Geneva: World Health Organization. 19 Summaries of papers 9 IVS Agiro and Huang 2020 Baum 2015 Del Boca et al 2016 Dervisevic, Lo Bue, & Perova 2021 Gennetian et al 2010 Publication Journal of Family and Economic Issues Journal of Labor Economics European Journal of Population The World Bank Journal of Health Economics ISFOL-PLUS (Istituto per lo Sviluppo della The data for this study come from the Formazione Professionale dei Lavoratori Indonesia Family Life Survey (IFLS), National Evaluation of Welfare-to-Work Data source Early Childhood Longitudinal Study, NLSY Participation, Labour, & Unemployment SAKERNAS survey in 1995, and data on Strategies Child Outcomes Study (NEWWS- Kindergarten Class (ECLS-K) Survey); child care center information Indonesian tariffs from UN Comtrade. COS) from the Istituto degli Innocenti 2008; 1992 data on the percentage of Years 2013 1986-1996 available places for public & private child 1997, 2000, 2007/08, and 2014/15 1991-1994 care 29,557 observations for the sample of children in two-parent households, and Sample size 1,591 4,944 12786 3435 Sample 32,214 observations in single mother households Location US US Italy Indonesia US (Grand Rapids, Riverside, Atlanta) Mothers who were between 23 and 30 in 1988. The low-income white oversample, Sample restrictions/ the military sample, and children who did children with at least one younger sibling one sample is single mothers welfare recipients Mother characteristics not live with their mother during their first three years were excluded from the analysis. Whether the mother engaged in any Dummy variable for working or not. Employed or not employed or not employed or not self-reports of employment marketplace work Could be full or part-time work. How employment Total number of hours worked in each is measured + Intensity weekly work hours week divided by 40 and the number of number of quarters employed additional weeks in which the mother worked characteristics of Number of weeks in which the mother maternal work Other (eg. Standard worked full time and when the mother schedule or not) first started working after giving birth Child ages when 1, 2 & 3 with Year 1 divided into quarters Children's Ages concurrent 0-2 6-18 mother was working in some specifications (in years unless tests given to children age 3 (PPVT) and 19-30, however grades at the end of high otherwise Ages at outcome in 2nd grade (ages 7-9) age 5+ (PIAT-R & PIAT-M) (children an school were when the student was 6-18 5-9 indicated) measurement average age of 5 yrs) approximate age 18 IV (local unemployment rate; potential earnings proxied by the local per capita income; the percentage of the local labor force that is female; the percentage of the local population that is urban and female; IV ( area of residence & regional supply of local population; percentage of the local child care: number of child care centres, IV (variation in youngest sibling's eligibility population with a high school education the slots available, and number of IV (reduction in tarriffs in female- IV (random assignment to an Methods Primary Causal for kindergarten) and a college education; percentage of children admitted) These variables used intensive industries) experimental welfare-to-work program) the local population employed; the as instruments for both child care percentage of the local labor force in attendance and maternal work. manufacturing, in wholesale/retail trade, in services, in government work, and self- employed; and per capita government transfer payments) high grades at the end of high school: dummy variable that equals 1 when the individual achieved the grade equal to or higher than 55/60 (for individuals who graduated before 1999), or 90/100 (for years of schooling, currently enrolled, and Measure individuals who graduated after 1999, on-grade ("age and schooling") when a law changed the high school grading scale); otherwise it is zero. Outcome: [Results robust to different grade cut- Schooling offs.] maternal work lowers the "age and schooling" indicator and increases years of schooling by around 1.8 years and the Findings null likelihood of being enrolled by around 35 percentage points with sufficiently low q- values to maintain significance with the FDR correction. PIAT-R & PIAT-M (5+ yrs) and PPVT (3-4 yrs) test scores. Test scores were Measure averaged if there were multiple Outcome: observations of the same test Cognition when found, generally a negative impact Findings of maternal work on cognition, but statistically insignificant in the IV models Measure Outcome: Behavior/ socio-emotional Findings Measure height for age, stunting Outcome: Height (haz) children from working mothers, compared to childen from now working mothers, have better height for age score Findings (around 1 standard deviation higher) and are around 49 percentage points less likely to be stunted Overweight based on BMI (percentile Measure wasting, underweight Outcome: used as robustness check) Weight (waz, bmi, obesity, overweight) Findings null null Parent-reported child health scale (1-5) turned into a dichotemous variable for Measure hemoglobin, lung capacity excellent or very good health in contrast to good, fair, and poor health Outcome: Health children from working mothers, Maternal employment reduces likelihood compared to childen from now working that child’s health is reported as excellent Findings mothers, have higher (around 2.2 gr. per of very good (as opposed to good, fair dl) levels of hemoglobin. Null results for and poor) lung capacity 20 IVS continued Greve 2011 Morrill 2011 Rashad and Sharaf 2019 Shajan and Sumalatha 2020 Publication Labour Economics Journal of Health Economics Oxford Development Studies Journal of Public Affairs Danish Longitudinal Survey of Children & Data source National Health Interview Survey (NHIS) Demographic and Health Survey (DHS) National Family Health Survey (NFHS) administrative register data Years 2003 1985-2004 2014 2015-2016 Sample size 4336 mothers and 4348 children 274,842 children 12,888 37,557 Sample Location Denmark US Egypt India children ages 7-17 with at least one Sample restrictions/ younger sibling whose age range is Mother characteristics around 5 years maternal employment dummy variable current employment & employment in Employed or not employed or not for work in the past 1-2 weeks the previous 12 months How employment is measured + Intensity average weekly working hours additional characteristics of maternal work Other (eg. Standard type of employment considered in OLS schedule or not) specification Child ages when mother 3 and 7 younger sibling is around age 5 concurrent Children's Ages was working (in years unless Ages at outcome otherwise indicated) 7 1/2 7-17, contemporaneous employment 0-5 0-5 measurement IV (cluster average of women's working IV for maternal work at age 3 with local IV (youngest child's eligibility for IV (proportion of women in the district Methods Primary Causal status, with the exclusion of the women's unemployment rate as instrument. kindergarten) who work) own employment status) Measure Outcome: Schooling Findings Measure Outcome: Cognition Findings Measure Outcome: Behavior/ socio-emotional Findings Measure Stunting HAZ Outcome: Height (haz) IV estimates are -0.21 (HAZ) of a standard Maternal employment increases the Findings deviation lower for children of employed probability of stunting by 18% . women. child overweight, defined by BMI cut-off Measure above 97th percentile; robustness using Wasting. overweight & underweight WAZ Outcome: the CDC cut-offs Weight (waz, bmi, obesity, overweight) Maternal employment increases the IV estimates are -0.17 (WAZ) of a Findings null probability of wasting by 13%. Null standard deviation lower for children of effects for overweight and underweight. employed women. Hospitalized overnight (last 12 months), Measure Asthma episode (last 12 months), Injury or poisoning episode (last 3 months) Outcome: Health Maternal employment results in an increase in the probabilty in overnight hospitalizations (4-9 percentage points), Findings injury/poisoning (5-7 percentage points), asthma episode (11-19 percentage points) 21 5 FE & LDV Dunifon et al 2013 Felfe & Hsin 2012 Pekkurnaz 2014 Pilkauskas et al 2018 Von Hinke Kessler Scholder 2008 Publication Developmental psychology Economics of Education Review North Carolina Psychology Unpublished dissertation, The University ofDevelopmental at Chapel Hill Health Economics Early Childhood Longitudinal Study-Birth Data source Fragile Families PSID CDS, & O*NET Fragile Families NCDS Cohort (ECLS-B) 2001/02, 2003/04, 2005/06, fall of 06, fall Years 2002 & 2004 1997 & 2002 ~2000, 2002, 2004, 2008 1958, 1965, 1969, 1974 of 07 Sample Sample size 2,367 1,630 1,450-8,900 depending on outcome 2011 3350 Location US (20 large cities) US US US (20 large cities) UK Cohort recruited from hospitals, oversampling nonmarital births. Half of Sample restrictions/ mothers who are working at the time of Cohort recruited from hospitals, the sample is non-Hispanic Black, one excludes single parents Mother characteristics the survey oversampling nonmarital births. quarter is Hispanic, and nearly one quarter is non-Hispanic White. Dummy variables for not working, working standard schedule, and working a non-standard schedule. The last group Indicators for full-time and part-time for How employment Employed or not full-time & part-time work dummies was divided into four non-mutually pre-school, age 7 and 11 employment is measured + exclusive categories: weekends, nights, additional evenings, and different times each week. characteristics of dummies for number of hours worked: 1- maternal work Intensity work hours hours worked in the past week number of months employed 19; 20-34; 35-44; 45+ Other (eg. Standard number of jobs working for 2 weeks or work hazards & work stress schedule or not) more Children's Ages Child ages when preschool, 7 & 11 for LDV or preschool, 7, concurrently for child FE, age 5 for LDV past 5 years concurrent 0-5 (in years unless mother was working 11 & 16 for FE otherwise Ages at outcome 3&5 5-17 0-4 5&9 16 for LDV or preschool, 7, 11 & 16 for FE indicated) measurement child FE & LDV (with LDVs instrumented with county- & state-level variables such as unemployment rate and poverty rate LDV (it controls for birth weight) & child Methods Primary Causal Child FE & LDV LDV & Child FE Child FE & LDV likely to affect working decisions, FE childcare prices and other factors influencing childcare availability) Measure Outcome: Schooling Findings Vocabulary, Reading, Applied problem Measure PPVTfrom Cognitive ability (standardized composite index & WJvarious tests) solving LDV: months employed between age 3 and 5 results in higher PPVT scores (0.005* SDs) and WJ scores (0.007** SDs) Consistent employment was not Outcome: FE estimates significant and negative (- significantly different from fewer years of Cognition 0.004- -0.005) for all outcomes employment. considering an additonal hour worked. FE: An increase in the number of months Findings Slightly smaller & not significant in LDV FE & LDV both null employed in the two-year period specification. Hazards significant in all between ages 1 and 3, and ages 3 and 5, specificaitons, about -0.05 reduction in resulted in a 0.004* SD higher PPVT score the standardized cognitive score. and a 0.006* SD WJ score. Once the number of jobs were controlled, the coefficient on months of employment was no longer significant, although the point estimates were very similar. anxious/depressed and aggressive Internal & External Behavior Indices Behavior problems (standardized internalizing & externalizing subscales of Measure subscales of the CBCL (Peterson & Zill) composite index) the CBCL Child FE: no significant impacts Outcome: LDV: Mothers' night shift work is Behavior/ associated an increase in anxious socio-emotional behavior (0.21 sd) and aggressive Results for behavior problems are null for Behavior problems reduce -0.096 for each behavior (0.13 sd) compared to not hours worked. An increase in internalizing Findings 10 hours worked in the LDV specification, Null results in all Child FE & LDV models working; mothers with other types of behavior (0.05) associated with work FE null work had children whose behavior had stress. insignificant differences in comparison to children whose mothers were not working. Measure Outcome: Height (haz) Findings Measure Obesity & overweight overweight (from BMI) Outcome: Weight (waz, bmi, obesity, overweight) ~5% increase in overweight at age 16 Obesity reduces -0.035 for each 10 hours Findings (LDV) and concurrently (FE) with full-time worked in the LDV specification, FE null work at age 7 Health status (excellent or very good), ear Measure infection, respiratory illness Outcome: Health Health status increases 0.029 with hours Findings worked in the LDV specification, FE null 22 5 Mother FE Aughinbaugh and Gittleman 2004 Haaland, Rege, Votruba 2013 Waldfogel, Han, Brooks-Gunn 2002 Ziol-Guest, Dunifon & Kalil 2013 Ermish and Francesconi 2013 Publication Journal of health economics CESifo Working Paper Demography Social Science and Medicine Journal of Applied Econometrics National Longitudinal Survey of Youth British Household Panel Survey Data source NLSY79 & young adult supplement Combined registry databases NLSY (NLSY) (BHPS) Years 1994, 1996, 1998, and 2000 1970-2007 1990, 1992, 1994, 1996 1994-2008 1991-1997 462 to 1414 depending on the outcome 1026 (full sample); 647 (restricted Sample size 165,957 children in 77,581 families 548 2914 siblings from 1247 mothers variable. sample) Location US Norway US (nationally representative) US UK (nationally representative) Sample Full sample: children who: (i) are aged 18 or more and were born between 1970 and 1979; serious Dropped families where siblings would disabilities; (iii) lived with their have differential exposure to parental biological, adoptive or step divorce or unusual living arrangements. women in the NLSY who ranged in age parent(s) during the first seven Sample restrictions/ overly represented young mothers based Exclude children whose mothers or from 16 to 32 at the time of their waves of the panel study; and (iv) Mother characteristics on child age fathers died before the child reached age children’s birth. have complete information 16 and children who do not have siblings employment patterns during represented in the sample. childhood and other variables related to her. Restricted sample: full sample aged 16-17 when they lived with their parents Employed during 1st year, 2nd or 3rd Employed or not Entry into full time employment year, from age 3 up to the year before How employment assessment, at time of assessment. is measured + Mother's hours of work were available Number of months spent in full- additional for the first 3 years of the child's life and average hours per week worked over time and part-time paid work. Intensity Hours worked during 1st year characteristics of for the 3 calendar years preceding the child's lifetime Worked at least 1 month when the maternal work adolescent interview child was aged 0-5 Other (eg. Standard schedule or not) Children's Ages Child ages when 0-3 and the 3 years prior to the 0-18 1st year, 2-3 yrs, 4+ years child's entire lifetime 0-5 (in years unless mother was working adolescent survey otherwise Ages at outcome 3-4 yrs: vocabulary; 15-18 27 or when enter military for boys 13-14 aged 18 and more indicated) measurement 5-6 yrs & 7-8 yrs: reading & math Methods Primary Causal Mother FE Mother FE Mother FE Mother FE Mother FE dummy (yes/no) for having years of education at age 27, college Measure achieved an 'A (Advanced)-level' attendance rates, log earnings age 29 qualification or higher qualification 5 additional years of full-time Negative and statistically significant Outcome: employment by one's mother reduces a effect of full and part time Schooling child's education by 0.065 years, which employment. Its point estimate amounts to 4 percent of a standard ranges between 4 and 11 (full time) deviation in our sample. For high-school Findings or between 3 and 7 (part time) completion, college attendance & income, percentage points lower probability effects are significant but small, for each additional year of full time increasing by 1.5%, 2% and 2.5% employment, depending on the respectively due to an increase of 5 years estimator. of mothers' work. IQ scores (boys only from military Vocabulary (PPVT-R), Reading (PIAT) & Measure records) Math (PIAT) Outcome: Cognition Primarily null; A significant (5% level) result is only found for the coefficient on Findings Insignificant impacts on IQ (boys only). reading (PIAT) ages 5-6 for employment during 2nd or 3rd year of life Incidence: Any risky behavior (of those that follow), drank alcohol, smoked cigarettes, used marijuana, used other drugs, sexual intercourse, convicted of Measure crime. Thresholds: Drank alcohol several times a month or more, smoked cigarettes everyday, used marijuana at least once a week, no birth control at last Outcome: intercourse Behavior/ No impact of maternal employment socio-emotional during adolescence on participation in risky behavior. More hours of maternal employment during ages 0-3 is associated with an about 9 percentage point Findings increase in the likelihood of ever having drunk alcohol with maternal FE model and a simlar likelihood of having used contraception the last time having sex with the grandparent FE model. Measure height (boys only from military records) Outcome: Height (haz) impacts on height are insignificant (boys Findings only) Measure BMI (boys only from military records) BMI Outcome: Weight (waz, bmi, obesity, overweight) impacts on BMI & are insignificant (boys Findings null only) Measure Outcome: Health Findings 23 7 Mixed Methods Afridi et al. 2016 Anderson et al 2003 Bishop 2011 Hubbard 2009 - Ch 1 Publication IZA Journal of Labor & Development Journal of Health Economics Economic Record Ph.D Thesis - The University of North Carolina at Chapel Hill Data source YLS NLSY HILDA ECLS-K fall 1998, spring 1999, fall 1999, spring Years 2007 & 2009-2010 biannually from 1986-1998 2,007 2000, spring 2002, spring 2004 16,650 child-year observations from 6283 Sample Sample size 3,725 mothers; sample size for FE ranges from 907 (IV), 430 (mother FE) 18,990 person-years for 6,330 indivdiuals 4,159 for child FE to 7919 for sibling FE Location India (state of Andhra Pradesh) US Australia US Sample restrictions/ Sample restricted to households in rural Mother characteristics areas in both periods Dummy taking value 1 if the mother Sample restrictions/ How employment participates in the labor market and 0 Mother characteristics is measured + otherwise additional mother's employment hours, divided into characteristics of average hours per week (when working) number of years working full and part- Employed or not not working (0-8), part-time (9-35), full maternal work and total weeks worked time time (36+) Intensity also if in childcare>5 hours per week 12-15 (The exception is 19-year-olds, for cumulative from beginning of life Children's Ages Other (eg. Standard whom adolescence is the 3-year period concurrently (although one simple probit considers concurrently (in years unless schedule or not) between the ages of 11, 13 and 15, as work before & after age 3) otherwise necessitated by the 7-year panel) indicated) Child ages when between 3 & 11; varies based on kindergarten, 1st, 3rd & 5th grades 5--14 15-19 mother was working specification (approx ages 5, 8, & 10) FE; some specifications have lagged Child FE & IV. Instruments used: lagged outcome variable; Discrete Factor Child FE, Mother FE, IV ( unemployment rainfall and lagged NGRES funs IV (local unemployment rate by industry Random Effects (with IV). Instruments Ages at outcome rate, child care regulations, wages of child Methods sanctioned at the beginning of each and typical hours worked by women in that are state or county & time variant: 7 measurement care workers, welfare benefit levels, and financial year. Both instruments are at that industry locally) & mother FE employment-related variables & 6 child- the status of welfare reform) the mandal (subdistrict) level. care-related variables; also instrument with child's birth weight time spent at school; school enrolment; grade's progression (defined as the actual Primary Causal grade attainment of a child divided by the grade the child should have completed at her/his age) The authors find that if the mother works, her child’s time spent in school Outcome: (significantly) increases by 6.506 h in a Schooling day. The probability of a child being enrolled in school (significantly) rises by Measure 47.2 percentage points. Moreover, there is a significant effect of mother’s work participation on her child’s grade progression: the gap between a child’s actual and ideal grade declines by 40.6 percentage points. Findings Outcome: Cognition Measure Findings Outcome: Behavior/ socio-emotional Measure Findings Outcome: Height (haz) Measure overweight, determined by the CDC's BMI overweight & obese, as determined by Findings BMI & Overweight guidelines BMI Outcome: Weight (waz, Child FE: children of mothers who work When mothers work full-time and do not bmi, obesity, an additional 10 hours per week while Not statistically significant relationship use child care, the risk of obesity overweight) Measure working face a 1.5 percentage point with both full & part time employmentin decreases by 1.5 percentage points and increase in the likelihood of being the IV & sibling FE specifications decreases risk of being overweight by overweight. IV: Null around 2 percentage points Findings Outcome: Health Measure 24 Mixed Methods, continued James-Burdumy 2005 Mendolia 2016 Reynolds, Fernald & Behrman 2017 Publication Journal of Labour Economics Journal of Family and Economic Issues Social Science and Medicine - Population Health British Youth Panel (BYP), a complement Longitudinal Survey of Early Childhood NLSY to the British Household Panel Survey (ELPI) (BHPS) Data source 1986 & 1988 1994-2006, yearly waves 2010 & 2012 Sample around 5,000 (number of observations Years 498 (PPVT), 1,761 (PIAT) 2476 reported, not number of children) Sample size US UK Chile Location mothers ages 21-27 in 1986 employed mothers Working mothers Sample restrictions/ How employment Mother characteristics is measured + additional Fraction worked of the two-year period characteristics of hours worked in the 1st, 2nd & 3rd year of Employed or not hours per week between surveys, calculated from both maternal work life hours work and employment duration Intensity Children's Ages Other (eg. Standard 0-3 concurrent 1-3 years (in years unless schedule or not) otherwise indicated) Child ages when 5-18 for PIAT, 3-5 for PPVT 11-15 1 & 3 years mother was working LDV, IV (mothers employed before child born, percent of other mothers with Ages at outcome Mother FE and IV (percent of county labor Methods Child FE & Mother FE children the same age working part & full measurement force employed) time, community gender equality index and traditional values index) Primary Causal Outcome: Schooling Measure PPVT (ages 3-5) & PIAT reading & math Findings PPVT, Battelle (ages 5-18) Average hours worked per year in first 3 years of child’s life do not significantly affects child's PPVT score. Analogously, no Outcome: significant effect of hours worked in the Cognition first, second or third yeard of the child's life on PPVT scores and PIAT readings scores. Measure null Only hours worked in year 1 are associated with lower PIAT math scores: Working 1 additional 8-hour day each week in year 1 (an increase of 416 hours for the year) would reduce the PIAT math score by 0.5 points relative to the mean of 96. youth reports of: smoker, low self Findings esteem, high life satisfaction, and Behavior (CBCL) intention to leave education at age 16 Outcome: Behavior/ socio-emotional Measure null null Findings Outcome: Height (haz) Measure Findings Outcome: Weight (waz, bmi, obesity, overweight) Measure Findings Outcome: Health Measure 25