Women’s Political Reservation, Early Childhood Development, and Learning in India

This article analyzes the long-term impacts of reservation of local political seats for women on children’s learning and nutritional outcomes in rural Andhra Pradesh. Using the random rotation of seats reserved for women over different election cycles and three rounds of a panel data set, we analyze the impact of exposure to political reservation during critical periods in early childhood. The article shows that the reservation policy for female leaders had the largest impact on learning outcomes in primary school when children were exposed to reservation in utero and very early in life.


I. Introduction
Early childhood experiences are often thought to be crucial determinants of children's cognitive and noncognitive development (Heckman 2008). Many children in the developing world are exposed to a multitude of risk factors, hampering their cognitive development from very early on (Grantham-McGregor et al. 2007). This is thought to lead to subsequent lack of human capital accumulation, resulting in poverty and continued inequality. Increasingly, policies therefore focus on improving health, nutrition, and educational outcomes during early childhood. And as there typically is a strong relationship between mother's characteristics and children's cognitive and educational outcomes, policies addressing constraints on the mothers of these children are often believed to be particularly promising (World Bank 2011). This is even more so if women have stronger preferences for human capital investments (as hypothesized, for instance, by Thomas [1990Thomas [ , 1993), and policies that increase women's decisionmaking power could potentially have large implications for the human capital of the next generation. Yet very little is known about the longer-term impact of such policies on children's human capital outcomes.
This article contributes by analyzing the impact of women's representation in elected bodies on children's learning and nutritional status more than 10 years later and tests whether political reservation for women resulted in improvements in children's longer-term learning and nutritional outcomes. Understanding the The data used in this publication come from Young Lives, a 15-year study of the changing nature of childhood poverty in Ethiopia, India, Peru, and Vietnam (http://www.younglives.org.uk). Young Lives is core-funded by UK aid from the Department for International Development (DFID) and cofunded from 2010 to 2014 by the Netherlands Ministry of Foreign Affairs. The views expressed here are those of the authors. They are not necessarily those of Young Lives, the University of Oxford, DFID, or other funders. We are grateful to the Young Lives team for assistance with the data and to the associate editor, two anonymous referees, Marc Gurgand, and participants at the 2013 Young Lives conference for helpful comments on an earlier version of this article. All remaining errors are ours.
impact of female reservation policy on children's human capital accumulation is of high interest given existing evidence on children's delays in physical growth, cognition, and learning outcomes in India. Stunting rates, for instance, have remained extremely high in India despite economic growth (Deaton and Dreze 2009;Jayachandran and Pande 2015), and lack of learning in school has been highlighted as a particular challenge (Pritchett and Beatty 2012;Muralidharan 2013).
The 73rd Amendment of the Constitution of India of 1992 prescribes that one-third of the seats in all local councils, as well as one-third of the leadership positions, must be reserved for women. 1 These local councils can affect children's outcomes through their responsibilities for public good provision in the Gram Panchayat (GP, the lowest administrative level). They also decide the allocation of private benefits, such as housing or toilet construction, public employment, or food subsidies for the poor. Exploiting the random allocation of the seats reserved for women in local elections, several studies have shown that reservations had important impacts on policy choices in the short term. Chattopadhayay and  show that female reservation increased investment in infrastructure related to the expressed development priorities of women in West Bengal and Rajasthan. Duflo and Topalova (2004) further show for a sample covering 24 states that GPs reserved for women invest more in public goods and that the measured quality of these goods is at least as high as in nonreserved GPs. This is particularly the case for drinking water infrastructure. Bhalotra and Clots-Figueras (2011) find that female leaders are more likely to invest in rural health infrastructure than men. And Beaman et al. (2009) show that reservation for women not only has led to a decrease in bias among voters against women candidates but also has resulted in a subsequent increase in the percentage of female local leaders. More broadly, political reservation for the Scheduled Caste, Scheduled Tribe, and women has been linked to more policy influence for these groups (Pande 2003). Pande and Ford (2011) further show the effectiveness of quotas in changing attitudes toward women leaders and giving them more policy influence. And exposure to female leaders can raises parents' aspirations for their girls (Beaman et al. 2012) and increase women's participation in political processes (Beaman et al. 2010). Reservation can also otherwise affect women's outcomes, with evidence by Jha, Nag, and Nagarajan (2011), for instance, documenting how improvements in water availability lead to higher participation of women in the labor market and more productive work. Ghani, Kerr, and O'Connell (2013) find a positive impact on female entrepreneurship. Focusing more specifically on child-related indicators, the available evidence suggests short-term positive impacts, but also substantial heterogeneity of such impacts. Clots-Figueras (2012) shows that female leaders have a positive effect on primary education attainment of children in urban India, though not in rural areas. Beaman et al. (2012) suggest that education investments might be particularly important for girls, as having a GP reserved for a female leader reduces the gender gap in school attendance. And Kumar and Prakash (2012) similarly find positive impacts of women leaders on child mortality in Bihar, but the impacts are found only for the wealthier quintiles of the population. Less is known about impacts on children's health and nutrition status during early childhood.
More generally, not all the available evidence points to positive impacts. Ban and Rao (2008) show that women's reservation in four southern states, including Andhra Pradesh, does not lead to more investments congruent with women's preferences. Bardhan, Mookherjee, and Para Torrado (2010) further show that female reservation can lead to worse targeting of private benefits within villages and, as such, potentially limit the impact of the public good investments. Yet Besley, Pande, and Rao (2005) find no such impact on targeting. Sathe et al. (2013) find better service delivery in villages with female leaders who have been in power for 3.5 years, but worse in those where female leaders only recently came to power. And relatedly, Afridi, Iversen, and Sharan (2013) find more leakage in National Rural Employment Guarantee Act programs with female leadership but improvement in governance as female leaders gain more experience. Deininger et al. (2015) also found that quality of service provision might be worse in the short run but better in the long run. Finally, Iyer et al. (2012) show how female reservation increased reporting of crime against women, which they interpret as a positive development.
Taken together the literature hence suggests a relatively wide variety of mechanisms through which the reservation policy could affect children's human capital. If female leaders indeed favor policies beneficial to human capital, they could increase investments in education and health care infrastructure or vaccination programs. They may also affect children's food intake, for instance, by increasing the number of households benefiting from the public food distribution system. Increased bargaining power resulting from higher earning potential can enhance some of these effects. But giving women political power does not necessarily lead to better human capital investments, as it is not automatically true that women would favor such investments (Duflo 2012). Indeed, both Chattopadhayay and  and Ban and Rao (2008) show that female and male preferences can be quite heterogeneous between GPs, so that one would not necessarily expect reservation to lead to similar investments in all GPs. And if reservation has positive impacts on women's labor market outcomes, this may come at the cost of time spent for child care. That said, even if women focus more directly on their own needs, such as their access to food and preventive health care, this could result in indirect benefits for children, in particular through better protection during the critical in utero period. Any longer-term impacts may well depend then on the ages at which children were exposed to potential benefits from having a female elected leader.
This article builds on the existing literature and contributes by showing long-term impacts of reservation of seats for women in the GP elections on children's learning and nutritional outcomes in rural Andhra Pradesh. We do not aim to attribute these results to specific policy choices by female leaders, but instead analyze potential critical periods during early childhood in which such policies can be particularly effective. Specifically, we analyze children's test scores at age 8 comparing children exposed to reservation in utero and very early in life with children exposed after age 5. Using the three waves of the Young Lives panel data set that follows children from 2002 to 2009, we show that women's political reservation improved nutritional and health outcomes very early in the children's life, and we measure the longer-term impacts on children's test scores capturing both math and language achievement and on their nutritional status. We are able to attribute causality to the reservation policy in the Panchayat as the seats to be reserved were randomly allocated through a rotating mechanism. Specifically, we focus on a cohort of children (12-17 months in 2002) who were exposed to reservation in utero if their GP was randomly selected to receive reservation between 1995 and 2001, while they benefited from reservation only after the age of 5 if their GP was randomly selected for reservation between 2006 and 2011. We then compare the outcomes for this cohort when they are 8 years old between GPs reserved in 1995 versus 2006. This allows drawing lessons regarding the relative impact of being exposed to reservation in utero and very early in life.
The results show that children in GPs that were randomly reserved for women prior to the child's birth have significantly better outcomes on test scores in 2009 when compared to GPs that were reserved after the child turned 5. As such, this article also contributes to the literature on the long-term impacts of early-life interventions (Barnett 1995;Garces, Thomas, and Currie 2002;Walker et al. 2005;Maluccio et al. 2009;Barham 2012) and the critical periods for such interventions (Barker 1990;Adair 1999;Doyle et al. 2009;Almond and Currie 2011;Barham, Macours, and Maluccio 2013).
The article is structured as follows: Section II discusses the data sources and provides details about the reservation policy, the random rotation of reserved seats, and the balance for the cohort of interest. It further discusses identification and the empirical specification. Section III then shows the long-term impact of having a woman legislator on child learning and nutritional outcomes, comparing exposure in utero and very early in life with exposure after age 5. Section IV provides further evidence for other cohorts, and Section V presents conclusions.
II. Data, Identification, and Empirical Strategy A. The Reservation Policy and the Cohort of Interest We use the Young Lives Survey (YLS) data set for Andhra Pradesh, a threewave panel data set (2002, 2007, and 2009) focused on childhood poverty covering 20 mandals in six rural districts. 2 The YLS survey instrument consists of a child questionnaire, a household questionnaire, and a community questionnaire. YLS follows 2011 children who were between 6 and 17 months of age in 2002. 3 For the initial sampling, each mandal was divided into four geographical areas, and one village was randomly selected in each area. YLS then randomly sampled 100 households with a 1-year-old child in each mandal. Our overall target sample in particular consists of 1,547 children, as we exclude 464 children from urban areas, where the GP reservation is irrelevant (the local election system being different for urban areas).
The Panchayat is a system of village-level (Gram Panchayat), mandal-level (Panchayat Samiti), and district-level (Zilla Parishad) councils, with membership determined through local elections. Their main responsibility is the administration of local public goods. Each mandal consist of various GPs, and each GP encompasses between 1,000 and 10,000 individuals in a group of villages (between one and 15). In 1992, the 73rd Amendment of the Constitution of India gave new powers to the Panchayats and provided that one-third of the seats in all Panchayat councils, as well as one-third of the leadership posi-tions, must be reserved for women. Seats and leadership positions were also reserved for the two disadvantaged minorities in India, Scheduled Castes and Scheduled Tribes, in the form of mandated representation proportional to each minority's population share in each district. 4 In Andhra Pradesh, for all practical purposes, this act came into effect with the Andhra Pradesh Panchayat Raj Act of 1994, which mandates the elections for the seat of the sarpanch, or the head of the GP. For each subsequent election, one-third of the total seats were reserved for women, and this was to be done by rotation. 5 Only women candidates could contest the election for these seats, thereby ensuring a woman sarpanch in the reserved GPs, without any exception. The first Panchayat or GP elections after the Andhra Pradesh Panchayati Raj Act were held in March 1995. The next elections took place in July-August of 2001 and 2006. Given the rotation rule, a GP reserved in 1995 could not be reserved in 2001 or 2006, and by 2006 every GP had been reserved once. There are hence three treatment episodes relevant for this article : 1995-2001, 2001-6, and 2006-11. The State Election Commission typically announced the dates for the elections 1-2 months in advance, and the term of the newly elected sarpanch started immediately after each election.
The first round of the YLS was conducted from September to December 2002, approximately 1 year after the sarpanch elected in 2001 came into power and 7 years after the first election with reservation. See figure 1 for a timeline of elections and data collection. At that point the children in the sample were 6-17 months old. The second round of data was collected from January to July 2007, that is, 6-12 months after the 2006 election. The third round of YLS was mostly conducted in the second half of 2009, that is, more than 3 years after the 2006 election. The YLS hence does not allow characterizing the situation prior to the start of the reservation policy. Given the randomized rota-4 In this article, we use the term "reservation" specifically for the reservation of the sarpanch position for women. We do not consider the reservation for minorities for lack of exogenous variation in this group. 5 The Andhra Pradesh Panchayat Raj Act of 1994 mandates that the office of sarpanch of the GPs in the state be reserved in the following manner: (a) Of the total number of GPs in the state, the state election commissioner shall reserve a number of GPs for Scheduled Castes (SC) and Scheduled Tribes (ST), such that the proportion of reserved GPs to total GPs reflects the ratio of the SC/ST population to the total population. (b) This reservation of seats for SC/ST has to be done by rotation, which implies that a GP reserved for SC/ST in the first election year is not to be reserved for SC/ST in the next election year. An exception to this case, though, is possible in theory subject to the constraint that the total number of seats reserved must reflect the proportion of population of SC/ST in the state. (c) A minimum of one-third of the GPs reserved for SC/ST must be reserved for women. This is also to be done by rotation. (d ) A minimum of one-third of the GPs, including the ones reserved for SC/ST, are to be reserved for women. This is also to be done by rotation (see http://www.apsec.gov .in/). tion of the reservation, we do not need such baseline information for identification. The lack of a baseline does imply that only the characteristics unlikely to have been affected by the reservation can be used as control variables. The major advantage of the YLS data is that they allow us to track the medium-to longer-term impacts of the reservation policy at the child level and notably to investigate to what extent impacts on health and nutritional gains, very early in life, are reflected in longer-term impacts on learning and nutritional status.
For identification purposes the key feature of the reservation policy is that the reserved seats were randomly allocated. It is this random allocation that most other papers on women's reservation in India-as reviewed above-have also relied on for identifying the causal impact of the reservation. Ban and Rao (2008) describe the process of assignment of women reservation to GPs in some states of South India, including Andhra Pradesh, as follows: GPs to be reserved for SC/ST in the first step and GPs to be reserved for women in the next step are drawn by arranging the GPs in descending order of the ratio of the population of the respective category for which it is to be reserved. Then the first, fourth, seventh, and so on numbered GPs on the list were assigned reservation for the 1995 election. The second, fifth, eighth, and so on GPs are assigned reservation for the 2001 election. And the third, sixth, ninth, and so on GPs are assigned reservation in the 2006 election. Note that this rule implies that GPs reserved in 1995, on average, should have slightly higher female population ratios but that conditional on these slight differences, assignment is random. This may nevertheless be a potential concern for identification if higher female population ratios are also related through other mechanisms to children's longer-term outcomes. We return to this point below.
Given that the YLS data sampled children aged 6-17 months in the fall of 2002, reservation had been in place approximately 6 years prior to their birth in the GPs reserved from 1995 to 2001. While they had limited direct exposure to reservation after birth, they may have continued to benefit from policies implemented during reservations. In addition, for the 12-17-month-olds, reservation was still in place when they were in utero. Children who were between 6 and 11 months old in the fall of 2002, on the other hand, had limited in utero exposure. In contrast, in the GPs reserved from 2001 to 2006, reservation starts almost at the same time or just before the children in the sample are born, and reservation was in place until they were about 5 years old. As it is likely that the results of reservation on policy implementation were not immediate, they may not have benefited much in utero or very early in life. Finally, in the GPs reserved from 2006 to 2011, reservation took place when children were older than 5, that is, at the moment they were entering primary school age.
Hence there are potentially important differences between cohorts as well and GPs in terms of being affected by reservation. In particular, we focus on children aged 12-17 months in the first round of the YLS survey: they were exposed to reservation in utero if their GP was randomly selected to receive reservation in the 1995 election, while they benefited from reservation after the age of 5 only if their GP was randomly selected for the 2006 reservation. By comparing the outcomes of this cohort between 1995 and 2006 reservation GPs in 2009, we can hence draw lessons regarding the relative impact of being exposed to reservation in utero and very early in life. By the time the third round of the survey is conducted in 2009, the children in the sample are about 8 years old. This allows evaluating the longer-term impacts of the reservation policy and, in particular, analyzing whether the differential timing of exposure leads to longer-term differences in learning and nutritional outcomes.

B. Random Rotation of Reservation and Balance
We first use data from the 1991 census to verify that villages in GPs reserved in 1995GPs reserved in , 2001GPs reserved in , and 2006 were indeed similar prior to reservation. To do so, we merge the 1991 census data with the GP election reservation data for Andhra Pradesh for the 660 villages in mandals included in the rural YLS sample. The election data are freely available and come from the Andhra Pradesh State Election Commission. We were able to obtain election data for the years 2001 and 2006. As expected, there are no GPs that were reserved for a female sarpanch in both years. Indeed, since the election reservation is done on a rotation basis and each election had female reservation for one-third of the GPs, these data also implicitly contain the election reservation data for the 1995 election (the GPs that did not receive reservation in 2001 and 2006 must have received reservation in 1995). Because of differences in village names between the election and the census data, the merge between the two data sources is imperfect, but we manage to merge 289 out of 660. For those that can be merged, we find very few differences between villages reserved in 1995 and 2006 and notably no clear differences in health, water, or education infrastructure. 6 As expected, the female population ratio in the GPs reserved in 1995 is slightly larger than in the GPs reserved in 2006. The difference is small (0.501 compared to 0.493) though significant at the 1% level. Overall, these results confirm previous findings in the literature regarding the random rotation of female reservation.
The YLS cannot be merged with the census data for confidentiality reasons. 7 We hence cannot repeat this exercise for the specific villages in the YLS sample. But given the random selection of villages into the YLS sample, there is no a priori reason to believe these results would be different. That said, as was the case for the census data, not all the YLS villages could be merged with the election reservation data (election data were merged for 59 out of 82 villages), and this imperfect match could create imbalances. To verify this, we use both the YLS village and household surveys of 2002 and focus on characteristics unlikely to have changed because of reservation policies. 8 Table 1 first analyzes whether villages that could not be merged to the election data differ from those for which the reservation status is known. Overall, there are few significant differences, with only one of the 16 village-level variables as well as one out of 10 from the household survey being significantly different. Overall, these results suggest no large selection bias into the sample. Table 2 then shows the same characteristics by the reservation status of the GP to which the village belongs. 9 It shows in particular that the villages that were reserved in 1995 and 2006 are very similar in observed characteristics, with the exception of the average number of female adults per household in the village. 10 Also at the household and child levels, variables unlikely to be affected by the reservation policies such as mother's age and ethnicity, child's gender, and birth order seem to be well balanced, suggesting there was no selective 6 We compared the 1995 reservation villages with the 2006 reservation villages for 84 characteristics covering demographics (including female literacy and labor force participation), education, health, water infrastructure, power, and other services. Three variables are significant at the 5% level and one at 10%. Similar results are obtained for the 2001-6 comparison or when using villages of all mandals in Andhra Pradesh. Detailed results are available from the authors. 7 The election reservation data were merged with the YLS data set by the YLS team, and the names of the GPs were removed to guarantee confidentiality and assure anonymity. 8 We exclude variables with very little variation in the data. For instance, we do not report whether agriculture is reported as one of the main income sources because this is the case in all villages. 9 Because of the YLS sampling design, every village in the sample corresponds to a different GP. We hence use village and GP interchangeably when discussing the results. 10 In contrast to the census data, the sample shows balance on the estimated female population ratio. The reason could just be the small number of villages or that the female population ratio in the YLS is a noisy estimate for the true female population ratio, as it is based on only the households in the sample. attrition between the 1995 and 2006 reservation villages. 11 Broadly similar results are obtained when focusing on children 6-11 months old, though for this age group there is an imbalance on children's age. We will therefore control for 11 Given the wide scope of potential impacts of reservation documented in the literature, the set of variables unlikely to have been affected is limited to demographic and village population characteristics. Note that even if reservation may well affect female education, we include parental education in these balance checks both because of its importance for children's human capital and because average ages of mothers and the extremely low level of education (with 70% of mothers having no education at all) suggest that their education levels observed in 2001 likely reflect 1995 education levels. monthly age fixed effects in all estimations. Note also that while infant mortality is not negligible for the population of interest, there are no significant differences between the 1995 and 2006 reservation villages in the likelihood that the mother has lost a previous child. 12 Hence mortality selection does not seem to be a major concern. Finally, attrition rates within the YLS matched sample of the 1995 and 2006 reservation villages are very low, with only 4% of children 12 On average, 12% of mothers have lost a child, but the difference between the two groups is 20.005 with standard error 0.047.  table 3). The differences are negative and so are unlikely to be a result of reservation policies. Given the importance of parental education for children's human capital (Behrman and Rosenzweig 2005;Carneiro, Meghir, and Parey 2012), this imbalance is a potential concern. It probably results from the relatively small Finally, we note that while the implementation of many policies affecting child outcomes happens at the GP level, budgets for such policies are determined at the mandal level, and separate reservation rules were used for the mandal authorities. We will therefore include mandal fixed effects in all estimations. GPs with 1995 and 2006 reservation status cover 16 mandals, and on average, there are 4.3 GPs per mandal. The number of GPs reserved per mandal, on average, was 1.5 in 1995.

C. Empirical Specification
We estimate simple intent-to-treat regressions, comparing outcomes of children born in GPs with reserved seats for women between 1995 and 2001 with children born in GPs with seats reserved from 2006 to 2011. Specifically, we estimate where Y igm is the outcome of interest for child i in GP g and mandal m. differential impacts of having been exposed to the 1995 reservation as compared to the 2006 reservation. As such they capture the impact of exposure in utero and very early in the child's life, as compared to after the age of 5. We will refer to this as "early" versus "late" exposure to reservation. The vector of child-level controls X includes child's gender and age (using monthly age dummies). 14 Given the lack of baseline, we add only controls that likely reflect 14 There are a few children in the data set younger than 6 months or older than 17 months, despite the fact that sampling was meant to include only children from 6 to 17 months old. As we use agemonth fixed effects, we round the age of 5-month-old children to 6 months and those of more than 17 months down to 17 months. Results are robust to exclusion of these observations. Our main interest is in the estimation of this regression for children aged 12-17 months in 2002, as this is the group of children born before the end of the reservation if they were born in a GP reserved in 1995. We also separately estimate the same specification for children aged 6-11 months in 2002 to look at differential impacts for this younger cohort, for whom exposure due to the timing of the elections was more limited. By using separate estimations for the two cohorts, we allow the mandal fixed effects and other controls to separately control for any factors that may be specific for these age groups. 16 The outcomes of primary interest are the longer-term impacts (in 2009) of reservation on children's test scores. The 8-year-old children were given the Early Grade Reading Assessment test, which measures the basic skills required for literacy acquisition in early years of schooling. It includes recognizing letters of the alphabet, reading simple words, understanding sentences and paragraphs, and listening with comprehension. 17 They were also given a math test with basic computing exercises appropriate for 8-year-old children, as well as the Peabody Picture Vocabulary Test, a popular test for assessing receptive vocabulary, often used as a proxy for cognition. The z-scores of these tests were calculated by subtracting the mean and dividing by the standard deviation. We also calculate the average of these tests to account for multiple inferences. The second main outcome we focus on is the children's nutritional status, in particular, their height-for-age z-scores and stunting levels, which can be tracked through the three rounds of the survey.

III. Long-Term Impacts of Early Exposure to Reservation
Prior to estimating the full model with all the controls and fixed effects, figure 2 first provides the raw evidence for the main outcomes of interest. The graph shows the height-for-age z-scores and the average test scores for all children 15 All results are robust to controlling in addition for the estimated female population ratio. Results are also robust to adding controls for the share of SC or ST. 16 Evidence presented by Brainerd and Menon (2015) indicates, for instance, that religion can be differentially related to nutrition for these two age groups. 17 See the US Agency for International Development's "Early Grade Reading" (https://www .eddataglobal.org/reading/index.cfm). in 2009. The figure clearly shows that the oldest children born in early reservation GPs are doing better than children of the same age born in late reservation GPs. This holds for both height-for-age z-scores and test scores. The difference is positive for children aged 12-13 months or older, and learning differences are the largest for the oldest children. This is consistent with the oldest children benefiting both from in utero exposure and from better earlylife health care. For the youngest children, the differences for height-forage are less clear, while the test score results suggest they might be doing worse in early reservation GPs. Table 4 then shows the main estimates for children 12-17 months old in 2002, comparing outcomes of children born in GPs reserved from 1995 to 2001 with children from GPs reserved from 2006 to 2011. Hence these estimates compare the main cohort of interest as they were exposed in utero and during very early childhood in the early reservation GPs but not in the late reservation GPs. The results show that 8-year-old children born in 2001 in GPs that have been randomly exposed to reservation in the 1995-2001 period have significantly better test outcomes than 8-year-olds who had reached the age of 5 before their GPs were exposed to reservation. The differences are positive for all tests, and on average, children in early reservation GPs have test scores that are 0.19 standard deviations (SD) higher than those of the same cohort of children in late reservation GPs. In addition, children in early reservation GPs have height-for-age z-scores that are 0.26 SD larger, a sizable impact given the low average height-for-age z-scores in the population of 21.60. The probability of stunting is also lower, though not significantly so. 18 Overall, these differences are substantial, in particular given that the 2006 reservation may still have affected the children in late reservation GPs. Table 5 shows that these outcomes reflect nutritional and health gains made earlier in the same children's lives. 19 Indeed, in 2007 (when the children were 5-6 years old) and in 2002 (when they were 12-17 months old) they already had higher height-for-age z-scores and a significantly lower probability of stunting. The impact on stunting is large, indicating a reduction of 12 percentage points (of a mean of about 40) for children born in early reservation GPs. The same children early on also showed a significantly higher probability of having received the major vaccinations by 2002 and hence were better protected against childhood diseases from very early on. Note also that the GPs with 2006 reservation had not yet had any reservation by the time of the 2002 Note. Standard errors (in parentheses) are clustered at the GP level. All estimations include only children in GPs who were reserved either in 1995 or in 2006. Controls include child's gender and monthly age dummies, dummy indicating whether child is first-born, mother's age, mother's ethnicity (dummies for SC, ST, and BC), mother's highest educational grade completed and dummy for no schooling, average adult females in household in GP, and mandal fixed effects. * p < .1. ** p < .05. *** p < .01.
survey. This allows us to interpret results for 2002 as the absolute short-term impact of being born in a GP exposed to reservation for the last 6 years. Overall, the results in tables 4 and 5 hence suggest that reservation in 1995-2001 led to an early advantage in health and nutrition status of children born toward the end of this period and that this advantage was largely maintained from 2002 to 2007 and 2009. When these children reach the age of 8 (in 2009), they also show substantially better test scores, consistent with better health and nutrition in early childhood leading to cognitive gains. These gains are still observed in 2009, despite the fact that in the early reservation GPs (with reservation in 1995-2001), reservation had not been in place already for 8 years, while reservation in fact was in place at that point in the late reservation GPs (with reservation in 2006-11).
To further understand these findings, table 6 shows a number of schooling outcomes. Previous literature on female reservation has indicated that female leaders often invest in schooling or school meals, and GPs with reservation in place in 2006 might therefore, a priori, be expected to have a schooling advantage. Table 6 indeed provides some indications of such advantages, as the children in the early GPs are significantly less likely to have school meals and are significantly more likely to have dropped out of school. They also appear to be in a lower grade (though not significantly so). This helps us to further interpret the test score results in table 4. Children exposed to early reservation have Note. Standard errors (in parentheses) are clustered at the GP level. All estimations include only children in GPs who were reserved either in 1995 or in 2006. Controls include child's gender and monthly age dummies, dummy indicating whether child is first-born, mother's age, mother's ethnicity (dummies for SC, ST, and BC), mother's highest educational grade completed and dummy for no schooling, average adult females in household in GP, and mandal fixed effects. The last column sums up vaccinations of BCG, measles, and polio. There are 20 observations with missing values for anthropometric outcomes because of missing measurements and trimming of values below 25 SD and above 15 SD. * p < .1. ** p < .05.
higher test scores despite apparent disadvantages in a number of schooling indicators in 2009. 20 Children's own perceptions of their performance are also significantly higher for the early than for the late reservation group, consistent with their higher test scores. Overall, the results suggest that reservation policies may have the highest potential for human capital accumulation when affecting the life circumstances of children very early in their young lives.
To put the magnitude of the test score results in context, table 7 shows the coefficients of a set of bivariate regressions of test scores on schooling indicators. The effect on test scores of 0.19 SD amounts to about half a grade of schooling of the child and to about 2.5 years of schooling of the mother. Given that this is a differential result, this suggests a sizable long-term impact of being exposed to reservation in early childhood. We can also compare the magnitude with other evidence on early childhood exposure to better nutrition or health environments. The differential impact of 0.19 SD is slightly higher than the 0.15 SD differential impact found in the study by Barham et al. (2013), who compare early versus late exposure to a 3-year conditional cash transfer pro-20 Note, however, that reservation policies in early reservation GPs may well have increased educational quality during the years of reservation and beyond in ways that are not captured in the data. Children born in GPs with earlier reservations could then have benefited for a longer period of such higher quality. When they entered the schooling system, the quality improvements may already have materialized, while for children born in 2006 reservation GPs, the quality improvements would not have been there when they entered. Note. Standard errors (in parentheses) are clustered at the GP level. All estimations include only children in GPs that were reserved either in 1995 or in 2006. Controls include child's gender and monthly age dummies, dummy indicating whether child is first-born, mother's age, mother's ethnicity (dummies for SC, ST, and BC), mother's highest educational grade completed and dummy for no schooling, average adult females in the household in the GP, and mandal fixed effects. ** p < .05. *** p < .01.
gram in Nicaragua after 10 years. On the other hand, it is smaller than typical impacts of intensive interventions combining nutrition, health care, and stimulation in other developing countries that can show short-term impacts of more than 1 SD (Engle et al. 2007). Evidence on impacts of such interventions when children are in primary school is scarcer. A potentially interesting comparison is, however, the well-known Jamaica early childhood home visiting program (Grantham-McGregor et al. 1997). This intensive intervention combining nutrition and stimulation in early childhood showed very small gains at age 7, even if large impacts were found both immediately after the 2-year intervention (when children were about 4 years old) and again when children were adults (0.6 SD; see Walker et al. 2011). If this means that cognitive gains are hard to observe for this age group, it could imply that our estimates provide a lower bound for the longer-term impacts of female reservation.

IV. Impacts on Other Cohorts
To further understand the relationship between exposure during early childhood and long-term learning and nutritional gains, table 8 shows results for the cohort of children that was born right after the 1995-2001 reservation, that is, the children aged 6-11 months in the fall of 2002. For this cohort, and in contrast to those children exposed in utero, we find no long-term differential effects on either average test scores or height-for-age z-scores. While the results of the two cohorts are not statistically different from each other, the contrast nevertheless is striking. Table 9 further shows that the short-term results on nutrition and health indicators are also not comparable with the older cohort. There are, in particular, no significant impacts on vaccination or stunting levels. This suggests that the policies causing short-term impacts of reservation on children's health in utero and during very early childhood did not persist after reservation was no longer in place.
The results in table 10 are further in line with these findings. Table 10 shows the results for the comparison of the children born in GPs reserved from 2001 to   Note. Standard errors (in parentheses) are clustered at the GP level. All estimations include only children in GPs that were reserved either in 1995 or in 2006. Controls include child's gender and monthly age dummies, dummy indicating whether child is first-born, mother's age, mother's ethnicity (dummies for SC, ST, and BC), mother's highest educational grade completed and dummy for no schooling, average adult females in household in GP, and mandal fixed effects. The last column sums up vaccinations of BCG, measles, and polio. Observations with missing values for anthropometric outcomes are due to missing measurements and trimming of values below 25 SD and above 15 SD. * p < .1.
reservation in the 2001 GPs, and potentially having benefited from reservation in their GPs the next 5 years, have significantly higher test scores and nutritional status compared to those in the GPs with reservations in 2006. As indicated earlier, these comparisons should be interpreted with caution, given the evidence of imbalance presented in table 3 and the overall low number of GPs reserved in 2001. While all estimates control for the variables that were found not to be balanced, it is possible that other unobservable characteristics are not balanced. But even if the evidence is less definite, the pattern in data further supports the main findings of this article, which indicate that having been exposed to reservation very early in life can lead to long-term learning and nutritional gains.

V. Conclusion
This article shows long-term impacts of reservation of local political seats for women on learning and nutritional outcomes in rural Andhra Pradesh. It re- Note. Standard errors (in parentheses) are clustered at the GP level. All estimations include only children in GPs that were reserved either in 2001 or in 2006. Controls include child's gender and monthly age dummies, dummy indicating whether child is first-born, mother's age, mother's ethnicity (dummies for SC, ST, and BC), mother's highest educational grade completed and dummy for no schooling, average adult females in household in GP, and mandal fixed effects. * p < .1. ** p < .05.
lies on three rounds of the Young Lives panel data set to analyze the impact of exposure to political reservation during different stages of young children's life.
We are able to attribute causality to the reservation policy in the Panchayat as the leadership positions to be reserved were randomly allocated through a rotating mechanism. Using this exogenous variation, we show that reservation policies can be particularly effective in increasing learning outcomes when they occur very early during a child's life. The results further show that political reservation improved the health and nutritional environment in utero and during the first years of life, which can help explain the long-term impacts on children's learning. The data used in this article do not allow analyzing in more detail which specific policy decisions resulting from the female reservation are driving the long-term impacts. The available evidence in the literature suggests, moreover, that policies favored by female leaders differ across GPs on the basis of heterogeneity of female preferences and needs, so that different policies might have been affected in different places. 21 The results in this article reveal that, on average, these policy choices led to better health status of children in 2002, which continued in 2007, and translated in higher learning by 2009. While the article does not investigate the causal pathway, we note that these results are at least consistent with female leaders prioritizing key constraints for children's human capital, constraints that may well differ from GP to GP. The results in this article further suggest that any potential advantages of being exposed to reservation-related policies affecting schooling during primary school years did not offset the early nutrition and health gains. 22 These findings go beyond the existing evidence on reservation policies by shedding light on their potential longer-term impacts. They suggest in particular that empowering female leaders might be especially important for improving the health and nutrition of very young children and that this in turn can lead to substantial longer-term impacts. The evidence also suggests that 21 As the first round of YLS data was collected only in 2002, it is not well suited to analyze specific policy choices during the early reservation period from 1995 to 2001. Analysis of a wide range of possible intermediate outcomes for the later reservation periods does not provide support for one particular mechanism. Similarly, data on village infrastructure changes in the different periods do not reveal any robust changes, consistent with reservation leading to different investments in different GPs. 22 An alternative interpretation of the findings could be the potential decrease in effectiveness of reservation policies to affect policy choices. When the reservation first came into place in 1995, it might have had a shock effect, and the impact of reservation might have diminished in subsequent cycles in 2001 and 2006. The reason could be that other political actors might have been able to respond to the initial shock of reservation and possibly made it ineffective in affecting policy choices by the time the third group of GPs benefited from reservation. gains from interventions early in a child's life can be observed many years later, indicating that evaluations of policies targeting early childhood may benefit from a sufficiently long horizon. From a broader policy perspective, the results provide support for interventions that improve children's conditions very early in their lives and suggest that when targeting improvements in learning outcomes, it can be worth considering interventions targeting early childhood in addition to policies directly targeting primary school children.