Female Labour Force Participation and Child Education in India: The Effect of the National Rural Employment Guarantee Scheme

We study the impact of India's National Rural Employment Guarantee Scheme (NREGS) on children's educational outcomes via women's labour force participation. Using data from the Young Lives Study and taking advantage of the spatial and temporal variation in the intensity of implementation of the NREGS, we find that greater participation of mothers in the program is associated with better educational outcomes of their children. Father's participation in the NREGS, on the other hand, has a negative effect on children's education. Further, the estimated impact of mother's program participation is over and above any income effect induced by the scheme and is robust to concerns about endogeneity of labour force participation and differences in economic trends between districts. We provide evidence which suggests that the mechanism through which children's educational outcomes improve is empowerment of mothers resulting from better labour market opportunities for females.


Introduction
The World Development Report (2012), focusing on gender equality, finds that women in the poorer regions of the world continue to suffer from disadvantages in the economic sphere. Although, significant progress has been made in reducing gender disparities in health and educational outcomes, economic opportunities continue to be limited for women. The Report underlines the policy priorities of closing gender differences in access to economic opportunities and earnings as well as increasing women"s voices within households as a means to reducing poverty in developing countries. In this paper we study the impact of one such policy initiative in Indiathe National Rural Employment Guarantee Scheme (NREGS) initiated in 2006. While the program"s main objective is to alleviate rural poverty by legally guaranteeing a minimum of 100 days of annual employment to households, it also has the potential to empower rural women through greater access to labour market opportunities.
From a gender perspective, there are two interesting features of this program.
First, the wage rate provided in this program is uniform across gender, and second, it gives priority to female employment and targets at least one third of the beneficiaries to be women. Thus, NREGS not only has the potential to raise female labour force participation rates by bringing employment opportunities almost to their doorsteps, the equal wage rates provided in NREGS program can potentially reduce any gender disparity prevalent in the rural labour markets. We, therefore, hypothesize that the introduction of this program should lead to greater labour force participation of women, either on the extensive or intensive margin or both.
3 A rise in women"s labour force participation can potentially impact individual and household behaviour on several fronts including marriage, fertility, and intra-household distribution of resources. This paper analyzes the effect of the exogenous policy shock of the implementation of the NREGS on children"s well being. Specifically, we explore whether an increase in participation of mothers in NREGS projects affects the educational outcomes of their children differently from that of fathers" participation in the program. If yes, we attempt to understand the mechanism through which this differential effect can be explained.
While an increase in either fathers" or mothers" labour supply could improve their children"s outcomes purely due to an income effect, greater labour force participation of mothers could impact children"s education through two additional channels. First, women (including mothers) are likely to have more alternative uses of their time than menmarket work, household chores and leisure. If children"s time in doing household chores substitutes for mother"s time then an increase in NREGS participation of mothers may lead to a decline in educational attainment of her children. 1 Second, mother"s say in household resource allocation decisions may rise due to her higher earned income.
Research suggests that this is likely to have a positive effect on her children"s schooling.
If an increase in women"s earned income is likely to translate into greater weight being attached to their preferences in resource allocation decisions of the household and mothers prefer to invest more in their children"s health and education (Blumberg 1988;Thomas 1990;Hoddinott and Haddad 1995;Quisumbing and Maluccio 2003), relative to 1 If mother"s and children"s time on household chores are not substitutes and child care services in the market are either unavailable or unaffordable, then it is more likely that children are in school when mothers are at work. If children attend school more regulary due to mothers working, then children"s educational outcomes might improve.
fathers, then we should see an improvement in child outcomes. Therefore, an increase in mother"s decision-making ability within the family can have a positive impact on her children"s welfare (Thomas 1990;Thomas et al. 2002). To sum, the net impact of a change in mother"s participation in the labour force on her children"s schooling depends on which of these two effects dominatesthe substitution effect or the effect of greater bargaining power, holding household income constant. 2 There exists relatively little empirical research on the impact of parental labour supply on children"s time allocation, particularly in a developing country context. Skoufias (1993) shows that an increase in female wages in rural India reduces the time in school significantly for girls only. Similar results were found by Grootaert and Patrinos (1999) in a cross-country study. However, Ilahi (1999) does not find any impact of female wages on children"s time use in Peru.
In contrast to the sparse literature on time allocation effects, there is considerable empirical evidence suggesting that households" resource allocation decisions are made in a "collective" (Chiappori, 1988) or bargaining framework (McElroy and Horney, 1981) where the final allocation usually depends on the bargaining power or weights attached to the preferences of the members of the household. The importance of labour income as a determinant of women"s bargaining power within the household has been highlighted recently by Anderson and Eswaran (2010). Using data from Bangladesh, the authors show that the effect of earned income on female autonomy is far greater than that of unearned income. Also, women who work on the household farm have no more autonomy than those who are housewives, while those who earn independent income 2 We are abstracting from any long term effects of changes in fertility due to increased labour force participation of women since we are looking at these changes over 2 to 3 years only.
5 have considerably greater autonomy. Luke and Munshi (2011) exploit data from tea plantations in South India where women are employed in permanent wage labor, to find that a relative increase in female income has a positive effect on their children"s education. Qian (2008) shows that a change in agricultural pricing policy in post Mao China which increased female labour income increased educational attainment of all children. However, when the policy increased male labour income, educational attainment for girls decreased but had no effect on boys' educational attainment.
Using data from the Young Lives Study (YLS) in the state of Andhra Pradesh in India and taking advantage of the spatial and temporal variation in the intensity of implementation of the NREGS within districts, we find that greater participation of mothers in the program, relative to fathers, is indeed associated with more time spent in school of children within households. We find that this effect on the educational outcome of children is over and above any income effect induced by the NREGS. Moreover, the impact is largely present for the poorest households and limited to the time spent in school by younger children and girls. We also find that for poorer households greater participation of mothers in NREGS has led to higher grade attainment of children. Our findings are robust to concerns about endogeneity of labour force participation and differences in economic trends between districts.
YLS data on household members" say in decision-making and control of income from various sources show that participation in the labour force by mothers significantly increases the probability that they have a say or control over utilization of earnings from those sources. This result, together with the negative impact of father"s days of NREGS work and our finding that girls tend to benefit more from an increase in mother"s 6 participation in the program, suggests that women"s preferences are the primary drivers of the improvements in educational outcomes of her children when her program participation is higher. Hence our results can be explained within the framework of a bargaining model of household resource allocation.
The findings of our study not only inform us about the impact of female labour supply on intra-household outcomes but it also addresses a broader policy question of the effect of public programmes on improving household outcomes in developing countries.
Specifically, our paper extends the current debate in India on the impact of NREGS on poverty (Ravi and Engler, 2009;Uppal, 2009) and finds evidence, albeit through the channel of women"s program participation, that supports preliminary findings of positive benefits of NREGS on households.
The paper is organized as follows. Section 2 gives the background on the National Rural Employment Guarantee Scheme and motivates the study. Section 3 describes the data and methodology used in this paper. Section 4 discusses the results and Section 5 concludes.

Background
The National Rural Employment Guarantee Act (2005)  Using data on NREGS participation of individual household members and comparing 2007 and 2009-10, we find that the overall female labour force participation in the age group of 16 to 60 years has increased substantially from 59 to 72 per cent while the same for males has fallen marginally from 92 to 89 per cent ( Figure 1). This rise in female labour force participation is largely driven by casual labour (public and private). Figure 2 shows that in 2007, 28 per cent of the females in this age group were involved in casual wage labour, but in 2009-10, this number has risen to 45 per cent. However, unlike females, participation of males in the casual labour market has not increased in this period; rather, it has remained almost the same. Further disaggregating the labour force participation rates across asset quartiles of the households, from With respect to participation in NREGS, the annual average number of days a household worked in NREGS has increased from around 11 days to 40 days ( Figure 3).  (Table 2).
Besides, the rise in female participation in terms of number of days worked in NREGS is also more noticeable in the poorer households.
The data from the YLS establish that casual labour force participation as well as NREGS participation has increased substantially more for women than men between We investigate next whether there has been any accompanying increase in labour force participation rates for 16-60 year old men and women in public casual labour. 8 From Figure 5 we see that there has been a drastic rise in labour force participation in public works, both for men and women. While participation somewhat declined between 6 In 2009-10, large parts of India suffered from drought and it can be contended that this weather shock would have affected labour outcomes in the rural economy. As a robustness check, therefore, we also looked at NSS data for 2007-08. Most results reported in this section go through. We report results with 2009-10 data since by then NREGS had been universally implemented in India. 7 For the country as a whole, the wage ratio, which rose from 1.1 in 1999-00 to 1.7 in 2004-05, has dropped back to 1.1 in 2009-10, thus exhibiting a potential declining trend post 2004. 8 We take into account casual labour done both as the principal occupation as well as a subsidiary occupation.
1999-00 and 2004-05 it has risen sharply to 6.41 per cent for men and 6.49 per cent for women in 2009-10 from almost no participation in 2004-05. 9 The analysis for private casual labour force participation does not exhibit the trends we observe for public casual labour and is more or less flat for both men and women although wage rates for casual private works have increased for both genders, possibly due to substitution of labour from private to public works (Imbert and Papp, 2011). These results are reported in the Appendix. To summarize, our analyses indicate that the trends we observe in casual labour force participation of women in the YLS may have been driven primarily by public works or specifically, the NREGS.

A. Data
In order to identify the effect of the NREGS on children"s education via their mothers" participation in the labour force, we conduct our empirical analysis at the level of the child using the two comparable waves of the YLS surveys -2007 and 2009-10. The panel data set we construct is restricted to children in the age group of 5 to 14 years in 2007, the school going age group. In order to construct our data set we use the following exclusion rules: first, we include only children living in rural areas in both periods. This rules out children who may have migrated to urban areas. However these form less than 1 per cent of our sample. Second, we exclude children for whom we cannot identify mothers in the sample (5 per cent of the original sample). Third, for econometric reasons explained 11 below, we restrict our attention to children present in both rounds of the survey; we thus drop 2.9 per cent of the children present in 2007. Finally, we exclude children for whom there is some missing information on relevant covariates in either of the years. Our data set, after these exclusions, contains information on 2893 children for both years. Table 3  doubled during this period, its contribution to total income was minimal in both years. 11 Table 3 also suggests that there is a rise in wealth over the period of our study. Using asset quartiles, we find that a larger proportion of children live in households in the 3rd and 4th quartile in 2009-10 as compared to 2007. 12 While this change is large, it is consistent with the rise in real incomes noted above. It is also important to note that the household size has remained more of less unchanged during this period.
While preliminary evidence presented above suggests that mother's NREGS participation and number of days of work have gone up, it would be incorrect to draw a causal link between that and changes in children"s time spent in school since decisions regarding labour supply of household members are endogenous. However, the 11 The reference period for non agriculture income is slightly different from agriculture income. While information on non agriculture income was collected over a reference of last 12 months, information on agriculture income was collected over the last agricultural year. Since the two periods have substantial overlap, we add the two to calculate total income. 12 We use the pooled sample of 2007 and 2009-10 to generate the asset index. Pooling over the years ensures that the asset quartile changes over years reflect absolute changes in wealth rather than a change in the relative position in the wealth distribution within any year.
13 introduction of the NREGS also leads to exogenous shifts in the demand for labour. One such measure of demand is the total number of "in-progress" NREGS projects in a mandal (or sub-district). This is unlikely to be determined by factors specific to the labour supply decision of any individual household or even a village. However, progress on a larger number of projects within a mandal indicates that there are relatively more work opportunities for households residing in that area. 13 The last row in Table 3 suggests that the number of such NREGS projects did increase during the period of our study.
Moreover, this increase was not uniform across mandals. Hence the change in demand for labour for NREGS projects varied both over time and across mandals in Andhra Pradesh.

B. Methodology
In this section, we specify our empirical model and discuss the estimation strategy we adopt to test our hypothesis.
To begin with, note that NREGS participation by household members can have two distinct effects on children"s time spent in school (TSS). 14 First, as household members work on NREGS, the total income of the household (INC) may rise. 15 In so far 13 It is important here to point out that an individual's decision to not work on NREGS project will neither affect the progress of the project (unless no person in the mandal wants to work) nor will it affect the total number of projects in a mandal on which progress has been made. 14 The time spent in school is recorded as hours spent in school on a typical day in the previous week. The total time spent on education on a typical day consists of time spent in school and time spent on studying outside school (private tuition and at home). The average time spent on education outside the school in the sample is less than 20 per cent of the total time spent on education on a typical day. 15 NREGS work is only one of many activities performed by members of the household. If NREGS work done by a member does not cause changes in the total labour supplied to other household activities, then total income will rise by the additional income from NREGS. However, the possibility of doing work as a part of NREGS is expected to change the allocation of labour to various activities. If households reallocate work rationally, then total income will tend (contd...) as households consider the education of children to be a normal good, this income effect may result in a rise in children"s time spent in school. 16 Second, NREGS could have an additional direct impact on children"s education due to greater labour force participation of mothers, but in two opposite directions: a negative substitution effect and a positive bargaining power effect. Thus, the net impact of a change in mother"s participation in the labour force on her children"s schooling is an empirical question. We posit that, controlling for INC and the number of days worked on NREGS by the father (FATHER_NREGS), a positive effect of the number of NREGS days by the mother (MOTHER_NREGS) on children"s educational outcomes would suggest that the latter effect dominates the negative substitution effect. 17 There are several child and household level factors that could confound the effects of NREGS participation of mothers on her children"s time spent in school. We, therefore, include Z -a vector of household variables that may change over time, in our empirical specification. For instance, households often smooth any income shocks with dissaving of assets. Hence we take into account household wealth represented by Asset Quartiles ( ). Further, since poorer households are more likely to work on NREGS, omitting asset ownership would confound the effects of parental labour supply on TSS. We also include land owned by the household as an indicator of wealth. Households" optimization to increase. However, in this case, the total income need not rise by just the amount earned through NREGS work. 16 Whether the income effect is significant or not is a function of the cost of schooling as well. If physical access to schooling is relatively easy and costs of schooling are subsidized (as in primary schooling), any effect of an increase in household income may be muted for the age group under study here. 17 As explained earlier, the number of days of NREGS work done by mothers and fathers have gone up over the period of the study. This is true not only for the districts where NREGS was implemented in 2008 but also for districts where it was implemented in 2006.

D i
ASSETS process is also a function of the size of the household and migration of members can cause substantial changes in the decisions taken by households. Hence we include the number of household members (HH_SIZE) in Z as well. Our analysis also controls for whether the household" reference week was a school holiday. However, the results hold up when we restrict our sample to only those children for whom the previous week was not a school holiday.
Next, we include X, which denotes the vector of child specific time variant variables that could affect TSS. As a part of X, we include the highest grade achieved by the child. This is to account for the fact that the time spent in school may systematically vary with grade. We include the square of GRADE (SQ_GRADE) to allow for non linearity in this effect. 18 Moreover, the time children can spend in school may depend on their age. Older children are more likely to spend time working outside or looking after their siblings. We allow for this effect to be non-linear in age by including age (AGE) and square of age (SQ_AGE) in X.
While the variables included in Z and X are observable to the econometrician, there may be unobservables at the geographic level (district, mandal and village), household level and there may also exist child specific unobserved heterogeneity. If these unobservables are correlated with the regressors on the right hand-side and they also affect time spent in school, it would lead to the issue of endogeneity and thereby inconsistency of our estimates. Our specification, therefore, includes time invariant child characteristics viz. ability ( ℎ ), household characteristics viz. parental preferences 16 for schooling ( ℎ ) , mandal level characteristics ( ) , and village characteristics viz. culture( ).
A potential problem for our empirical exercise is the phased implementation of NREGS. Districts that had NREGS earlier (Phase 1) may be different from those that had NREGS later (Phase 2). Moreover, these districts may have different economic growth trajectories as well as trends in educational attainment. To take into account these concerns, we allow for district specific intercepts and introduce district specific time trends We also control for a secular time trend , that allows for increases in demand for and supply of schooling. 19 In addition to district specific trends, there could be trends that are driven by rising awareness of rights due to social audits. For example, social audits that make households aware of their rights may also lead to a demand for public schools. Hence NREGS participation and children"s time in school could be driven by this rising awareness. To control for this, we allow the trend to depend on the number of social audits that have taken place in the mandal prior to the date of the survey ( _ * ).
More formally, we estimate the following specification: where the subscript refers to a child in household in village in mandal m in district . refers to time, which takes the value 0 for the year 2007 and 1 for the period  This period is also one of renewed efforts to universalize primary education, through district level government policies (viz., Sarva Shiksha Abhiyan or SSA). Given this specification, and using data on a balanced panel of children over the two time periods, we estimate a child fixed effects model. In doing so, we eliminate ℎ , ℎ , and as well as . The child fixed effects model uses the temporal variation for each child and demeans variables from the child level mean.
This eliminates all observable variables that are time invariant; therefore such child (e.g. the gender of the child) and household level variables (e.g. education of parents that we may have possibly considered) are eliminated. If we assume that the deviation of the observed variables from their mean values are not correlated with the deviation of the error term from its mean values, this estimation procedure would yield consistent estimators of 1, 2 and 3 .
The main concern with our estimation strategy is that parental labour supply decisions are likely to be made simultaneously with investments in children"s education.
Thus, father"s and mother"s participation in the NREGS may remain endogenous even though we account for unobserved, time invariant household and child level heterogeneity. To address this simultaneity issue, we adopt a 2SLS estimation procedure using mandal (sub-district) level rainfall shocks in the month of May and June and variation in the demand for NREGS labour as instruments. We define a rainfall shock MOTHER_NREGS) and three instruments (RAIN, PROJECTS, RAIN*PROJECTS), our estimating equation is just identified. We discuss the validity of our instruments next.

C. Validity of instruments
Agricultural production in India continues to be dependent upon rainfall. The choice of rainfall in May-June of the reference period as an instrument is, thus, driven by the nature of agricultural activity in the region of our study. Rice is the main crop cultivated in Andhra Pradesh. Using the YLS data we find that among rural households, the crop which the largest proportion of households cultivate (almost 36 per cent across rounds 2 and 3) is rice. 22 The cultivation of rice is highly water-intensive. The crop is mainly cultivated in flooded, standing water fields. But prior to cultivation in the paddy fields, the rice seedlings (which cannot survive in flooded fields) are grown in nurseries. They are then manually transplanted into the flooded fields. It is therefore expected that rainfall in the pre-monsoon season will promote the development of rice seedlings enabling 20 The variable capturing rainfall shocks (RAIN) is constructed from the precipitation data available from the Center of Climatic Research at the University of Delaware. The data includes monthly precipitation values at 0.5 degree intervals in latitude and longitude. To match this data at the mandal level, the nearest latitude-longitude to each mandal headquarter is taken. To construct the rainfall shock at the mandal level, the long term (1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008) average mandal level rainfall in the months of May and June are estimated. Standard deviation of rainfall for the same period is also calculated at the mandal level. Then rainfall shock is defined as the deviation of actual rainfall in the months of May and June in the last year from the long term average, divided by the standard deviation. 21 Data on the number of ongoing projects at the mandal level is obtained from the Andhra Pradesh Government"s website on NREGS (http://nrega.ap.gov.in/). 22 Groundnuts is a distant second, with about 16% of rural households engaged in its cultivation.
19 farmers to increase their cultivation of rice during the monsoon season. This in turn would create greater demand for labour for transplanting. Majority of the transplanting work is done by women because it is delicate work and is a highly labour-intensive activity (Mies, 1986). 23 Our assumption, therefore, is that, ceteris paribus, demand for female labour for agricultural work will be higher if premonsoon rainfall is high.
The monsoon typically arrives in Andhra Pradesh in mid June. Hence the premonsoon rainfall falls mostly in the month of May and early June. Furthermore, schools are closed for summer vacations from the last week of April to mid June every year in Andhra Pradesh. 24 Thus it is unlikely that rainfall in this period will have a direct effect on time spent in school or grade attainment either due to households" labour substitution decisions or supply-side factors such as teacher attendance.
Our second instrument is the number of NREGS projects currently in progress in the mandal. 25 Before we discuss the validity of this instrument it is imperative to discuss the administration of the NREGA projects in general and in AP in particular. There are three tiers of administration of NREGA projects in Andhra Pradesh -district, mandal and village. While the district administration is responsible for overseeing the implementation of the NREGA program, the Mandal Parishad Development Office (MPDO) is the main agency for administering each NREGA project and sanctioning all financial payments for projects undertaken in that mandal. The village council"s role is limited to recommending the list, and priority, of NREGA projects to be implemented to the MPDO. Since the NREGA is envisaged as a demand-driven program, households are expected to apply for work and once a critical mass of demand is generated in a gram panchayat (a collection of 1 to 3 villages) in a mandal, a project has to be selected from the approved list of works and sanctioned by the district administration. Thus the main concern with the IV"s validity is that current, individual demand may determine the number of projects in progress, partly or fully, at the mandal level.
Note that our instrument is defined at the level of the mandal -a collection of 11 to 39 gram panchayats (in the YLS sample) -and the sanctioning of projects is at the district and mandal level. Furthermore, although the NREGA envisages a demand driven programme, the reality is quite different according to several recent studies. Imbert and Papp (2012) report that "many people are unaware of their full set of rights under the programme"; "in practice, very few job card holders formally apply for work while the majority tend to wait passively for work to be provided." Other research on NREGS districts in Andhra Pradesh (Ravi and Engler, 2009;Afridi et al., 2012) also indicates that the programme is supply rather than demand driven. 26 Hence, given the fact that the program is driven by the supply of projects at the district and mandal level and that our intrument is defined at the level of the mandal, it is unlikely that there are significant effects of current household demand on program intensity at the mandal level.

21
The concern that remains then is whether temporal changes in awareness of NREGA entitlements (including demanding work; Khera, 2011) are accompanied with changes in the demand for public schooling (quality or quantity). 27 On the other hand, say there is no increase in awareness but the administration is learning how to implement NREGS, which improves between 2006-2009 and this learning spills over to the provision of the public good of interest to useducation. In either case, our IV will not meet the exclusion restriction as it would have a direct effect on educational outcomes.
We address the latter concern first. In Andhra Pradesh, school participation is near universal. 28 According to the Annual Survey of Education Report (ASER, 2006), the percentage of out of school rural children in the 6-14 age group was between 0 to 5 per cent in all the YLS districts except West Godavari where it was between 5 to 10 per cent in 2006. Learning levels were higher than the average for the country and have remained more or less steady during this period (ASER, 2006 and2009 implementation at the grassroot level and which helps expand capacity for the program is different and delinked from that required for public schooling (see Figure A3 in the appendix for more details have been any significant changes in political will for implementation of public programmes either. To address the former concern, we first use data from the YLS to check whether political participation or participation in community led demand for certain public goods was correlated with the occurrence of social audits. The timing, frequency and conduct of social audits in a mandal is determined centrally by an independent bodythe Society for Social Audit, Accountability and Transparency (SSAAT)in Andhra Pradesh.
Hence, the number of audits conducted in a mandal should be exogenous to the village and household. We find an insignificant effect of the occurrence of social audits on awareness between the second and third round of the YLS surveys in a household fixed effects model (see Table A2 in the appendix). Nevertheless, as discussed in the empirical model above, we include a variable "number of social audits that took place in the mandal between the two survey rounds" in all our baseline regression analyses to control for any direct effect of "awareness" improvements on children"s schooling. Our results are robust to the inclusion of this variable.

A. Overall impact on children's time in school
Our hypothesis is that once we control for households assets, total income and other child, household village, mandal and district level characteristics, the coefficients of MOTHER_NREGS would be positive. Results in Table 4 show that this is indeed the case for time spent in school. Column 1 reports the results of an OLS-FE regression of child"s time spent in school on parental participation in NREGS, accounting for unobservable heterogeneity in child characteristics and differences in trends across districts. We find that the coefficient of MOTHER_NREGS and FATHER_NREGS are positive but insignificant. However, as pointed out above, this specification does not account for the possible endogeneity of labour force participation of parents and household income. We, therefore, look at the 2SLS results in column 2. 29 Instrumenting for the three endogenous variables in column 2, we find that the coefficient on MOTHER_NREGS is positive and significant at 1 per cent. This result validates our hypothesis given that the change in days of NREGS work by mothers 29 The first stage results (Table A1 in the appendix) suggest that our instruments are good predictors of the endogenous variables. First stage regression equations have F statistics ranging from 42.55 to 71.47. Moreover, at least one instrument is significant in each regression equation making their use valid. The first stage results suggest that an increase in the number of on-going NREGS projects in a mandal increases the household income and parents" participation in such projects, as hypothesized. The coefficient on RAIN turns out to be insignificant for female NREGS participation while significant for male participation, where as the coefficient for RAIN*PROJECTS is positive and signficant and more or less similar for both female and male participation in NREGS. These results point out that the response of females to NREGS is lesser than that of males when the rainfall in May and June is good. This is consistent with the fact that, in the years of good pre-monsoon rains, females are more likely to work in rice fields than men. The coefficient of RAIN is negative though insignificant for annual household income. This insignificance could be because agriculture income is a very small part of total annual household income. Moreover, here we use rainfall in only May and June as an instrument. However, a good rain shock for the summer crop may well be followed by a bad rain shock during the winter crop, resulting in an insignificant (or even negative) effect of May-June rainfall on total annual agricultural income.
24 between the two survey rounds represents a substantial increase in the proportion of household earnings attributable to mothers. Note that since mother"s NREGS participation is likely to have two opposing effects on children"s educational outcomes, as discussed in the previous section, a positive coefficient on MOTHER_NREGS indicates that the bargaining effect dominates any labour substitution effect. Thus, in so far as our estimated coefficient is a consequence of higher bargaining power of mothers, it may be a lower bound on the impact of the bargaining effects of NREGS on child outcomes.
In contrast, the effect of FATHER_NREGS is negative and significant at 1 percent significance level in column 2. Disaggregation of this effect, discussed in the next section, shows that this effect is greater for poorer households and for older boys, 30 We recognize that any additional time spent in school could be substituted by less time spent studying outside school leading to an insignificant effect of mother"s NREGS work days on total time spent on education on a typical day. In an alternate specification, therefore, we consider the total time spent on education (including time spent studying outside the school) as the dependent variable. Our results are unchanged. potentially indicating substitution of fathers" time on non-NREGS work with that of children"s time. 31 In so far as NREGS income is a part of total income, any NREGS work by parents may lead to a rise in the time spent in schooling. In column 1 we find that the coefficient on total household income is insignificant. However the point estimate is positive and the p value is 0.17. Once we account for the endogeneity of household income in column 2, we get a similar positive but much larger point estimate for income.
However, the p value is still around 0.17. This suggests that, even if there are income effects, there may be too much heterogeneity in this effect, so as to render the coefficient insignificant. The insignificant coefficients for asset quartiles echo the same argument. In rural areas, land is usually a good predictor of wealth. We find that, in the OLS-FE specification, the effect of land is positive and significant. However, this result turns insignificant when we look at the 2SLS-FE results in column 2. In a later part of the paper we will examine this issue more by stratifying households by their base wealth levels.
As pointed out, children"s time spent in school and parental NREGS participation may covary because of increasing awareness, through social audits. While the OLS-FE results estimate this effect to be negative, the 2SLS-FE results find this effect to be positive but insignificant. Recall that the variable, "number of NREGS social audits in the mandal between the two survey rounds x time" allows for different trends in time 31 The negative coefficient on father"s NREGS work is also consistent with our result on mother"s NREGS work. For a given level of income and mother"s NREGS participation, a rise in father"s NREGS participation might imply that a larger share of household income is earned by the father. If this larger share of income bestows greater bargaining power to fathers relative to mothers, and fathers prefer to invest resources in goods other than children"s education, then the time spent in school by children could fall. spent in school depending on the number of audits that have taken place in the mandal before the survey. Our results point out, that if anything, the change in time spent in schooling is lesser in mandals with more social audits.
The coefficient on household size is negative but insignificant in both columns.
The coefficient on time is positive and significant in column 1 but is insignificant in column 2. In both cases, the point estimates are large representing the effect of increasing age of the child over time. While the child"s age drops out as it is collinear with time, we find that there is a non-linear effect of age. The square of age turns out to be negative in columns 1 and 2. The greater the age, the lower the increase in time spent in school. This reflects the higher opportunity cost of time in school for older children. The coefficient of the term linear in GRADE is negative and significant while that on the quadratic term is positive. The argument for GRADE is similar to that of age (and may indeed be picking up some of the effect). As children progress to higher grades, they are more likely to drop out of school, leading to a fall in time spent in school in the current year. However, for those who continue to attend school, as they move into higher grades, the time spent in school goes up.

B. Heterogeneity of impact on children's time in school
The reported average effect of NREGS participation by mothers may hide large heterogeneity of impact across households belonging to different wealth strata. To address some of these issues related to heterogeneity of effects we run our regressions by two indicators of household wealth -asset quartiles and land ownership. We construct four sub-samples of children who belong to households in each of the four asset quartiles 27 in 2007 (recall that the quartiles are still based on the pooled sample of 2007 and 2009-10). Doing so not only allows us to look at heterogeneous impacts across wealth levels but also addresses the issue of transition in wealth quartiles. In addition, we classify households into those whose land ownership was less than the median land ownership and more than the median land ownership (based on 2007 distribution of land in the sample).
The results in Table 5

29
The days of NREGS work by the father has no impact on the female children.
However it is negative (-0.139) and significant for the male children. 32 We also find that in the case of male children, there is a positive secular trend (3.412) indicating a general rise in schooling for male children over time.
Columns 3 and 4 further disaggregate the effect of NREGS work by parents by the age of the child. We divide the sample of children into two groups: those who were in the age group 5-9 years in 2007 and those who were 10-14 years old. Results show that the effects indeed vary across the two age groups. The days worked by the mother in NREGS is positive and significant (0.225) for the younger age cohort. The rise in days of work by the mother (4.4 in 2007 to 20.4 in 2009-10 for this sub-sample) lead to a 3.6 hour increase in time spent in school for this younger cohort. The number of days father worked on NREGS is negative and significant for both the age groups but it is larger for the 10-14 age group (-0.276 for 10-14 year age groups relative to -0.168 for the 5-9 year age group).
Therefore, the overall impact of parental work on NREGS is greater for the younger age group than for the older age group. There could be two reasons for this.
First, since the opportunity cost of being in school goes up for higher age groups, mother"s NREGS income or greater bargaining power may not be able to fully compensate for these higher costs. Second, as discussed earlier, the substitution effect of 32 Our results are in keeping with the findings of existing research on the impact of parental resources on children"s outcomes. Previous literature suggests that the impact of mother"s influence on household decision-making may differ by the gender of the child (Thomas 1990;Murthi et al. 1995) but the literature is not conclusive on whether it exacerbates or reduces gender differences. For instance, Thomas (1990) finds that in Brazil women"s education has a significantly stronger effect on girls" health while educated fathers prefer to invest more in boys. In Java (Thomas et al. 2002) and Cote d"Ivoire (Haddad and Hoddinott 1994), on the other hand, women with greater earned income allocate more resources to sons" health.
mother"s NREGS work may kick in more strongly for older children whose time on household chores is likely to be a closer substitute for mother"s time. This is also apparent from the insignificant effect of mother"s NREGS work on older girls.

C. Impact on children's grade progression
In the previous sections, we have shown that mothers" work on NREGS projects positively affects children"s time spent in school. In this section, we delve into whether an increase in attendance rates in school has translated into higher grade attainment. But To find the determinants of GRADE, we consider a slight modification of the empirical model presented above. We estimate the following specification: The dependent variable is grade attainment of a child (subscripts follow the same convention as in equation 1). The only difference in regressors between this specification and specification 1 is that GRADE is now the dependent variable. Hence we refer here to the modified vector of child level characteristics, which does not include GRADE in ′ ℎ .
We report results of OLS-FE and 2SLS-FE in Table 7. 33 For the overall sample, we find that while MOTHER_NREGS is positive and significant in OLS-FE specification in column 1, it becomes insignificant for 2SLS-FE in column 2. Similarly,

FATHER_NREGS is negative and significant when we run OLS-FE but is insignificant
in the case of 2SLS-FE. Therefore, for the overall sample, we find no effect of NREGS work (by either parent) on grade progression. However, when we stratify the sample by households" land holdings in 2007, we find that MOTHER_NREGS is positive and significant for households with less than median land holdings (columns 3 and 4). The coefficient of 0.031 implies that the increase in the days worked by the mother on NREGS projects, among this sub-sample (3.87 days in 2007 to 22.2 days in 2009-10), has lead to an increase in grade attainment by 0.568, which is more than half an academic year. The days worked by the father has no significant effect on grade attainment. For households with more than median land holdings, there is no effect of either parent"s 32 working on NREGS. This result substantiates what we have also observed for time spent in school -that the effect of days of NREGS work by mother is more visible in the lower economic strata. 34 To sum, our results for grade attainment are muted but consistent with what we observed for time spent in school. It suggests that the days of work by the mother on NREGS, ceteris paribus, has lead to better educational outcomes for her children.

D. Impact of mother's work status on empowerment
Our results suggest that an increase in labour force participation of mothers has beneficial effects on her children but not so of fathers". The time allocation hypothesis indicates that there should be a negative or zero effect of mother"s labour force participation on children"s educational outcomes, particularly girls. While we do find that there is an insignificant effect of mother"s program participation on older children and boys, 5 to 9 year olds and girls tend to benefit. While the former can be explained in terms of nonsubstitutability of younger children"s time with mother"s time on household chores, the latter effect suggests that women"s preference could be coming into play.
In order to further test our hypothesis of improvements in women"s decisionmaking abilities within households as a result of increased participation in the labour market, we use data from the second round of the Young Lives survey. The second round 34 When we classify the sample by children"s age group, we find that the coefficient on MOTHER_NREGS is significant and positive for the age group 5-9 years, whereas there is no significant impact for the older age group of 10-14 years. We do not find any significant impact of parental participation in NREGS when we classify the sample by any other individual or household characteristic (viz. gender or household assets).
of the YLS data contains detailed information on various decision-makers within a household. 35 Our dependent variable is the binary response to the following questions: a."Is the caregiver responsible for making the key decisions about any of the plots?" b. "Does the caregiver control the use of the earnings from the sale of goods or rent from any of these plots?" (Earnings from land) c. "Is the caregiver responsible for making the key decisions about any of these work for wages activities?" (Wage activities) d. "Is the caregiver responsible for controlling the earnings from any of these from work for wages activities?" (Earnings from wage activities) The sample is restricted to caregivers who are mothers in age group 16-60 years. 36 Our main variable of interest is whether the woman works. We expect the woman"s work status to be affected by the number of "in-progress" NREGS projects in the mandal and the long-term deviation of rainfall in May-June of the reference period. We, therefore, instrument work status by these two variables and their interaction. Controls for individual and household characteristics such as -age, education, household"s asset quartiles and household sizeare included within a district fixed effects specification.
The results are reported in Table 8.
The positive and significant coefficient on "working" across all outcomes suggests that greater participation of mothers in the labour market does increase the say and control these women have on important decisions being made within the household. In a rural setting earnings from land and from wages are likely to be the two most important sources of income for households. 37 This result, therefore, bolsters our claim that an increase in work opportunities for women is likely to have a positive effect on their decision-making abilities within the household. The positive impact of mother"s NREGS work on girls" time in school and our analysis here indicates that our findings cannot be explained purely within a unitary framework of the household. 38

Conclusion
The role of increasing women"s bargaining power within households as a means of reducing poverty has been emphasized in discussions on development policy. In this paper, we look at one such policy initiative in India -the National Rural Employment Guarantee Scheme. While the scheme has been conceived primarily to provide households a guaranteed income through employment on public projects, it is sensitive to issues of gender discrimination in the labour marker. Given that private casual wages for women are often less than those of men, the scheme stipulates equal wage rates across gender. It also gives priority to female employment and targets at least one third of the beneficiaries to be women. Thus the scheme aims to increase and improve rural women"s labour market opportunities.
In this paper we contend that, ceteris paribus, an increase in participation of a mother on NREGS projects could affect her household"s outcomes such that they reflect her preferences better. Using panel data collected by the Young Lives Study in a large 37 We find no impact of work status of mothers on their participation in decisions related to earnings from livestock and self-employment activities of the household. 38 In order to test for the possibility that schools substitute for day care for working mothers, we control for the demographic composition of the household. Under this hypothesis, the effect of mothers working on children"s time in school should be insignificant if there are older siblings in the household to take care of the younger ones. But our results are unchanged when we control for demographic composition of the household. See Table A3 in the appendix for details. southern state of India (Andhra Pradesh) and taking advantage of intra district variation in rainfall shock and the number of NREGS projects "in-progress", we find that greater participation of women in NREGS works has a positive effect on her children"s time in school. Moreover we find that this effect is largely on children in the poorest wealth group, for girls and younger children in the household.
Our findings of the positive effect of mothers" program participation on children"s time spent in school carries implications for their educational attainment as well. Our results suggest that grade attainment of younger children and those in less landed households improves due to mothers" NREGS participation, implying that more time in school translates into better educational attainment for some groups of children.