Cash Transfers and Gender Differentials in Child Schooling and Labor: Evidence from the Lesotho Child Grants Programme

Abstract

Family and child allowances constitute about 16 percent of total spending on cash transfers (CTs) worldwide (Honorati, Gentilini, and Yemtsov 2015). These programs often focus on increasing investments in children's human capital, particularly in nutrition and schooling, with the goal of reducing the intergenerational transmission of poverty. Other old-age social pension programs and poverty-targeted CTs have similarly targeted human capital investment objectives. To this end, the impacts of CTs on child welfare outcomes have been widely studied (De Hoop and Rosati 2014), showing overall positive results on schooling and in some cases a reduction in child labor.1 The bulk of such evidence on both conditional and unconditional CTs shows that they have substantial impacts on child enrollment and attendance, particularly in secondary schooling, where attendance tends to be lower in poor households (World Bank 2014). A remaining important question about CTs, both conditional and unconditional, is whether their impacts on human capital investments are equitable between boys and girls vis-à-vis the use of their labor. The bulk of studies available on CTs show no consistency on whether impacts in education benefit girls or boys more. Gender differences in the impacts of CTs on child labor also remain inconclusive. Differences in outcomes by gender, given household access to CT programs, have important implications for gender equality in human capital accumulation and economic growth. Therefore, when designing CT programs to mitigate constraints faced by households in investing in children, public policy must also consider factors leading to unequal investments by parents and caretakers in children based on sex-specific preferences. This article contributes to the literature on child well-being by examining gender-differentiated impacts on child schooling, labor, and time use by comparing impacts on outcomes for boys and girls across married male-headed households (MHHs) and de jure unmarried female-headed households (FHHs), and by the sex of cash transfer recipients. For the empirical analysis, we use impact evaluation data from the Child Grants Programme (CGP) in Lesotho, which consists of a CT provided to poor and vulnerable rural households with children. As in many sub-Saharan African countries, most rural households in Lesotho are engaged in agriculture, and the vast majority of their children are employed in crop and livestock production activities; this engagement is an important determinant of school enrollment and schooling outcomes (Kimane 2006). Rural households tend to rely on family labor and face more constraints when allocating the time that children dedicate to agricultural activities, household chores, and schooling. The context of Lesotho is also characterized by the HIV pandemic, which has affected the structure of households significantly, reducing its adult labor capacity and potentially further constraining children's time in school. Conditional CTs (CCTs) mandate child school attendance (among other requirements) for qualification. There is clear evidence that CCTs, including large programs like Brazil's Bolsa Familia and Mexico's Progresa (Bourguignon, Ferreira, and Leite 2003; Cardoso and Souza 2004; Handa et al. 2009; Skoufias et al. 2001), have positive impacts on children's schooling, especially among students in secondary school (see Martorano and Sanfilippo [2012] for the case of Chile). There is also evidence that social pensions and unconditional CTs (UCTs) improve child schooling. Edmonds (2006) analyzed pensions for the elderly in South Africa, finding significant increases in schooling and declines in labor participation for children, mostly for boys. Examining a monthly UCT for the ultra-poor in Malawi, Miller and Tsoka (2012) found improved education and reduced labor among children in beneficiary households. More recently, Akresh, de Walque, and Kazianga (2013) found increased school attendance rates as a result of participation in a UCT in Burkina Faso, and Handa et al. (2016) found increased school enrollment, particularly among older children, and decreased child wage labor in a UCT in Zambia. A number of studies have suggested that education and labor outcomes are influenced by parental expectations of future labor market outcomes relative to the current opportunity cost of boys’ and girls’ time (World Bank 2014). It is therefore plausible that these factors also influence decisions on how to use CTs, particularly UCTs. In addition, for agricultural households facing nonseparable production and consumption decisions, the impact of CTs on household production—and therefore on labor decisions of both adults and children—are expected to be jointly determined with other outcomes such as schooling investment decisions (Benjamin 1992; Bardhan and Udry 1999; Handa et al. 2010). In Lesotho, child labor is about 23 percent—usually young men involved in the task of livestock rearing. Although no data are available for young children, the Lesotho Demographic and Health Survey (LDHS) shows that in 2009, about 76 percent of boys aged 15–19 participated in agricultural activities, whereas women of the same age-cohort worked in agriculture at a lower rate (36 percent). In relation to household decision making in child investment by sex, child preference also plays a role in the use of CTs in child investments. Since the seminal work of Becker (1965; 1981), economists have built on his theory of choice framework to analyze intrahousehold and intergenerational resource transmission. The findings of Emerson, Souza, and Portela (2002) in Brazil provide strong evidence that parental child preferences may generate a gender bias in child human capital investments. They find that while both father's and mother's schooling had strong impacts on sons’ education and labor, only mother's schooling affected the probability that a daughter works. In addition, nonlabor income (transfers) for either parent had an impact on sons’ school attendance, but not on that of daughters. A strand of the literature investigating the impacts of gender-based program features remains inconclusive on the policy implications. For example, Mexico's Progresa provided larger transfers to households with girls to reduce the gender gap in schooling enrollment (Handa et al., 2009). However, empirical evidence has not confirmed whether the observed larger impacts on girls derived from lower initial enrollment rates for girls or from the higher payments made to them. Various studies have already shown that child welfare is improved when women have control of a greater share of household resources, either through income (Thomas, Strauss, and Henriques 1990; Quisumbing and de la Brière 2000) or dowry (Quisumbing and Maluccio 2003), thus making the case for women to be designated cash recipients. However, there is scarce evidence comparing outcomes by sex of transfer recipient. The little existing research on child sex–differentiated impacts by sex of household recipient has in some cases suggested prevalent gender bias in intrahousehold resource allocation (Duflo 2003; Akresh, de Walque, and Kazianga 2013). More recently, a randomized controlled trial on male and female cash recipients of an education grant in Morocco found that girls had marginally higher schooling outcomes when mothers received the transfer instead of fathers. However, this difference was not observed within a UCT applied in the context of the same experiment (Benhassine et al. 2015). Others studies have made the case for a strong association between cash given to mothers and child schooling, nutrition, and general welfare (Behrman and Hoddinott 2005; Manley, Gitter, and Slavchevska 2012; De Brauw et al. 2014). Most of these studies, though, failed to compare these outcomes to a scenario with male cash recipients. The role of household structure on differences in human capital investments on girls and boys as a result of CTs has not been widely studied either. Constraints derived from lower labor capacity are higher for agricultural households, as they tend to rely on family labor, including that of children. An important question is whether FHHs are more likely to contribute to the intergenerational transmission of poverty, as they face higher constraints for substituting child labor for child education. Empirical evidence on this question and on the social and economic factors that mitigate these poverty dynamics is essential for sub-Saharan Africa, where 26 percent of households are estimated to be headed by a woman and their prevalence has increased since the 1990s (Milazzo and van de Walle 2015), due to its changing population structure,. The extent to which FHHs are disadvantaged relative to MHHs in terms of poverty, labor capacity, access to land and livestock, and lower credit and education varies greatly across studies and contexts (Kossoudji and Mueller 1983; Handa 1996; Quisumbing 1996; Buvinić and Rao Gupta 1997). These factors can also vary between de jure FHHs, which are run by single, widowed, divorced, or separated women, and de facto FHHs, in which a husband is temporarily absent—for instance, because he is working and living abroad. Further, qualitative evidence from Lesotho shows that women enact social and caregiving roles within a gendered family and household context (Harrison, Short, and Tuoane-Nkhasi 2014). In the context of sub-Saharan Africa, the age of the head of the household is very relevant, as (due to the HIV pandemic) FHHs sometimes consist of elderly women caring for their grandchildren. In Lesotho, households face particular constraints caused by the HIV pandemic. Starting in the 1990s, the pandemic reduced life expectancy at birth from 59 years in 1990 to 48 years in 2000 (World Bank n.d.), and life expectancy has not yet recovered to pre-1990 levels. According to the LDHS, in 2009, about 23 percent of adults were infected with HIV; in the same year, the proportion of rural households with foster and orphan children reached 47 percent. It is common for grandmothers to take charge of orphan children from their kin, also bringing additional foster children into their care. Lesotho's CGP, the CT analyzed in this article, aims to help households (and children in particular) who have been hit the most by these circumstances. About half of our sample of CGP households are headed by a woman. The CGP in Lesotho is a UCT that targets poor rural households with orphans and vulnerable children. Its primary objective is to improve the living standards of such children—to reduce malnutrition, improve their health status, and increase their schooling. At the beginning of the program in 2009, the transfer value was set at a flat rate of LSL120 (US$12) per month per household and was disbursed every quarter. This amount corresponded to around 19 percent of the median consumption of an eligible household. Since April 2013, the size of the transfer was increased and indexed to the number of children, ranging from LSL120 to LSL250 (US$25) per month. Program beneficiaries are selected through a combination of proxy means testing and community validation and are registered in the National Information System for Social Assistance (NISSA) (Pellerano et al. 2014). As of December 2017, 26,600 households were benefitting from the CGP, making it the second largest social protection intervention of the Government of Lesotho after the old-age pension. Phase 1, Round 2 of the program was evaluated through a randomized experiment. A detailed questionnaire was administered to control and treatment households in July–August 2011 (baseline) and during the same months in 2013 (follow-up), so as to avoid seasonality issues. More details on the evaluation design can be found in Pellerano et al. (2014). Although the CGP is unconditional, program messaging did affect child schooling (Pace et al. 2018). Further, an impact evaluation study carried out by Oxford Policy Management (2014) found that the CGP had a large effect on the proportion of children aged 6–19 who were attending school. The impact was driven mainly by a large decline in enrollment among older boys aged 13–17 in the control group. Enrollment for 13–17-year-old boys was 6–10 percent higher among beneficiaries. Impacts of the CGP on girls’ schooling outcomes were not statistically significant but followed a trend similar to that for boys. A qualitative study of the CGP found that children are commonly taken out of school to engage in labor activities, including farm work for boys and washing and child care for girls, especially in households engaged in agricultural activities (Oxford Policy Management 2014). Our analysis extends existing studies in several ways: 1 it investigates the impact of the CGP on gender inequality in schooling; 2 it investigates whether household characteristics, in terms of the sex of the household head and labor capacity, affect the impact of CGP on gender inequality in schooling; and 3 it investigates whether these impacts are shaped by the sex of the cash transfer recipient.2 Our study aims to test four hypotheses. First, we tested whether the CGP has positive effects on child investment in schooling. Positive impacts would be manifested in an increase in children's time in school and a decrease in children's time in labor, particularly among older children (those aged 13–17). Such children tend to be more vulnerable to disinvestment, due to their higher labor value (e.g., in agriculture) as well as their higher schooling costs: primary schooling in Lesotho is free, while secondary schooling is not. Household decisions to invest in child education depend on marginal costs (forgone earnings from child labor and direct educational costs) and marginal benefits (higher expected earnings as an adult as they enter the labor market). CTs may reduce the marginal costs of education by reducing the relative value of children's time in work and leisure compared with schooling. The agricultural household model (Benjamin 1992; Bardhan and Udry 1999) predicts that by alleviating household credit constraints, an exogenous increase in income provided by CTs may affect simultaneously both adult and child labor. If CTs increase labor demand (say through greater employment opportunities on the farm), an increase or a decrease in child labor are both possible, depending on the elasticities of adult and child farm labor with respect to income. However, if child and adult labor are imperfect substitutes, then a decrease in child labor is to be expected. Further, if CTs increase adult participation in wage labor off of the farm, then child labor could increase or decrease, depending on the income effect of the transfer and a household's propensity to hire outside labor. Although the CGP is a UCT, the program included strong messaging about spending money on the needs of children. Hence, we expected to observe an increase in child-specific investments, particularly in education, and a decrease in child participation in agricultural and household labor. The second hypothesis to be tested was whether the CGP reduced gender inequalities in schooling in Lesotho by generating higher impact in schooling among boys vis-à-vis girls, as boys tend to be at a disadvantage with respect to schooling. The unconditional nature of the transfer, coupled with the vulnerability of recipient agricultural households, could lead households to prioritize different needs over investing in all children equally. Therefore, we expected to observe sex differences in the outcomes of time invested in schooling and time invested in both agricultural labor and household chores by child sex and age. If parents expect higher lifetime wages and better employment opportunities for boys than girls, then the marginal benefit of one extra year of education for boys is higher than for girls, all else held equal. If this were the case, we would expect to find CTs having a larger impact on boys than on girls. However, if the marginal costs of child education in terms of forgone earnings remain relatively higher for boys than for girls despite the transfer, then we would conclude that girls benefit more from the transfer than boys. Baseline differences between boys and girls in our sample from Lesotho showed that secondary school–aged boys are more likely to miss and repeat school and are vastly more likely to participate in crop and livestock activities than are girls. Boys aged 13–17 spend on average one additional hour on a typical day working (mostly) on farm activities or household chores compared with girls, which is consistent with the national-level data presented earlier. Among poor households participating in the CGP, boys appear to be more disadvantaged than girls with respect to educational prospects due to their participation in income-generating activities. A third hypothesis was that household composition determines investments in children's schooling, by the sex of the household head and by the household's labor capacity. We thus would expect to find gender differences in child investment impacts due to differences in the value of human capital relative to the cost of present forgone earnings for boys and girls, by household structure. We would also expect to observe more gender-equitable outcomes among MHHs—positive impacts in child schooling, as well as higher impacts among the more disadvantaged children at baseline—as these households tend to be less labor-constrained and, with an increase in income by the CT, are more able to substitute for child labor in favor of more time for schooling. FHHs, on the other hand, tend to be more labor-constrained and, in the context of Lesotho, to be formed by one adult female, usually elderly, and children, usually their grandchildren or foster children. The fourth hypothesis was that the control provided by assigning a CT recipient influences decision making on households’ child schooling investments. In addition to examining differences in investments in boys and girls by household structure, we analyzed heterogeneous impacts by the sex of the CT recipient, who could be the father, mother, grandmother, or caretaker. To test the assumption of unitary household decision making, we compared child outcomes by the sex of the transfer recipient within married MHHs only, in which intrahousehold resource allocation decisions can be made solely or jointly. While we expected to observe gender differences in schooling outcomes according to sex of the recipient, we did not expect to see marked gender preferences. A global study on family preferences based on demographic and health survey (DHS) data suggests that in the case of Lesotho there was no statistically significant difference in girl-boy preference (18–19 percent each) and the vast majority (57 percent) prefer a balanced family in terms of girls and boys (Fuse 2010). The empirical analysis used both baseline and 24-month follow-up data. These surveys were representative of Phase 1 (second round) of the CGP pilot, which covered five districts—Qacha's Nek, Maseru, Leribe, Berea, and Mafeteng—in 10 community councils made up of 96 electoral divisions. Electoral divisions were split equally into treatment and control groups through public lottery events in each community council. Two criteria were used to determine households’ eligibility for CGP: having at least one resident child aged 0–17, and being among the poorest households in the community.3 Our sample is represented by the cohort of children 13–17 years of age living in de jure unmarried FHHs and married MHHs. It included children from both panel and attrition households, and (given the two years’ lag between baseline and follow-up) half of the sample consisted of children appearing in both rounds, including those from split households.4 Overall, the final sample was 2,144 children, 1,066 from the baseline and 1,078 from the follow-up. As shown in Table 1, which describes the sample size and the selection process, approximately 60 percent of the households in the original study had at least one child between 13 and 17 years of age. Among these households, the vast majority (around 90 percent) were either married MHHs or de jure unmarried FHHs. Table 2 presents summary statistics at baseline in 2011, across treatment and control households. Given the restriction of the sample to unmarried FHHs and married MHHs with children of secondary school age, some differences between the treated and control groups are to be expected, despite the randomized nature of the original design. Household composition for adult members over 18 and members aged 0–5 differed by 0.31 and 0.13 members, respectively, between the treatment and control households, and this difference was statistically significant. As a result, treatment households overall had 0.49 more household members than control households. Controlling for differences in household composition is likely to be important for measuring the impact of CTs on child outcomes, as this reflects labor composition. We also found a significant difference across treatment groups in household engagement in crop production, with control households being 6.6 percentage points less likely to participate and producing on average 0.24 fewer goods, including crops, fruits, and vegetables. Both crop production and livestock rearing were important household economic activities for the poor and vulnerable households sampled in Lesotho, with 74–80 percent and 59–62 percent engaged in crop production and livestock rearing, respectively.5 Table 3 compares the samples of de jure unmarried FHHs to married MHHs and shows statistically significant differences in characteristics of the household head and in household attributes. Household heads in FHHs are on average four years older than those in MHHs. FHH heads are also more educated and have 1.8 years more schooling than MHH heads. Other significant differences include larger households, with more members over age 18 in MHHs than in FHHs. Further, MHHs are relatively more engaged in crop production and livestock rearing, produce fractionally more fruits and vegetables and owning more livestock than FHHs. No significant differences were observed at baseline with respect to nonfarm business operations and engagement in wage labor by family members. Within household structure groups, differences between treatment and control groups were minor. (Descriptive statistics not reported here are available upon request.) In Table 4, we compare how girls and boys differed before CGP payments started, particularly in the outcome variables of interest with respect to children aged 13–17. About 52 percent of girls were enrolled in the last three grades of primary school (years 5 to 7), compared with 58.5 percent of boys in the same age group. However, in the same age category, 39 percent of girls were in secondary school, compared with 25 percent of boys. At baseline, 71 percent of boys aged 13–17 had ever repeated a grade in school (12 percentage points more than girls), and 38.9 percent of boys had missed school in the 30 days prior to the baseline survey (6.5 percentage points more than girls). Hence, schooling among older boys appeared to be more volatile and less favored than for girls. For rural households, especially those engaged in agriculture, this implies that for a large share of older boys, the value of their current earnings relative to the opportunity cost of schooling may be considered greater than the value of their future earnings, resulting in a lower share of boys in school. In addition, researchers have observed that boys in Lesotho have lower school enrollment rates than girls and that in the context of the HIV pandemic there has been growing pressure for boys to support households economically (Nyabanyaba 2008). In terms of labor and time use, 35.8 percent of boys participated in their own crop or livestock activities in the week prior to the survey, compared with only 7.9 percent of girls. In addition, boys in this age-group spent on average two days per week on such activities, while girls spent just 0.36 days. However, though girls aged 13–17 spent roughly 95 minutes on a typical day engaged in household chores, boys devoted roughly 35 minutes less on such activities. This confirms well-established gender roles in rural households among secondary school–aged boys and girls, which is seen not only in Lesotho but in many rural settings. Supporting this dichotomy of gender roles, on a typical day, boys participated in farm activities and household chores on average nearly one hour more than girls did. This difference was statistically significant and would add up to a large difference between secondary school–aged boys’ and girls’ participation within a week. Hence, older boys were typically more disadvantaged than girls among poor rural households in Lesotho, in relative time spent on nonleisure and nonschooling activities, and to a less extent in schooling participation.6 We next examined differences in observed child characteristics across FHHs and MHHs. For secondary school–aged children (Table 5), there was a stark contrast in terms of their relationship to the household head. Specifically, 67.7 percent of children in MHHs were the sons or daughters of the household head, while only 43.4 percent in FHHs had this relationship. Further, only 19.4 percent of boys and girls in MHHs were the grandchildren of the head, as opposed to 48.9 percent of grandchildren in FHHs. Grandmothers may view the value of the human capital relative to the opportunity cost of time differently than mothers and fathers. Moreover, households headed by a female elder may face very different constraints in terms of labor capacity and access to assets and services than households headed by younger males. We did not observe meaningful differences in educational outcomes between MHHs and FHHs for these children. Only 32.5 percent of secondary school–aged children were enrolled in junior secondary school (forms A–C), while most of them (54–56 percent) were enrolled in primary school (years 5–7), below the school grade that they should be in given their age. This indicates a lack of resources to remain in school for children in this age-group, most likely due to household economic constraints and a high level of grade repetition. Further, there were no significant differences across MHHs and FHHs in regard to other key schooling indicators, either in the likelihood of repeating school (69 percent vs. 67 percent) or in the likelihood of having missed school days in the month prior the baseline survey (36 vs. 35.5 percent). Interestingly, boys and girls from MHHs were nine percentage points more likely than those from FHHs to have a sister enrolled in school in the current year, a statistically significant difference. Consistent with the above, we also observed a large and significant difference in the likelihood of secondary school–aged children participating in farm labor in the seven days prior to the survey (26 percent in MHHs, vs. 19.6 percent in FHHs). On a typical day, such children in MHHs spent 60 minutes on farm activities, while those in FHHs spent just 31 minutes. However, the same children in FHHs spent on average 83 minutes on chores, while those from MHHs spent 69 minutes on them. Children from MHHs also spent less time at school and doing homework than did those from FHHs. Most of these differences were significant, suggesting that farming activities take precedence in MHHs, where livestock rearing is more prevalent, and take up more time among male children. From the summary statistics, labor activities in MHHs for this group of children were likely to lead to greater substitution away from schooling relative to FHHs. Children from FHHs spent more time on household chores, most likely because children in FHHs were less likely to engage in livestock rearing and more likely to substitute time on chores, including fetching water, sibling care, cleaning, cooking, washing, and shopping. The above equation differed from the first equation only in its incorporation of the Girli indicator, denoting if the sample individual is a girl.7 Here, we were interested in the coefficient β2, which represented the impact of the program on individual-level outcomes (schooling, labor, and time use) for boys, and coefficient β3, which measured the differential impact in outcomes for girls with respect to boys. In our impact estimates table, we also reported for simplicity β2+β3, which represents the impact for girls. Panel A of Table 6 presents the results from the estimation of equation 1 on the impact of the CGP on children. We observed an overall reduction in the likelihood of repeating school years (9.5 percentage points), though the effect was statistically significant only at p<0.10. Further, we found that children aged 13–17 were 8.8 percentage points more likely to be enrolled in school and 14.6 percentage points less likely to have missed any days of schooling in the last 30 days (columns 2 and 3, respectively). Both results were significant at p<0.05 and seemed driven by girls, as shown in Panel B of Table 6, which shows the heterogeneous impacts by gender obtained by estimating equation 2. Magnitudes for girls were slightly higher (in absolute terms), though the β1 coefficient (the interaction term) was always statistically nonsignificant. Looking at the impact of CGP on the time use of girls and boys (Table 7, Panel A), the CGP caused a reduction by 22 minutes in the time spent on household chores for children on a typical day. This represents a large reduction time-wise relative to the baseline average. In addition, from column 3, children also spent approximately 41 minutes more time at school on a typical day. These changes were statistic