Child Labor under Cash and In-Kind Transfers

Abstract

1. Introduction

More than 200 million children between 5 and 17 years of age are economically active in the world (ILO 2017). Employment at younger ages is often regarded as one of the main causes of the perpetuation of poverty in the developing world as it typically forces children to abandon school at an early stage, thus interfering with their human capital development. Several explanations for the existence of child employment in poor countries have been proposed in the literature.¹ However, there is ample consensus that children tend to work less when a household’s economic conditions improve.² This view has motivated the inclusion of poverty alleviation programs in the policy toolkit against child employment. The vast majority of such programs provide benefits to vulnerable households which are delivered either in-kind or in cash (World Bank 2015). Yet little is known about the effects of one transfer scheme against another on the time allocation of children. This paper investigates the relationship between the economic activity of children and the provision of in-kind versus cash transfers by exploiting experimentally induced variation in the transfer modality.

Most of the recent literature has focused on the evaluation of poverty alleviation programs whose benefits are delivered conditional on children’s school attendance. Conditional cash transfers are often found to lead to increased school participation and lower levels of economic activity (Fiszbein and Schady 2009). Conditional in-kind transfers can also change the time allocation of children. Ravallion and Wodon (2000) find that a food subsidy program conditional on children’s attendance at primary school increased school participation and reduced economic activity in Bangladesh. Edmonds and Shrestha (2014) show that an in-kind stipend conditional on school attendance reduced hazardous child labor in Nepal, although the effect did not persist once the incentive was removed. However, the literature on conditional transfer programs cannot inform about how the time allocation of children responds to changes in household’s economic conditions, since the conditionality requirement changes the opportunity cost of schooling. In other words, it is not possible to determine whether the reduction in child employment is due to the transfer per se or if it is driven by the program conditionality.

The empirical evidence about the effects of unconditional transfers, either in-kind or in cash, is relatively more scarce. Edmonds (2006) finds that child employment declines and schooling increases after an anticipated expansion of a pension scheme in South Africa. Edmonds and Schady (2012) document large reductions in economic activity for families receiving a cash transfer in Ecuador. Similarly, other studies in Malawi and Mexico have found higher school attendance rates among children in cash-recipient households (Baird, McIntosh, and Özler 2011; de Brauw and Hoddinott 2011). However, to my knowledge there is no evidence in the literature about the impact of unconditional non-productive transfers in-kind on the time allocation of children.³

The first contribution of this paper is to provide novel evidence about the effects of an unconditional transfer in-kind on child employment. Second, and more importantly, by exploiting experimental variation in the transfer modality, this is the first paper to study whether the impact of welfare programs on child economic activity and schooling depends on whether the transfers are given in-kind or in cash. To investigate these issues, the paper studies how child time allocation responds to the Programa de Apoyo Alimentario (PAL), a governmental program providing either a cash transfer or a food basket to poor households in rural Mexico. The food basket includes common staples in the Mexican diet (e.g., rice and beans), as well as less frequently consumed luxury foodstuffs (e.g., canned fish). The cash transfer is set to equalize the government’s purchasing cost of the basket in wholesale markets, and corresponds to approximately 8 percent of household baseline expenditure. Both transfers are given to a woman with the objective of enhancing female participation in decision-making by increasing women’s control over household resources.

The evaluation design of the program relies on an experimental trial in which about 200 villages were randomly assigned to receive either the cash transfer or the food basket, or to a control group that received nothing. Pre- and post-intervention household surveys collected information on children’s economic activity, schooling, and working hours per week. Therefore, the PAL experiment provides a unique setting for the purpose of investigating the effects of cash versus in-kind transfers on child time allocation.

The empirical analysis focuses mostly on children of high-school age (15–16 years old at baseline), as for these children local labor regulations are non-binding and the opportunity cost of not working is higher. Results show that the cash transfer reduces participation in paid employment by 12 percentage points (a 32 percent reduction as compared to the control group) and increases schooling by 10 percentage points (27 percent increase). Unpaid employment in the family farm or business increases by 7 percentage points, suggesting that cash-recipient households partially substitute paid with unpaid child employment. As a result, total hours of work per week decreases by 40 percent as compared to children in the control group. In contrast, the estimated impacts of the transfer in-kind on the time allocation of 15–16-year-old children suggest a much smaller, and not statistically significant, reduction in economic activity. Comparing children in cash and in-kind treatment villages, the null hypotheses of the transfers having the same effect on paid employment, hours of work, and school attendance are rejected.

The larger impact of the cash transfer on the time allocation of children is unlikely to depend on the fact that, unlike cash, the food basket cannot be directly spent on education or other non-food items. Economic theory predicts that infra-marginal in-kind recipients (i.e., those who would consume at least as much food as given in-kind) would treat the transfer as cash. Even if the transfer is extra-marginal (i.e., it results in over-consumption of food), in-kind recipients could partly reduce purchases of food out of non-transfer income to sustain school enrollment costs. Previous evaluations of PAL suggest that the food basket is infra-marginal with respect to total food consumption, and that cash and in-kind recipients spend a similar share of their budget on food versus non-food consumption (Cunha 2014).⁴ Thus, differences in child time allocation between cash and in-kind recipients do not seem to be related to substantial differences in the fungibility of the transfers. Moreover, the empirical analysis presented in this paper does not provide support for these differences being driven by liquidity constraints, differences in social taxation rates across transfers (Jakiela and Ozier 2016), or the in-kind transfer displacing local food suppliers.

The most likely explanation for this finding is that different transfer modalities have different effects on intra-household decision-making. Collective household models predict that household spending depends on the distribution of resources between partners since control over resources leads to control over decision-making (Browning and Chiappori 1998). Consistent with this prediction, several empirical studies have found that cash transfers targeting women increase household spending on children as compared to cash transfers targeting men (Lundberg, Pollak, and Wales 1997; Duflo 2003). In the context of rural Mexico, women are usually financially dependent on husbands and have limited autonomy over the administration of household resources (García and de Olivera 1994). By targeting women, the PAL cash transfer might thus strengthen their position within the household, resulting in increased spending on child welfare and lower child employment. Indeed, both the economic and the anthropological literature have found that cash payments to Mexican women from the well-known Oportunidades program lead to enhanced female autonomy and decision-making (Adato et al. 2000; Radel et al. 2017; Urbina 2020), and there is evidence that Mexican households with higher female decision-making power exhibit lower levels of child employment (Reggio 2011). In contrast, the PAL food basket might have a more limited effect on female decision-making power (and, in turn, on child employment) since, even if the transfer frees up household income which is no longer spent on food, this income is typically controlled by husbands and the transfer does not change the share of cash income (from which child schooling would be paid) under women’s control.

Two findings provide support for this hypothesis. First, consistent with the argument that income controlled by women increases spending on child welfare, PAL cash recipients spend significantly more on education goods and child clothing as compared to in-kind recipients. In contrast, there is no statistically significant difference in household spending on other non-food items, suggesting that this result is not driven by a larger propensity to consume non-food out of the cash transfer but could rather indicate that household spending more effectively aligns to women’s preferences when transfers are given in cash. Second, consistent with the argument that cash has a larger effect on women’s decision-making power, the differential impact of the cash vis-a-vis the in-kind transfer on child employment is statistically significant only for those households in which the woman is in a weaker position according to several proxies of the intra-household distribution of power between partners.

Taken together, the results in this paper contribute to the literature on child employment by showing that cash transfers have larger impacts on the time allocation of children as compared to in-kind transfers, and providing suggestive evidence that the non-neutrality of the transfer modality with respect to women’s decision-making power is a plausible mechanism for this finding. Moreover, the paper is also related to a long-standing literature on the relative merits of cash versus in-kind transfers (see Currie and Gahvari (2008) for a review). Whereas cash transfers are usually praised for having lower administrative costs and for giving freedom of spending to the recipient, in-kind transfers might be preferred when there are externalities from consumption of merit goods (Garfinkel 1973), to induce the non-poor to self-select out of social protection programs (Nichols and Zeckhauser 1982; Blackorby and Donaldson 1988), or because in-kind transfers theoretically provide lower disincentives to work than cash-equivalent transfers (Murray 1980; Leonesio 1988; Munro 1989; Gahvari 1994).⁵ Several studies from this literature used data from the PAL program to investigate the effect of cash vis-a-vis in-kind transfers on adult labor supply (Skoufias, Unar, and González-Cossío 2008), health behavior (Avitabile 2012), consumption and nutrition (Cunha 2014), local food prices (Cunha, De Giorgi, and Jayachandran 2019), learning outcomes of primary school children (Avitabile, Cunha, and Meilman Cohn 2019), and household welfare (Tagliati 2022). This paper contributes to this active area of research by studying child time allocation responses to cash and in-kind transfers.

2. Conceptual Framework and Existing Evidence

The literature reviewed in the previous section suggests that cash and in-kind transfers could affect child economic activity. However, the extent to which the elasticity of child employment is the same across transfer modalities depends on several factors which are related to the characteristics of the transfers and to the mechanisms which induce children to work in the first place. This section reviews these mechanisms and highlights several channels through which child time allocation might respond differently to different transfer modalities.

The classic theoretical framework to study the effects of cash and in-kind transfers on labor supply follows early work by Murray (1980), Leonesio (1988), and Munro (1989). In these studies, utility typically depends on consumption of both subsidized and non-subsidized goods and on leisure. If leisure is a normal good, a cash transfer reduces labor supply through a standard income effect. The in-kind transfer also generates an income effect, but its magnitude relative to a cash-equivalent transfer depends on whether the subsidy is infra-marginal or extra-marginal to the recipient. If it is infra-marginal, optimal consumption and leisure patterns are not distorted and the transfer has exactly the same effect on welfare of an equal-value cash transfer. In contrast, an extra-marginal in-kind transfer results in over-provision of the subsidized good and in a welfare loss compared to a cash-equivalent transfer. In this case, the in-kind transfer reduces labor supply by less than an equal-value cash transfer if there is substitutability between the subsidized good and leisure. The same result holds if utility is separable in leisure and consumption (Gahvari 1994).

Although this framework has been used to explain labor supply effects on adults, it could easily be adapted to study child time allocation by assuming that household utility depends on child leisure or schooling. In such a framework, children might work if preferences for child welfare obey a luxury axiom as in Basu and Van (1998). A cash transfer, or an infra-marginal in-kind transfer, would reduce work participation if it allows the household to move from below to above subsistence consumption. Instead, an extra-marginal transfer would distort the shadow price of child leisure or schooling, resulting in a smaller reduction in child labor supply if there is substitutability between child leisure and the subsidized good within household preferences.

Liquidity constraints represent another channel through which cash and in-kind transfers can have different effects on the time allocation of children. If children work because credit markets are imperfect and households cannot borrow against future earnings (Baland and Robinson 2000; Ranjan 2001), a cash transfer could allow households to sustain the fixed costs related to child welfare (e.g., schooling fees) which would not be affordable without the subsidy. Evidence on the importance of liquidity constraints is presented in Edmonds (2006), who documents large increases in school attendance when poor households in South Africa become eligible for a fully anticipated increase in income. Whereas an in-kind transfer frees up household income which would no longer be spent on the subsidized goods, the transfer might be a less effective buffer against temporary income shocks which could push children out of school since it does not increase household’s cash on hand as much as a cash-equivalent transfer.

Cash and in-kind transfers could also affect the time allocation of children through a household production channel. In a model with imperfect labor markets, children in poor households work in the family business as an imperfect substitute for the lack of access to employment in the market (Bhalotra and Heady 2003; Basu, Das, and Dutta 2010; Dumas 2013). In this context, a transfer (cash or in-kind) can increase child labor supply if it is used to buy productive capital which is complementary to child work in household production.⁶Edmonds and Theoharides (2020) find evidence of a similar mechanism when rural families in the Philippines are granted a productive in-kind asset transfer, as recipients increase adolescent labor to convert the asset into income. In-kind transfers provided by external agents (such as governments or development agencies) might also displace local suppliers as the demand for the subsidized goods or services is at least partially covered by the transfer (Cunha, De Giorgi, and Jayachandran 2019). For a household whose income depends on the production of these goods, this can have two opposite effects on child time allocation. On one hand, the child might work less as a result of the lower opportunity cost of his/her time. On the other hand, the household might try to compensate the profit loss with increased child employment.

Recipients of both transfer types might face redistributive pressure from members of their social network (Jakiela and Ozier 2016), but different transfer modalities could be subject to different social taxation rates. Cash transfers are more easily hidden than in-kind transfers, especially when the latter involve the delivery of food or other physical goods. Social taxation could be particularly high for transfers of large quantities of food, as there is evidence that households frequently forego discounts from purchasing food in bulk to avoid social pressure from sharing (Dillon, De Weerdt, and O’Donoghue 2021). If in-kind transfers are taxed at a higher rate than cash transfers, they could result in lower effects on child time allocation outcomes.

Finally, cash and in-kind transfers can also have different effects on the intra-household decision-making process. Transfer programs to poor households are often paid to women, based on the argument that they would result in increased empowerment for women and better child outcomes (Duflo 2000). As control over resources is related to control over decision-making (Browning and Chiappori 1998), a cash transfer provided to a woman in the household could lower child economic activity if women have higher preferences for child welfare and schooling than men.⁷ In contrast, a non-productive in-kind transfer might be less effective in increasing women’s weight in household decision-making (and, in turn, child schooling) for at least two reasons.⁸ First, the in-kind transfer, unlike cash, cannot be used directly to pay for education expenses and it does not increase the share of cash income (from which these expenses would be paid) controlled by a woman. As a result, household spending might less closely align with women’s preferences when gender-targeted transfers are provided in-kind rather than in cash. Second, in-kind transfers of certain goods, such as food, provided to women might not be as empowering as cash since they could reinforce gender identity norms which cause a disproportionate allocation of home production among women (Akerlof and Kranton 2000; Bertrand 2011).

3. The PAL Program and the Data

3.1. The PAL Program and Experiment

PAL is a social protection program launched by the Mexican government in 2003 with the objective of improving the living conditions of the targeted population. Eligibility for the program was determined through a two-stage procedure. First, villages were deemed eligible if (a) they have a population of fewer than 2,500 inhabitants, (b) they are highly marginalized, (c) they do not receive other transfer programs, such as Liconsa or Oportunidades,⁹ (d) they are accessible and close enough to a store managed by DICONSA, the governmental agency in charge of administering the program.¹⁰ Second, within eligible villages, all households that scored below a means-test poverty threshold were offered the program.

Concurrent with the nationwide implementation of the program, 206 eligible villages were randomly selected to participate in an experimental trial. Each village was randomly assigned to one of three treatment arms: (a) an in-kind treatment arm (103 villages), (b) a cash treatment arm (53 villages), (c) a control group, which received nothing (50 villages). Villages in the in-kind treatment arm received a monthly food basket containing 10 commodities, which were selected by nutritionists to provide a balanced diet. These include cereal-based products (corn flour, rice, pasta soup, cookies, breakfast cereals), pulses (beans, lentils), vegetable oil, and luxury animal-source items (canned fish, powdered milk). In contrast, villages in the cash treatment arm were offered a monthly cash transfer of 150 pesos (US$13), which corresponded to the purchasing cost of the food basket to the government in wholesale markets.¹¹ The transfers were not conditional on family size and, whenever possible, they were given to a woman (typically the spouse of the household head) with the aim of enhancing female participation in household decision-making by increasing women’s resource control.

An additional feature of the program is the fact that, for a random half of villages in the in-kind treatment arm and for all villages in the cash treatment arm, the transfers were intended to be conditional on adult members’ participation in monthly classes, which covered topics related to healthy eating, nutrition, and hygiene practices.¹² However, although the courses were meant to be a mandatory requirement to receive the transfers, no household was ever denied benefits for not attending (Skoufias, Unar, and González-Cossío 2008).¹³ In addition to the lack of enforcement, classes were also taught in villages in the in-kind without classes treatment arm. Because of the contamination of this program component, in the subsequent empirical analysis all villages that received transfers in-kind are pooled together, irrespective of whether they were originally randomized in or out of class participation.

It is worth remarking that the conditionality requirement of PAL is very different from other programs in the literature, which typically require children from beneficiary households to comply with specific schooling requirements. While such programs provide a direct disincentive towards child economic activity by changing the relative price of schooling, this mechanism is not present in the context of PAL in which the only requirement is parental (as apposed to children) class attendance. Nevertheless, one might be worried that class participation might have some direct effect on the time allocation of children. This might occur if parents are forced to reduce their participation in the labor market in order to attend the classes, possibly compensating the reduction in earnings with increased child economic activity. However, this seems very unlikely since parents were only required to attend one class per month and class participation was not enforced. Indeed, previous evaluations of PAL have not found any effect of the program on adult labor supply (Skoufias, Unar, and González-Cossío 2008). For this reason, in the subsequent analysis the estimated effects of PAL on the economic activity and schooling of children are interpreted as arising from purely unconditional transfers.

3.2. Data and Transfer Comparison

In each of the 206 villages included in the experiment, approximately 33 households per village were randomly selected to participate in pre- and post-intervention surveys. The baseline survey was conducted between October 2003 and April 2004, while follow-up data were collected from October to December 2005. The PAL transfers began to be delivered after the completion of the baseline survey. The survey respondent was usually the spouse of the household head, being typically the most knowledgable household member about the economic activity and consumption of the household. The survey provides information on household demographics, expenditure, asset ownership, and the economic activity of each household member older than 12, including school attendance, the main occupation, and the total number of working hours in the last seven days.

The follow-up survey also collected self-reported information about receipt of PAL transfers. About 90 percent of households report receiving transfers from PAL in any treatment arm (see supplementary online appendix S1 for a detailed discussion about program take-up). Due to the lack of administrative data on household eligibility, it is not possible to determine whether the remaining 10 percent of households did not receive the transfers because of ineligibility or imperfect compliance.

In both survey rounds, enumerators visited local shops to collect information on local food prices, from which it is possible to assess the value of the PAL food transfer and compare it to the cash transfer. Although the procurement cost of the in-kind transfer was 150 pesos, the basket was significantly more expensive in recipients’ local markets.¹⁴Table 1 reports the monthly allotment of each commodity in the basket (column 1) and the average post-program value of each transfer item, which is constructed as the product of the PAL commodity allotment and the median price at the village level (column 4). The food basket is worth, on average, 208 pesos (US$18). This implies that the 150 pesos cash transfer could only buy approximately 72 percent of the food basket in recipients’ local markets.

Compared to total household expenditure, both transfers are quite sizable: the in-kind (cash) transfer represents, on average, 11 percent (8 percent) of household baseline expenditure. Estimates of child wages in PAL villages are not available, but it is possible to relate the size of the transfers to the average wage that a child working full time could earn in other poor rural villages in Mexico. This suggests that the cash transfer represents between 25 and 30 percent of the average child wage, while the in-kind transfer represents between 35 and 42 percent.¹⁵

Table 1 also reports the average monthly endline consumption among households in the control group (column 2) and the percentage of households in the control group with post-program consumption below the transferred amount (column 3). Compared to average household consumption, the transfer is clearly infra-marginal for basic necessity items such as beans, vegetable oil, and rice. However, the allotment of luxury animal-source products (canned fish, powdered milk) and of some grain items which are rarely consumed (corn flour, lentils, breakfast cereals) exceeds recipient typical consumption for more than 80 percent of households.

Overall, table 1 suggests that the in-kind transfer was more valuable at face value than the cash transfer, but it was extra-marginal for several food items (i.e., it distorted household consumption). In this case, there is a positive deadweight loss from the in-kind transfer or, in other words, the cash-equivalent value of the transfer is below its market value. Previous evaluations of the PAL program estimate the cash-equivalent value of the food basket at values between 145 and 160 pesos (Cunha 2014; Tagliati 2022). As the cash transfer is set at 150 pesos, the average income effect of the PAL program should be very similar for in-kind and cash recipients. Absent other mechanisms which affect child time allocation, the two transfer modalities are therefore expected to have similar effects on child economic activity and schooling.

3.3. Sample and Baseline Balance

Of the original 206 experimental villages, 9 were excluded from the analysis for various reasons: 2 localities were excluded because households started to receive PAL prior to the baseline survey; 2 villages are geographically contiguous, possibly violating the Stable Unit Treatment Value Assumption (SUTVA); 2 villages refused to participate in the program; 1 control locality was excluded because it received the in-kind treatment; 2 localities were dropped because all households in these villages were receiving Oportunidades, contrary to program rules.

Within the remaining 197 villages, the sample includes 2,591 children aged 12 to 16 at baseline who are observed in both surveys. The choice to restrict the sample to this age group is dictated by two reasons. First, there is no information on the labor supply of children younger than 12. Second, by the age of 16, children should have progressed into high school (preparatoria or bachillerato), which usually ends around age 18. Since the follow-up survey was taken after one and a half years, and because enrollment into university is extremely rare within this sample, choosing age 16 as a cutoff guarantees that a substantial number of children in this age group face a decision between enrolling or completing high school and starting work.

Attrition is rather low: 88 percent of baseline households were resurveyed at follow-up. Although attrition is higher in control localities (14.6 percent) than in cash and in-kind localities (respectively 10.1 percent and 10.7 percent), reassuringly the difference in the attrition rates between the cash and in-kind treatment groups is insignificant (p-value = 0.67). Attrited households with one or more children aged 12–16 are on average smaller and less educated than non-attrited households. Attrited children are on average older and less likely to be in school. The Results section and supplementary online appendix S3 present several robustness checks to account for potentially non-random attrition.

The first three columns of table 2 show the means of several child characteristics by treatment group. Columns 4 to 6 report the mean difference between any treatment group and another. Households are quite large, with six members on average. The sample is poor and low educated. The average schooling of the household head is four years. The average value of household monthly expenditure is 2,200 pesos (US$195) and the share of food consumption corresponds to 67 percent of total consumption. Sample children have completed six years of formal education on average. Children in different treatment arms are overall balanced at baseline in terms of household composition and per capita expenditure levels of the household. However, there are significantly fewer male children in the cash treatment group as compared to the control group. Children are also significantly older in localities in the in-kind and cash treatment arms. Despite these differences, children in any of the treatment groups do not work significantly more than those in the control group, both on the intensive and on the extensive margin: although work participation and hours worked are slightly higher for cash and in-kind recipients, the difference from the control group is not statistically significant.¹⁶

3.4. Child Economic Activity: Data and Institutional Framework

For any individual who is at least 12 years old, the survey asks first to report the main activity in the last seven days; it then asks whether the respondent was involved in any occasional working activity in addition to the main occupation. The dummy variable “economic activity” in table 2 combines both answers and takes the value 1 if the child worked in any activity in the last seven days. Approximately 18 percent of children in the baseline sample work. This definition of economic activity incorporates both paid employment and unpaid work in the family business (but it excludes housework). The empirical analysis also looks at paid and unpaid work separately.¹⁷ This distinction is important since work in the household farm or business is typically easier to adjust on the intensive margin, whereas hours of work in paid employment might be indivisible and thus more likely to displace schooling. In the sample, on average 11 percent of 12–16-year-old children work for pay at baseline, while 7 percent work in an unpaid activity in the family business. Differences across treatment groups in the prevalence of paid versus unpaid work are insignificant (table 2). The average number of weekly working hours, conditional on working, is 34. Full-time work is relatively extended: 43 percent of children worked 40 hours or longer in the last week. Children working in paid activities work on average more hours per week than children working in the family farm or business (38 versus 31 hours per week, respectively).

Figure 1a shows economic activity and schooling participation rates at endline among children in the control group, by age of the child at baseline. Similarly, fig. 1b reports average endline hours of work per week by baseline age among children in the control group. Child labor supply increases markedly with age at baseline: while 23 percent of 12-year-old children are economically active at follow-up, the share of economically active children rises to 60 percent by age 16. This increase is almost entirely driven by higher rates of paid employment at older ages. In contrast, the proportion of children working in unpaid activities remains relatively constant as children get older, at rates between 7 percent and 12 percent. Economic activity increases not only on the extensive margin, but also on the intensive margin: the average number of unconditional working hours at endline goes from 6 hours for children who were 12 at baseline up to 25 hours for children who were 16 years old. Whereas this increase in part reflects changes on the extensive margin, hours conditional on working also show an increasing pattern with age. Finally, school attendance rates at endline are above 57 percent for children who were 14 or younger at baseline, whereas for children in the 15–16 age group, participation in school at endline drops to 38 percent on average.

Figure 1.

Economic Activity and Schooling at Endline by Baseline Age.

Source: Author’s analysis based on original survey data.Note: Outcomes are measured at endline. The sample includes children in the control group who were 12–16 years old at baseline. “Economic activity” is an indicator for the child working in any activity (paid or unpaid) in the last seven days.

Open in new tabDownload slide

The decline in school attendance and the increase in economic activity past age 14 at baseline is related to child labor regulations and to the Mexican education system. Child employment in Mexico is regulated by the Ley Federal de Trabajo. At the time of the intervention, the law prohibited any form of employment for children below age 14, as well as for children older than 14 but younger than 16 who had not finished compulsory school.¹⁸ In the Mexican education system, compulsory school comprises primary and secondary education. The latter is expected to end at age 14–15, although it is not uncommon for children to delay completion because of grade repetition and late enrollment. After completion of secondary school, typically children either start high school (preparatoria or bachillerato) or they enter the labor force. There are additional restrictions to employment for 14–15-year-old children, including mandatory medical examinations, exclusions from working in hazardous activities, and a limit of six working hours per day. Statistics in fig. 1 suggest that child employment laws might not be fully enforced, but nevertheless they limit the work participation of younger children to some extent. As a result, employment of 15–16-year-olds at baseline (and thus 16 to 18 at follow-up) is expected to be more elastic to changes in household income as compared to 12–14-year-olds who would still face legal restrictions to employment in the endline period.

4. Results

The main statistic of interest is δ^Cash, which measures the differential effect of cash vis-a-vis in-kind transfers on the relevant outcome. The parameter δ^Treat measures the impact of the in-kind transfer on child time allocation with respect to the control group. The sum of both parameters, δ^Cash + δ^Treat, provides an estimate of the effect of the cash transfer as compared to the control group. Self-reported data on transfer receipt suggests that about 90 percent of households actually received the transfers.¹⁹ As a result, (equation 1) identifies Intention-To-Treat (ITT) estimates of providing transfers in cash or in-kind.

The main findings reported in the next subsection focus on children who were 15–16 years old at the time of the baseline survey. This choice is dictated by two factors. First, as discussed in the previous section, there are legal restrictions to employment which could limit the extent to which the economic activity of 12–14-year-olds respond to changes in household income. In contrast, for children who were 15–16 years old at baseline, child labor laws are not binding in the follow-up period. Second, and more fundamentally, the study is generally underpowered to detect treatment effects on the time allocation of 12–14-year-old children given the low incidence of paid employment in this age group. For these reasons, the results for 12–14-year-old children are reported in the supplementary online appendix (table S2.2, panel B).²⁰

4.1. Main Findings

Figure 2 presents the estimated impact of PAL on several measures of child time allocation. The point estimates correspond to the following coefficients from (equation 1): δ^Cash (labeled “Cash − Kind”), δ^Treat (labeled “Kind − Control”), and δ^Cash + δ^Treat (labeled “Cash − Control”). The plot in the top-left corner considers any type of economic activity in the last seven days, whereas the top-middle and top-right plots consider, respectively, paid and unpaid activities. In the bottom-left corner, economic activity is measured on the intensive margin (hours of work in the last seven days). The bottom-middle plot shows results for school attendance.²¹

Figure 2.

Estimated Impact of PAL on Child Time Allocation.

Source: Author’s analysis based on original survey data.Note: The figure shows the point estimates and 90 percent confidence intervals of the treatment effects of PAL based on estimation of (equation 1). Each plot shows estimates from a different regression with the plot title as the dependent variable. “Cash − Kind” indicates the coefficient and confidence interval on an indicator for the cash treatment group. “Kind − Control” indicates the coefficient and confidence interval on an indicator for any treatment group. “Cash − Control” indicates the sum of both coefficients and the associated confidence interval. Outcomes are self-reported for the last seven days. “Economic activity” is an indicator for the child working in any activity (paid or unpaid). Hours of work are coded as zero for children not working in the last seven days. “Child time index” is a measure constructed as in Anderson (2008) including the following outcomes: economic activity, paid work, hours of work, and school attendance. The estimation sample includes children aged 15–16 at baseline. All regressions control for the baseline outcome, month of the interviews dummies, and the following variables measured at baseline: age and gender of the child, age and education of the household head. Confidence intervals are calculated based on standard errors clustered at the village level.

Open in new tabDownload slide

Estimated treatment effects show large differences in child time allocation responses to different transfer modalities. Compared to in-kind recipients, children in cash-recipient households experience a 10 percentage point reduction in paid employment, which corresponds to a 25 percent decrease with respect to the control group mean in the endline period. This reduction is only partially compensated by an insignificant 4 percentage point increase in the probability of working in the family farm or business. As a result of substituting paid with unpaid employment, children in the cash treatment arm are 5 percentage points less likely to be economically active than children in the in-kind treatment arm, a sizable, although not statistically significant, reduction of 12 percent with respect to the mean in the control group in the follow-up period. On the intensive margin, cash-recipient children work 6 fewer hours per week than children in the in-kind group. Moreover, the null hypothesis that the two transfers have the same effect on school attendance is rejected (p-value = 0.012), as cash recipients are 10 percentage points more likely to attend school than in-kind recipients.

The treatment effects for each transfer modality with respect to the control group provide more insights into these results. Although in-kind recipients experience some substitution from paid to unpaid employment, the point estimates are small and the null hypotheses of no difference between the in-kind and control treatment arms cannot be rejected. Similarly, the in-kind transfer reduces work participation on the intensive margin by two hours a week as compared to children who do not receive the transfer, but the estimate is insignificant. As a result, the effect of the in-kind transfer on child economic activity and schooling is very close to zero. In contrast, children from cash-recipient households are significantly more likely to substitute paid with unpaid employment, reduce weekly working hours, and increase school attendance as compared to children in the control group. The treatment effects of receiving the PAL cash transfer are large. Paid employment decreases by 32 percent while the intensive margin estimates correspond to a 40 percent reduction as compared to the average weekly working hours in the control group. Effects of this size on the time allocation of children are not unprecedented in the literature. For example, Edmonds and Schady (2012) find that a cash transfer program in Ecuador of approximately the same size as the PAL cash transfer caused a 41 percent reduction in paid employment among 12–16-year-old children.²²

Several robustness checks for these results are presented in supplementary online appendix S3. First, estimates are robust to the exclusion of the baseline outcome variable from the estimating equation (table S3.1). Second, about 12 percent of households in the sample are beneficiaries of the conditional cash transfer program Oportunidades.²³ As this program changes the opportunity cost of schooling by providing cash transfers conditional on school attendance, one might be concerned that the estimated effects on child time allocation might be capturing the effect of Oportunidades rather than the effect of PAL. However, results are robust to the exclusion of Oportunidades recipients from the estimation sample (table S3.2). Third, results are unchanged when estimating an alternative econometric specification which uses two separate indicators for the in-kind treatment arms randomized in or out of the education classes (table S3.3).

Fourth, to alleviate concerns about multiple hypothesis testing, (equation 1) is estimated using, as the dependent variable, an aggregate measure of child time allocation summarizing the independent information contained in the child outcomes in the first five plots of fig. 2. This measure is constructed following the methodology in Anderson (2008). The bottom-right plot of fig. 2 shows that the differential effect of the cash vis-a-vis the in-kind transfer is statistically significant, and that the cash transfer has a large and significant effect on the child time index as compared to the control group. The main results are also robust to corrections of the p-values for multiple hypothesis testing (table S3.4) following Romano and Wolf (2005, 2016).

Finally, two bounding exercises have been conducted to alleviate concerns about the larger attrition rate in the control group vis-a-vis the cash and in-kind treatment groups. In the first, child time allocation outcomes have been bounded following Lee (2009). Given the minimal difference between the attrition rates of the cash and in-kind treatment groups, Lee bounds on the main statistics of interest (the differential effects of cash versus in-kind transfers) are narrow and mostly statistically different from zero (table S3.5).²⁴ The second bounding exercise constructs lower and upper bounds of the impact estimates for the pairwise comparisons across treatment arms following Kling and Liebman (2004). Missing observations have been imputed using the mean within the treatment arm plus or minus the arm specific standard deviation multiplied by 0.1 or 0.25. To construct lower (upper) bounds for the differential effect of cash and in-kind transfers, this amount is subtracted from (added to) the cash treatment group and added to (subtracted from) the in-kind group. Bounds for the cash and in-kind treatment effects with respect to the control group are constructed similarly.²⁵ Estimates in table S3.6 show significantly larger impacts of the cash transfer with respect to both the in-kind transfer and the control group for most outcomes, with estimated impacts on working hours and the child time index which are significant across all bounds. Taken together, the results of both bounding exercises suggest that estimated treatment effects are unlikely to be driven by potentially non-random attrition.

4.2. Mechanisms

The analysis presented above suggests that PAL has substantially larger effects on the time allocation of children when transfer are given in cash rather than in-kind. These differences in child outcomes are unlikely to be driven by the deadweight loss associated to the extra-marginality of the in-kind transfer. As noted in the PAL Program and the Data section, previous evaluations of PAL estimate that the cash-equivalent value of the in-kind transfer approximately coincides with the value of the cash transfer (Cunha 2014; Tagliati 2022).

Other PAL evaluations have also studied the impact of the program on consumption and nutrition (Cunha 2014), adult labor supply (Skoufias, Unar, and González-Cossío 2008), and learning (Avitabile, Cunha, and Meilman Cohn 2019). These outcomes are all potentially related to the relative shadow price of schooling (e.g., learning effects could change the expected return from schooling, or there could be complementarities between schooling and food consumption in the production of the child’s human capital). However, estimated differences across transfer modalities in these outcomes are likely too small to explain the large differential in child time allocation between cash and in-kind recipients. Despite some differences in the composition of food consumption, which are mainly driven by in-kind recipients substituting away from close substitutes of the subsidized good, Cunha (2014) finds that cash and in-kind recipients spend a similar fraction of the transfer on food and non-food consumption, and that nutritional and health outcomes do not significantly differ across treatment groups. Similarly, estimated effects on adult labor supply are not statistically different from zero for both transfer types (Skoufias, Unar, and González-Cossío 2008). As for learning effects, Avitabile, Cunha, and Meilman Cohn (2019) document lower test score results among children from cash vis-a-vis in-kind recipient households. The focus of their analysis is however on primary school children, and therefore their findings cannot shed light on the existence of learning effects for children of high-school age (which are instead the focus of the previous section), nor on whether such effects could affect their employment and schooling decisions.²⁶ Thus, the rest of this section discusses several other channels which could explain the larger impact of the PAL cash transfer on the time allocation of 15–16-year-old children.

4.2.1. Liquidity constraints and social taxation

In the presence of liquidity constraints, cash and in-kind recipients might differ in their ability to mitigate temporary income shocks which might induce households to pull children out of school. Savings are an obvious mechanism through which households could ensure against such shocks. Even if receiving benefits in-kind, households could in principle save part of the transfer to sustain education-related costs. Data on savings is not available, but the survey reports information on the durable items owned by the household, which could represent another channel through which PAL households accumulate savings. Higher durable goods ownership has also been associated with looser liquidity constraints (Jacoby 1994), and there is evidence that poor households use durable assets as collateral for accessing credit markets (Beegle, Dehejia, and Gatti 2006). Thus, differences in durable goods accumulation across treatment arms could indicate differences in the severity of liquidity constraints for cash versus in-kind recipients.

The first column of table 3 reports treatment effects, based on estimation of (equation 1) at the household level, on the number of durables owned by the household.²⁷ Cash and in-kind recipients accumulate significantly more durables than the control group, suggesting that both transfers seem to loosen liquidity constraints. Indeed, the null hypothesis of no difference in durable ownership between cash and in-kind recipients cannot be rejected. Column 2 also looks at differences in the probability of reporting any outstanding debt. The source of debt is not specified, and could therefore include both formal credit and borrowing from individuals in the household social network, but differences across treatment arms could be indicative of different transfer recipients having differential access to credit markets. Results suggest that cash and in-kind recipients do not differ in the probability of having an outstanding debt. Overall, these findings are consistent with previous evaluations of PAL suggesting that cash and in-kind transfers have similar income effects, and with liquidity constraints not being the main channel for the observed difference in child time allocation across transfer modalities.

As discussed in the Conceptual Framework and Existing Evidence section, differences in social taxation across transfer modalities could create a wedge in the amount of the transfer that cash and in-kind recipients could effectively dispose of, which could in turn affect child time allocation. Community sharing could occur either by directly transferring a fraction of the subsidy to members of the household social network, or indirectly by sharing meals or other resources. While the former is hard to observe, the survey reports the number of non-household members eating with the family in the past week and the number of meals which have been shared with them. Columns 3–5 of table 3 show that in-kind recipients are not more likely to share meals with individuals not belonging to the household as compared to cash recipients (coefficients on the cash treatment dummy are actually positive, although insignificant). Although there could be other forms of community sharing which are not accounted for, these results do not point to a higher social taxation rate for in-kind recipients.

4.2.2. Household production

Cash and in-kind transfers could also have different effects on the economic activity of children through their interaction with household production. The provision of a food transfer from the government could displace local food suppliers. Consistent with this hypothesis, Cunha, De Giorgi, and Jayachandran (2019) find evidence that, in PAL villages in the in-kind treatment arm, the transfer put downward pressure on the local prices of the subsidized commodities due to the increased supply of such goods through the program. This could result in a profit loss for households whose income depends on the production of the transferred goods (or of close substitutes of the transferred goods), which could in turn induce food-producing households to mitigate the negative income shock by increasing child employment.

Food production is not observed in the data. However, a proxy measure can be constructed by defining an indicator equal to 1 if, at baseline, at least one household member was employed in agricultural activities in the past year. The sample is then split into agricultural and non-agricultural households, and (equation 1) is estimated separately on each group of households. Results are shown in fig. 3 (see also table S2.3). Compared to the control group, both agricultural and non-agricultural cash recipients reduce child participation in paid employment and hours of work, although estimates are significant only for the latter group. The treatment effects of receiving the in-kind transfer with respect to the control group indicate some small reduction in paid work and total hours for children in non-agricultural households, but the estimates are not significant. Looking at the differential effect of cash versus in-kind transfers, this is statistically significant for most outcomes and for both agricultural and non-agricultural households. Therefore, these results do not support the hypothesis that the larger reduction in child employment under the cash transfer is driven by food-producing in-kind recipients increasing child employment in response to negative income shocks from the transfer.

Figure 3.

Estimated Impact of PAL on Child Time Allocation by Sector of Activity of the Household.

Source: Author’s analysis based on original survey data.

Note: The figure shows the point estimates and 90 percent confidence intervals of the treatment effects of PAL based on estimation of (equation 1). Each plot shows estimates from different regressions, estimated separately for agricultural and non-agricultural households, with the plot title as the dependent variable. “Cash − Kind” indicates the coefficient and confidence interval on an indicator for the cash treatment group. “Kind − Control” indicates the coefficient and confidence interval on an indicator for any treatment group. “Cash − Control” indicates the sum of both coefficients and the associated confidence interval. Outcomes are self-reported for the last seven days. “Economic activity” is an indicator for the child working in any activity (paid or unpaid). Hours of work are coded as zero for children not working in the last seven days. “Child time index” is a measure constructed as in Anderson (2008) including the following outcomes: economic activity, paid work, hours of work, and school attendance. The estimation sample includes children aged 15–16 at baseline. Agricultural households are defined based on a dummy equal to 1 if at least one household member worked in agricultural activities in the past year. All regressions control for the baseline outcome, month of the interviews dummies, and the following variables measured at baseline: age and gender of the child, age and education of the household head. Confidence intervals are calculated based on standard errors clustered at the village level.

Open in new tabDownload slide

4.2.3. Intra-household decision making

Another possibility is that the transfer modality is non-neutral with respect to intra-household decision-making. As discussed in the Conceptual Framework and Existing Evidence section, a food transfer targeted to women might be less effective than a cash transfer in shifting household resource control and decision-making towards women either because the transfer does not change the share of cash income (from which child schooling would be paid) controlled by the female partner, or because of a potential reinforcement of gender identity roles. In the absence of exogenous variation in the gender of the recipient for both types of transfers, the analysis below presents indirect evidence that different effects on women’s resource control are likely important to explain the difference between the impact of cash and in-kind transfers on child economic activity.

Several studies in the literature show that cash transfers given to women tend to be spent in a way which benefits children. For example, Duflo (2000, 2003) and Macours, Schady, and Vakis (2012) find that income controlled by women leads to higher investments in child health, whereas Lundberg, Pollak, and Wales (1997) show that paying a child allowance to women rather than men increases spending on child clothing. Thus, the first test for the hypothesis of the non-neutrality of the transfer modality with respect to intra-household decision-making looks at differences in household expenditure on child-related goods between cash and in-kind recipients. Table 4 shows estimated treatment effects on per capita expenditure on child education (column 1) and child clothing (column 2). For both items, the null hypotheses that cash and in-kind transfers have the same effect are rejected. In contrast, there is no difference in adult clothing expenditure or other non-food expenditure (columns 3 and 4), suggesting that this result is not driven by a generalized increase in expenditure on non-food items among recipients of the cash transfer. In fact, cash recipients spend a significantly larger share of their income on child education and child clothing as compared to in-kind recipients (columns 5 and 6). Overall, these results are consistent with increased spending on children after an increase of income controlled by women, and with the hypothesis that women’s decision-making power increases more when transfers are provided in cash.

If the intra-household distribution of power is indeed relevant for child welfare and depends on the transfer modality, differences in child time allocations between cash and in-kind recipients are expected to be wider for households in which the woman is in a weaker position, since for these households the cash transfer should increase proportionally more women’s weight in household decision-making. Thus, a further test for the proposed hypothesis looks at the heterogeneity of the economic activity estimates for several proxies of the intra-household distribution of power among partners.

The relative strength of each partner in the decision-making process is unobserved, but common proxies from the literature are age and education differences between husband and wife (Browning et al. 1994; Schady and Rosero 2008). The age difference measure is constructed by subtracting the man’s age from the woman’s age, and defining an indicator for the age difference being above the median age difference in the sample. When the age difference is above the median, the woman is relatively older and, therefore, she is expected to have a relatively higher weight in decision-making. Educational differences are measured by an indicator for the wife having more years of schooling than the husband. Households with a larger education gap between husband and wife are larger, are more likely to be indigenous, and tend to be poorer. There is also a negative correlation between the female education gap and spending on child schooling and clothing (table S2.8). These household attributes are suggestive of women with a higher education gap being in a weaker position regarding intra-household decision-making. Finally, another measure used in the subsequent analysis is an indicator for whether the female member in the couple was working at baseline, as this could proxy for gender roles in the household (Alesina, Giuliano, and Nunn 2013; Armand et al. 2020).

The sample is restricted to households in which both the head and spouse are present, and (equation 1) is estimated separately for each group of households. Figure 4 plots the estimated difference between the cash and in-kind transfers on the relevant outcome, together with 90 percent confidence intervals.²⁸ For households presenting characteristics associated with lower female decision-making power, the treatment effect difference is significant for almost all outcomes. For example, when the wife is less educated than the husband, the PAL cash transfer reduces paid employment by 16 percentage points more than the in-kind transfer. Similarly, the effect on total working hours implies that children in cash-recipient households work 10 fewer hours than children in in-kind recipient households. In contrast, when the wife is more educated than the husband, the estimated differences in the treatment effects are smaller in magnitude and the null hypotheses that cash and in-kind transfers have the same effect on child time allocation cannot be rejected. The same patterns are observed when using alternative proxies of the intra-household distribution of power between partners. The differential effect of cash versus in-kind transfers on the time allocation of children is larger for households in which the woman’s control over resources is expected to be lower, such as households with age difference below the median and in which the woman does not work. Overall, the results in fig. 4 provide support for the hypothesis that women-targeted transfers are more empowering to women, and thus have larger effects on child employment, when provided in cash. In the context of Mexico, evidence in favor of the link between the intra-household distribution of power and child employment is also provided by Reggio (2011), who finds that an increase in female bargaining power reduces child employment, especially among girls. This occurs also in the PAL sample. Program effects on economic activity and schooling are overall larger for girls compared to boys (table S2.7).

Figure 4.

Estimated Difference between Cash and In-Kind Treatment Effects by Proxy Measures of the Intra-Household Distribution of Power between Partners.

Source: Author’s analysis based on original survey data.

Note: The figure shows the point estimates and 90 percent confidence intervals of the treatment effect difference between the PAL cash and in-kind transfers based on estimation of (equation 1). Each point estimate and confidence interval is from a different regression with the plot title as the dependent variable, and corresponds to the coefficient and confidence interval on an indicator for the cash treatment group. Outcomes are self-reported for the last seven days. “Economic activity” is an indicator for the child working in any activity (paid or unpaid). Hours of work are coded as zero for children not working in the last seven days. “Child time index” is a measure constructed as in Anderson (2008) including the following outcomes: economic activity, paid work, hours of work, and school attendance. The estimation sample includes children aged 15–16 at baseline in which both the head and the spouse are present. All regressions control for the baseline outcome, month of the interviews dummies, and the following variables measured at baseline: age and gender of the child, age and education of the household head. Confidence intervals are calculated based on standard errors clustered at the village level.

Open in new tabDownload slide

5. Conclusions

Poverty alleviation programs are a popular policy option to contrast child employment. While such programs often include a household transfer provided in cash or in-kind, little is known about whether the effectiveness of such programs depends on the transfer modality. To investigate this issue, this paper studies the effects of cash versus in-kind transfers on the time allocation of children by exploiting the experimental design of PAL, an unconditional transfer program which randomly provided either a food basket or a cash transfer of approximately the same value to poor households in rural Mexico. The empirical results show that economic activity among children of high-school age decreases significantly more when transfers are given in cash. Whereas the estimated impacts of the in-kind transfer are negative but not statistically different from zero, cash recipients experience a 32 percent reduction in paid employment, a 40 percent reduction in working hours, and a 27 percent increase in schooling, compared to children who do not receive PAL benefits.

Both cash and in-kind transfers were given to a female member of the household with the objective of increasing women’s control over household resources. The food transfer does not provide the same freedom of spending as the cash transfer, and might thus have a more limited effect in enhancing female decision-making power over how much of household income should be spent on child schooling. Consistent with this hypothesis and with the literature suggesting that income controlled by women is spent more on children than income controlled by men (Lundberg, Pollak, and Wales 1997; Macours, Schady, and Vakis 2012), cash-recipient households spend a significantly larger fraction of their budget on child education and child clothing compared to in-kind recipients. Moreover, differences in the response of economic activity to cash versus in-kind transfers are larger for households in which women are expected to be in a weaker position, such as households in which the wife is less educated than the husband or does not work. These results are suggestive that the difference between the child time allocation effects of cash vis-a-vis in-kind transfers might be driven by cash transfers being more effective in aligning household preferences towards women’s, resulting in a higher fraction of resources being devoted to children’s well-being.

6. Data Availability Statement

The data underlying this article will be shared on reasonable request to the corresponding author.

Footnotes

Notes

Federico Tagliati is an economist at the Bank of Spain, Madrid, Spain. His email address is [email protected]. The author thanks Aureo de Paula and Valérie Lechene for invaluable advice and suggestions and Alex Armand, Orazio Attanasio, Ciro Avitabile, Antonio Cabrales, Valerio Dotti, Antonio Guarino, Laura Hospido, Yuliya Kulikova, Aitor Lacuesta, Carlos Sanz, Marcos Vera-Hernandez, Ernesto Villanueva, the editor Eric Edmonds, three anonymous referees, and conference participants at the IIPF Annual Conference and SAEe for helpful comments. The author is grateful to Orazio Attanasio and Giacomo De Giorgi for providing access to the data. Financial support from the EPSRC, Fondazione Luigi Einaudi, and Fondazione per il Territorio BPN is gratefully acknowledged. A supplementary online appendix is available with this article at The World Bank Economic Review website

References