The field experiment was “natural” in the sense that job-seekers applied to a position in the exact same way that they would have done in the absence of the research team. Because households themselves selected into the “mother applied” and “father applied” groups, I analyze the two sub-samples separately. In section 7, I analyze selection into the two groups. Tasks performed by men and women on flower farms can differ somewhat so that male and female applicants did not compete for the same jobs. The randomization was thus effectively stratified on gender. The research design allowed us to survey only one member of the household, the applicant. A limitation of the data is thus that questions about, for example, the time use of other household members were answered by the applicant. Both unemployed and employed parents appear very knowledgable about their families’ time use, however. I discuss potentially resulting biases below. Summary statistics are presented in table 2. As expected, there are no statistically significant differences between the treatment and control groups in household characteristics, nor in the outcomes analyzed below, cannabis vertical farming in either of the two sub-samples. Literacy rates are low, especially for women. Income and wealth indicators, such as the material that the applicant’s floor is made of, indicate the severe poverty of the sample.
It appears that individuals are unlikely to apply to a flower farm job if they have infants or very young children. In addition to socioeconomic background questions, the survey included a basic expenditure module and questions about the time use of the household members. Because, as is typical for Ethiopia, some families in the sample are very large, average expenditure and time use values for younger sons and younger daughters were recorded, in addition to individual values for the oldest son and the oldest daughter. The primary focus in this paper is on intra-household time use substitution. Table 3 lays out the time use of an average mother, father, daughter and son, in each of the two subsamples, at baseline. Fathers spend more hours than mothers on paid work, although both fathers and mothers appear to be underemployed. Notably, mothers spend almost thirty more hours on house-work per week than fathers do. Agricultural work is included in the definition of house-work: the figures in table 3 essentially suggest that men in rural Ethiopia have a fair amount of leisure time, at least at the time of our baseline survey. Daughters are on average 9.21 years old and sons on average 9.70 years old at baseline, but there is substantial variation in the age of children in the sample. Sons spend on average 12 hours per week in school.
Daughters spend 12 hours per week in the sub-sample in which mothers applied to a flower farm job, and seven hours in the sub-sample in which fathers applied. Travel times and the fixed cost of attending school on a given day are significant: many school-children in the sample did not attend school every day of the week. Parents face both financial and opportunity costs of sending their children to school. On average daughters do seven hours of house-work per week and sons three hours, but some children spend significantly more time on house-work. Childrens’ time use is a primary determinant of human capital accumulation and also has direct welfare consequences of interest. We have seen that parental employment involves a significant increase in household income, a shift in parents’ relative incomes towards the newly employed spouse, and, in the case of mother’s employment, a large decrease in the time the employed parent is able to devote to house-work. If childrens’ time use responds to household income, the relative economic position of the mother and father, or parents’ time use, sons’ and daughters’ activities may thus be affected when a parent gets employed. I now analyze the impact of parent’s employment on children’s time use. In table 4, I aggregate the time that all daughters and all sons, respectively, spend in a given activity. Children aged 5 or older are included in the schooling regressions. Daughters spend six fewer hours in school per week when mothers get employed, a highly significant decrease of 24 percent.
The reason is a need to spend more time helping out at home: daughters’ spend nine, or 48 percent, more hours on house-work when mothers get employed. Sons’ school-time, on the other hand, increases by two-and-a-half hours, or nine percent, when mothers get employed. This increase appears to be unrelated to time use substitution: sons’ house-work time is unaffected by mothers’ employment. The only significant effect of father’s employment on children’s time use is an increase in sons’ school-time of three-and-a-half hours, or 11 percent. Recalling that children’s time use was reported by the applicant, a potential alternative interpretation of the results in table 4 is that employment leads to systematic changes in the accuracy of a treated parent’s beliefs about children’s time use rather than, or in addition to, actual changes in children’s time use. While I cannot entirely rule out such a possibility, comparison of the answers given by mothers and fathers in the seven households in which both parents applied suggests that any resulting biases are likely to be minor. First, at baseline mothers and fathers have remarkably similar beliefs about childrens’ time use. Second, at follow-up there were only negligible changes in the difference between the answers given by a spouse who was randomly chosen for employment and the answers given by his or her spouse who also applied but was not chosen employment. Table 5 investigates the impact of mother’s employment on daughters’ time use in more depth. The dependent variables in columns 1 – 6 are indicators that take value 1 if the household-member primarily responsible for a given house-work task is/are daughter. Mother’s employment significantly increases the probability that daughters are responsible for fetching water, grinding grains / cooking, cleaning / washing / ironing, food shopping, and caring for children. In panel B, I separately analyze how mother’s employment affects the extensive and intensive margin of the oldest and younger daughters’ school-time. The oldest daughter is most affected: the probability that she is enrolled in school decreases by 11 percentage points, or 12 percent, and her hours of schooling, conditional on being enrolled, fall by almost four hours, or 24 percent, when the mother starts working. The estimated effect on the probability that younger daughters are enrolled in school is also negative, but not significant. There is a marginally significant 15 percent decrease in the school-hours of enrolled younger daughters when mothers get employed. The evidence in tables 4 and 5 provides a clear picture of intra-household time use substitution in rural Ethiopia. Part of the reason why sons are not required to take over duties from the father when he gets employed appears to be that fathers spend little time on housework in the first place. Even in households in which fathers spent substantial time on housework before applying to a flower farm, drying cannabis sons do not take over those responsibilities when the father starts working, however. Moreover, sons’ school-time increases both when a mother gets employed and when a father gets employed. The increase in schooling thus appears unrelated to the shift in parents’ relative bargaining power that likely follows employment. The evidence we have seen indicates that sons’ school-time increases when a parent gets employed due to the resulting increase in household income.
Primary school is supposed to be free in Ethiopia, but parents face costs associated with uniforms, material and clothing for school – costs that can be significant for households as poor as those in this paper’s sample. The picture for daughters is a different one. The results in tables 4 and 5 indicate that daughters take over several of the mother’s house-work tasks when she gets employed and end up making up for about two thirds of the decrease in the mother’s house-work time. To do so, daughters are forced to attend school less. In contrast, daughters’ schooling is unaffected by father’s employment. These results leave little doubt that intra-household time use substitution is key to schooling outcomes in rural Ethiopia. The evidence in table 5 that some daughters drop out of school entirely when mothers get employed is particularly worrisome because school attendance is path-dependent: it may be difficult for a daughter to return to school later on once she has dropped out. The reason why daughters, unlike sons, do not attend school more when fathers get employed is not immediately clear. It may be that parents have “lexicographical” preferences over childrens’ schooling and prioritize sons before daughters. It is also possible that fathers weigh sons’ schooling relative to daughters’ schooling more than mothers and that employment increases fathers’ influence over child schooling decisions. In the next section I present a simple theoretical framework of the household that captures the time use effects we have seen so far and derive auxiliary predictions that further illustrate the trade-offs that arise when the house-work of different family members is substitutable.The magnitude of the effect of mother’s employment on daughters’ schooling is far from uniform across households. In table 6, I test predictions 2 – 4 by interacting proxies for the household characteristics that the framework above suggests should induce heterogeneity in the response of daughter’s schooling to mother’s employment with the treatment. While the tests are thus indirect – focusing on the mitigating effect of a given covariate on the response to employment rather than the covariate itself – they provide informative evidence on the intuition behind each prediction. As a proxy for mother’s bargaining power I use her share of baseline earned income, a “distribution factor” commonly used in empirical research that has been shown to influence individuals’ control over household decisions . In the first two columns we see that the impact of mother’s employment on daughters’ house-work is two hours bigger, and the impact on daughters’ school-time three hours more negative, for every one standard deviation increase in the mother’s initial bargaining power. Prediction 2 thus finds empirical support: it appears that, relative to fathers, mothers prefer daughters to take over more house-work when mothers get employed and are left with less available time to spend on house-work duties. In columns 3 and 4 I proxy for mother’s weight on daughters’ well-being with daughters’ initial expenditure share. Daughters’ expenditure share will reflect a combination of mother’s and father’s weight on daughters’ well-being . The interaction between the treatment and daughters’ weight is negative but not significant in the housework regression. The decrease in schooling is significantly smaller, by 3 hours, for every one standard deviation increase in daughters’ expenditure share, as predicted by prediction 3. It thus appears that the negative effect of mother’s employment on daughters’ schooling is dampened in households that value daughters more. If the negative effect of mother’s employment on daughters’ schooling is due to gender specific time use substitution, as I have argued, then the effect for a given daughter should be smaller the more females there are in the household. The reason is of course that the house-work previously done by the mother can be spread across females. In columns 5 and 6 we see that, while the impact of mother’s employment on daughter’s time use remains significant even when other adult women are present, the impact is greatly reduced in such households. In columns 9 and 10 we see that the effect on the oldest daughter’s house-work time is two-and-half hours reduced, and the effect on the oldest daughter’s school-time one and-half hours reduced, for every additional daughter that is present in the household. It is thus clear that other female household-members share the extra house-work burden that mother’s employment implies, as the framework predicts. In sum, the evidence in table 6 indicates that the extent of time use substitution between mothers and daughters depends on parents’ preferences and on the number of females available for house-work. The framework above can thus account for important determinants of heterogeneity in how children’s time use responds to a parent’s employment. In the next section I explore how child time use decisions are made.If time use substitution between parents and children is significant and parents are imperfectly altruistic, then parent-child agency issues are likely important for children’s well-being . The framework above follows the literature in modeling parents as the relevant decision makers in the household.