UC Berkeley

This dissertation combines three empirical studies of household behaviors as they relate to investment in health and human capital in developing countries. The first explores how changes in children's nutrition in Uganda correspond to composition of a household's income. The second studies measurement activities in a cookstove intervention in Darfur, Sudan, with insights into what may be missed in traditional evaluation approaches as well as how technology adoption may benefit from an unintended “nudge.” The third evaluates the impacts of a conditional cash transfer program in El Salvador, with a focus on how program compliance and benefits change time allocations among household members.

Chapter 1 explores the relationship between a household's income source (e.g. wage vs. farm) and children's nutrition in Uganda, in a joint work with Talip Kilic and Calogero Carletto. The analysis uses the three annual waves of the Uganda National Panel Survey and features a series of panel regressions for child height-for-age z-scores (HAZ) under age 5. We control for time-invariant child-level heterogeneity and other time-variant observable characteristics using fixed effects. The analysis finds no impact of short-term changes in total gross income on height scores overall. Sector-differentiated analyses indicate that compared to wage earnings, only the share of income originating from non-farm self-employment exerts positive effects on HAZ, while agriculture is more negative. Within agriculture, the income shares from (i) household's consumption of own crop production and (ii) low-protein crop production appear to underlie a negative effect seen from the share of income originating from crop production. We see that results are driven by the older and poorer cohorts, whose diets may be more influenced by shifts in income and production. Overall, any effects are small, given that coefficients represent a change from 100 percent wage income to 100 percent of the other source in a context where many households experience limited changes from year to year. We also cannot say that these relationships are causal, given that observed changes in income likely reflect changes in endogenous livelihood decisions from year to year. Still, the results suggest the possibility of stickiness of crop production to own consumption. While this may be nutrition-supporting in some contexts, it is possible that income growth in the production of low-protein crops in Uganda, which is known for a low-protein diet, may crowd out consumption of other goods and services that have the potential to serve as better nutritional investments. These results suggest a need for more information about how children’s diets or childcare patterns accompany income changes.

Chapter 2 studies fuel-efficient cookstove adoption in Darfur, in a joint work with Daniel Wilson, Jeremy Coyle, Javier Rosa, Omnia Abbas, Mohammed Idris Adam, and Ashok Gadgil. In this study, we used sensors and surveys to measure objective versus self-reported adoption of freely-distributed cookstoves. Our data offer insights for how effective measurement and promotion of adoption, especially in a humanitarian crisis. With sensors, we measured a 71% initial adoption rate compared to a 95% rate reported during surveys. No line of survey questioning, whether direct or indirect, predicted sensor-measured usage. For participants who rarely or never used their cookstoves after initial dissemination (``non-users''), we find significant increases in adoption after a simple followup survey (p = 0.001). The followup converted 83% of prior ``non-users'' to ``users'' with average daily adoption of 1.7 cooking hours over 2.2 meals. This increased adoption, which we posit resulted from cookstove familiarization and social conformity, was sustained for a 2-week observation period post intervention. Given that most dissemination programs do not employ objective measurement of adoption to inform design, marketing, and dissemination practices, our findings suggest that self-report information may lead programs to over-estimate impacts. A lack of reliable data is likely to prevents insights and may contribute to consistently low adoption rates. Our findings also suggest a potential role for low-cost followup actions that may facilitate learning for a subset of the target population that could benefit from the new technology.

In Chapter 3, I use panel data from El Salvador to examine short-term responses in time use to the Comunidades Solidarias Rurales conditional cash transfer program during 2007/2008, applying difference-in-differences and regression discontinuity methods. Th program was introduced in stages based on observable municipality traits that precluded household-level influence over eligibility. This design allows for a selection-on-observables estimation approach. Because baseline analysis shows significant differences in a few characteristics between earlier and later phases, I use fixed effects specifications to control for time-invariant differences between groups. With only one baseline period, however, I cannot provide evidence against differences in time-variant trends. For each specification, I present results using two bandwidths from the treatment cutoff. To address the small number of municipalities in the sample, I apply wild cluster bootstrapping and present the resulting p-values along those obtained from clustered standard errors as typically applied for larger samples, and show that standard methods would lead to over-rejection of the null hypothesis in multiple instances. I use clustering at the municipality level in both cases. Overall, many of my results are small and somewhat variable across alterative specifications, potentially due to measurement error, a small number of clusters, or simply a small response in the short run to a program offering a relatively small sum of $15-20 a month. Despite these caveats, my findings suggest that for children 6-12, the program appears to have increased school attendance for girls by a small amount relative to boys. There were no gains in enrollment in most specifications, though this may not be surprising in a context where primary school enrollment is already around 90 percent. At the household level, the program may result in a slight reduction of household labor (defined to exclude housework or time allocated to program compliance) for wealthier households relative to poorer households, but a more important change seems to be the shift of productive labor from adult females toward men. Given the total number of statistical tests, however, multiple inference penalties reduce confidence in these few findings.