Yuma Noritomo

This paper studies a fundamental trade-off faced by low-income households in developing countries: whether to invest in productive assets to improve productivity and build resilience, or to invest in human capital to enable entry into the skilled labor market. We study quantitatively how perceptions of risk and returns associated with climate change and structural transformation affect investment decisions. We conduct two field experiments with over 2,000 pastoralist parents in rural Kenya and Ethiopia: a hypothetical choice experiment to estimate parental preferences and subjective beliefs over their children's future education, occupation, and livestock production, and an information intervention that exogenously shifts beliefs about these returns to examine how households adjust actual investment decisions. Together, these analyses allow us to test and quantify how risks associated with productive assets shape human capital investment.

We study the long-run effects of temporary catastrophic drought insurance coverage on pastoralist households in Kenya and Ethiopia. Instrumenting for insurance purchase with randomized premium discounts, we estimate the impacts on households re-surveyed a decade after the start of the experiment. Effects depend on baseline herd size. Among households with smaller baseline herds, insurance reduced holdings of small, liquid livestock like goats -- typically held as a precautionary reserve -- freeing resources for crop production and driving a sharp long-run rise in children's education due to reduced demand for child herding labor. Among households with larger baseline herds, insurance instead facilitated a significant increase in holdings of higher-return large-animals, like cattle, with no comparable effect on schooling. Both patterns strengthened over the following decade, well after insurance itself became unavailable to sample households. These results are consistent with poverty-trap dynamics under which households on opposite sides of an unstable herd-size threshold respond to reduced risk exposure in very different ways resulting in very different outcomes. Even a temporary intervention that reduces uninsured risk can generate self-sustaining gains in human capital and productive capacity long after the intervention itself has ended.

Productive assets are central to poor households' income generation and to many social protection programs, yet they may also raise the opportunity cost of schooling. I study this trade-off among pastoralist households in Kenya and Ethiopia, where children commonly participate in family livestock production. Combining household panel survey data with more than 20 years of satellite measures of rangeland conditions, I exploit plausibly exogenous spatial and temporal variation in drought exposure and differences in children's ages at exposure, relying on household and cohort fixed effects. Drought exposure before and around primary school entry is associated with significantly more years of schooling completed by endline, despite losses in income. Mechanism analysis suggests that persistent herd depletion may reduce current and future demand for children's labor. Droughts are associated with substantial, statistically significant declines in herd holdings that persist for several years. Children respond by reducing livestock work and increasing schooling in the short run. These findings highlight the importance of accounting for complementarity between productive assets and children's labor when designing social protection policies that build or protect these assets.

Social science research often relies on field data drawn from non-randomly selected study sites. 15 In rural development studies, inadvertent exclusion of underrepresented, vulnerable communities 16 risks misguiding policies intended to support them. To examine potential bias in study site 17 selection, we constructed two new datasets in the Peruvian Amazon, where poor accessibility 18 significantly limits the feasibility of field-based research: (1) a universe of communities covered 19 by survey-based rural development studies over the past 40 years through a systematic review 20 (99 studies); and, (2) a large-scale, regionally representative community census (919 21 communities) and household survey (near 4000 households). We find that studied communities 22 (those in 1) have systematically higher development status, household welfare, and inequality 23 than the unstudied communities. This bias is likely due to purposive site selection related to 24 study cost (e.g., access) and targeting (e.g., subject). Our findings underscore the critical 25 importance of study site selection bias in policy-relevant research.

We study how drought and cattle raiding interact to affect child marriage in pastoralist communities in Kenya and Ethiopia, where marriage payments are prevalent, and how these risks narrow adolescent girls' educational and economic opportunities.

We study whether governments can reduce the harms of child labor by prohibiting hazardous jobs without banning child work altogether in Cambodia.

We evaluate a large-scale randomised control trial across 436 pastoral communities in Kenya and Ethiopia testing whether insurance and paid apprenticeships can reduce cattle raiding by addressing the economic conditions that drive conflict.