Can Education be Standardized? Evidence from Kenya
Most recent version appears as Becker Friedman Institute Working Paper 2022-168, June 2022
We examine the impact of enrolling in schools that employ a highly-standardized approach to education, using random variation from a large nationwide scholarship program. Bridge International Academies not only delivers highly detailed lesson guides to teachers using tablet computers, it also standardizes systems for daily teacher monitoring and feedback, school construction, and financial management. At the time of the study, Bridge operated over 400 private schools serving more than 100,000 pupils. It hired teachers with less formal education and experience than public school teachers, paid them less, and had more working hours per week. Enrolling at Bridge for two years increased test scores by 0.89 additional equivalent years of schooling (EYS) for primary school pupils and by 1.48 EYS for pre-primary pupils. These effects are in the 99th percentile of effects found for at-scale programs studied in a recent survey. Enrolling at Bridge reduced both dispersion in test scores and grade repetition. Test score results do not seem to be driven by rote memorization or by income effects of the scholarship.
Most recent version appears as Williams Economics Working Paper 2021-13, July 2021
Joint with Pamela Jakiela
Languages use different systems for classifying nouns. Gender languages assign nouns to distinct sex-based categories, masculine and feminine. We construct a new data set, documenting the presence or absence of grammatical gender in more than 4,000 languages which together account for more than 99% of the world’s population. We find a robust negative relationship between prevalence of gender languages and women’s labor force participation and educational attainment both across and within countries. We also demonstrate that grammatical gender is associated with both weaker legal support for women’s equality and reduced female bargaining power within the household.
Joint with Pamela Jakiela, Heather Knauer, and Lia C. H. Fernald
We model household investments in young children when parents and older siblings share caregiving responsibilities and when investments by older siblings contribute to young children’s human capital accumulation. To test the predictions of our model, we estimate the impact of having one older sister (as opposed to one older brother) on early childhood development in a sample of rural Kenyan households with otherwise similar family structures. Older sibling gender is not related to household structure, subsequent birth spacing, or other observable characteristics, so we treat the presence of an older girl (as opposed to an older boy) as plausibly exogenous. Having an older sister rather than an older brother improves younger siblings’ vocabulary and fine motor skills by more than 0.1 standard deviations. Viewed through the lens of our model, the empirical pattern we observe suggests that: (i) older siblings’ investments in young children contribute to their human capital accumulation, and (ii) households perceive lower returns to investing in older girls than in older boys.
Evaluating the Effects of an Early Literacy Intervention
Conditionally accepted based on Stage 1 Pre-Results Review at Journal of Development Economics
Joint with Pamela Jakiela, Lia C. H. Fernald, and Heather Knauer
We conduct a cluster-randomized evaluation of an early literacy intervention that provided Kenyan parents with illustrated children’s storybooks and modified dialogic reading training. Rural communities were randomly assigned to treatment or control. Within treatment communities, households were further randomized to receive children’s storybooks in either Luo (the mother tongue of all children in the sample) or English (a national language, and the primary language of instruction in grade 4 of primary school and beyond). We estimate the impacts of treatment on children’s vocabulary and literacy skills. Our design also allows us to document household responses to the intervention including behavioral responses by parents and older siblings and overall impacts on parental time investments in children.
A Firm of One’s Own: Experimental Evidence on Credit Constraints and Occupational Choice
(Also appears as IZA Discussion Paper 10583.)
This study presents results from a randomized evaluation of two labor market interventions targeted to young women aged 18 to 19 years in three of Nairobi’s poorest neighborhoods. One treatment offered participants a bundled intervention designed to simultaneously relieve credit and human capital constraints; a second treatment provided women with an unrestricted cash grant, but no training or other support. Both interventions had economically large and statistically significant impacts on income over the medium term (7 to 10 months after the end of the interventions), but these impacts dissipated in the second year after treatment. The results are consistent with a model in which savings constraints prevent women from smoothing consumption after receiving large transfers – even in the absence of credit constraints, and when participants have no intention of remaining in entrepreneurship. The study also shows that participants hold remarkably accurate beliefs about the impacts of the treatments on occupational choice.
Perils of simulation: parallel streams and the case of stata’s rnormal command
Large-scale simulation-based studies rely on at least three properties of pseudorandom number sequences: they behave in many ways like truly random numbers; they can be replicated; and they can be generated in parallel. There has been some divergence, however, between empirical techniques employing random numbers, and the standard battery of tests used to validate them. A random number generator that passes tests for any single stream of random numbers may fail the same tests when it is used to generate multiple streams in parallel. The lack of systematic testing of parallel streams leaves statistical software with important potential vulnerabilities. This paper shows one such vulnerability in Stata’s rnormal function that went unnoticed for almost four years, and how to detect it. It then shows practical implications for the use of parallel streams in existing software.