Working Papers

Standard evaluations of early childhood education measure returns to the individual attendee, ignoring the possibility that skills acquired in early childhood propagate to peers through sustained social interaction. If such spillovers exist, the social returns to early childhood investment are systematically underestimated. We test this hypothesis using nationally representative panel data on middle-school students in China, exploiting random classroom assignment to estimate the causal effect of exposure to kindergarten-attending peers on mental health, social adjustment, and academic outcomes. Effects are negligible after three to six months of peer exposure. After fourteen or more months, a ten-percentage-point increase in peer kindergarten share improves non-cognitive outcomes by 0.074 standard deviations and academic achievement by 0.072 standard deviations. Drawing on the technology of skill formation, we interpret these duration-dependent results as evidence that behavioral norms and socio-emotional competencies require sustained peer interaction to propagate. Effects are concentrated among disadvantaged students, suggesting that deliberate classroom integration represents a low-cost lever for reducing socioeconomic disparities, especially in non-cognitive skill formation.

Many languages have sex-based or gendered grammar, that is, they require speakers to mark male/female distinctions as part of regular everyday conversations. For example, in Spanish book is masculine (el libro rojo), while table is feminine (la mesa roja). We ask whether women who speak such gendered languages report different gender attitudes and household outcomes. Using data on more than 1.5 million women residing in 53 countries and speaking 151 languages, we find that women speaking gendered languages report more intimate partner violence (IPV) (1.2 percentage points) and lower participation in household decisions (4-6 percentage points). This pattern is consistent with linguistic theory: repeated gender marking in everyday conversations may make gender categories more salient, and hence shape real world outcomes. We then examine whether the association operates primarily through long-run cultural transmission (native language) or through more immediate cognitive salience (interview language). We find evidence consistent with both channels, with somewhat stronger support for the cultural-transmission channel. We also show that the pattern remains after controlling for different cultural and institutional factors.

India's schools were closed for 82 weeks during COVID-19; when they reopened in early 2022, children returned far behind on foundational skills. We ask whether the cheapest channel available, a text message to any mobile phone, can help them catch up, and how much a brief human follow-up adds. In a cluster-randomized trial in 1,162 villages across 19 Indian states, run in the first weeks after schools reopened, about 27,000 children in Grades 3-5 who could not yet subtract received one numeracy activity a day by SMS for four weeks, matched to each child's level and written for a caregiver to do with the child. In a second arm, a field worker from the implementing NGO, already known to the family, also called once a week to ask whether the activities were being done, but did not teach. Messages alone raise arithmetic by 0.14 standard deviations and the share of children who can subtract from 14 to 19 percent; with the calls, the gains rise to 0.24 standard deviations and 23 percent. There is also a spillover to reading, which the program did not address, of 0.07-0.15 standard deviations, and the gains are present in households with only a basic phone. Where one-way texting has mostly failed, level-matched activities from a trusted organization succeed, and a brief call with no instruction adds a further 0.10 standard deviations. The messages were virtually free to send, and the calls cost well under a dollar per child.

We study the long-run consequences of a major early-life exposure to a medical innovation: the rapid diffusion of sulfonamide antibiotics in the late 1930s. We leverage large cross-state differences in pre-sulfa mortality from maternal infections, scarlet fever, and influenza/pneumonia to construct plausibly exogenous variation in the intensity of early-life exposure among U.S. birth cohorts. Linking these baseline differences to more than 34 million NCHS death records and 5 million Census-Numident observations, we find that children aged 0-4 in high-mortality states at the time of sulfa's introduction experience longer adult lifespans, roughly 1.4 months per standard-deviation increase in baseline disease burden. The estimates are robust across specifications, appear in both administrative datasets, and are concentrated among children from low-SES families and rural areas. Mechanism analyses reveal reductions in infant mortality and adult cognitive, ambulatory, and functional disabilities, alongside modest but significant increases in adult education and income. The results show that major medical innovations can generate both immediate mortality gains and persistent improvements in longevity over the life course.