Research

Abstract: This paper investigates how vulnerable households respond to natural disaster shocks. Using spatial variation in Haiti’s 2021 earthquake intensity and applying matching techniques to national social registry data, I examine associations between earthquake exposure and household coping strategies. Affected households were significantly more likely to adopt costly strategies, including borrowing (+8.7pp), reducing education/health spending (+7.4pp), and depleting seed stocks (+11pp). Female-headed households responded differently, and financial services access reduced reliance on adverse coping by 4-7pp. Robustness checks reveal strongest causal evidence for agricultural liquidation strategies. Findings highlight the importance of cash assistance and financial infrastructure in mitigating shock impacts.

Abstract: This paper studies how drought affects children's education in urban Ghana and traces the household-level mechanisms behind the effect. Using four waves of the Ghana Socio-Economic Panel Survey matched with CHIRPS precipitation data and exploiting within-individual variation in the Standardized Precipitation Index, we find that a one-standard-deviation drought reduces children's school attendance by two to two-and-a-half hours per week, roughly seven percent of instructional time, while enrolment, which is near-universal under Ghana's fee-free basic education, and child health are unaffected. Point estimates indicate that households trim the non-fee components of educational spending, uniforms, books, levies, extra classes, while school fees, zero for most children by policy, show no response. The education effect reflects a budget squeeze that arrives largely uninsured: drought lowers adult weekly earnings by about twelve percent, concentrated among men, without any offsetting rise in employment or working hours, and simultaneously pushes households off free water sources onto paid alternatives, raising water expenditure with highly inelastic demand. 

Abstract: We study how rural households in semi-arid India reallocate labour over the months follow- ing two distinct weather shocks, rainfall deficits (drought) and extreme heat, using the high- frequency Village Dynamics in South Asia (VDSA) panel (2010–2014) merged with India Me- teorological Department gridded weather data, and the local-projections framework of Jordà (2005, 2023) adapted to a micro panel following Crofils et al. (2024). Two patterns emerge. Extreme heat is contractionary, it lowers total days worked and public-works (MGNREGS) participation and sharply curtails off-farm work during the heat-sensitive rabi (winter-crop) harvest, with the largest effects on women. Rainfall deficits instead induce reallocation: when the monsoon fails during the labour-intensive cropping season, local hired-farm work con- tracts and men shift toward out-migration, work outside the village, and public works three to five months later. Mobility is overwhelmingly male; women adjust in place through idle days and domestic work. On the demand side, on-plot labour-hours fall by roughly a fifth four months after a heat shock that strikes during the harvest stage, an effect absent during the grow- ing stage and borne mainly by hired labour, confirming that the contraction operates through harvest-stage labour demand. Timing of the response is governed by the agricultural calendar, identical shocks produce sharply different responses depending on whether they strike during the cropping season or the fallow.

Texts don ’t nudge: An adaptive trial to prevent the spread of COVID-19 in India – A reproducibility and replicability assessment of Bahety et al. (2021), with Florian Neubauer, and Julian Rose

Abstract: The paper by Bahety et al. evaluates an adaptive randomized controlled trial of SMS-based messaging aimed at promoting COVID-19 preventive behaviors in rural Bihar. The intervention shows no statistically significant effects on key outcomes, including social distancing and handwashing. A full reproduction, combined with extensive multiverse robustness checks, yields consistently null results across specifications, including after correcting coding errors. Given limited statistical power, we supplement conventional inference with equivalence testing. For most outcomes, we can rule out effects larger than 15% of the control mean, though smaller effects remain indistinguishable from zero.