(Mis)measuring Local Weather Shocks: Evidence from Drought and Migration in Ethiopia
with Michiel Gerritse and Frank van Oort · Under review
Presented at the 6th GCD Workshop on Regional Development and Innovation (A Coruña, 2024),
the Regional Science Workshop on Remote-Sensed Identification and Networks (Oviedo, 2024), and the
PSE Summer School on Migration (2023).
Extreme weather shocks strike locally, yet their effects are typically estimated
from exposure aggregated across districts, regions or countries. We show that this aggregation
systematically biases impact estimates and propose a simple correction. Using high-resolution
rainfall and satellite-derived population data for more than 13,000 Ethiopian settlements
(2000–2016), we find that drought reduces population in vulnerable settlements by about
2.5 percent, concentrated among working-age adults and coinciding with falling food production,
income and health. Measured at progressively coarser scales, the impact disappears, even at
resolutions common in the literature. We decompose the bias into three sources (partial exposure
misclassified, local vulnerability averaged away, area weights misrepresenting where people live)
and derive a fix: measure each shock at fine scale, weight by local vulnerability, then aggregate.
The correction recovers estimates in areas roughly four times larger than those at which the
standard binary indicator fails; simulations confirm its performance. Building exposure locally
before aggregating is a low-cost change that can substantially improve empirical practice.
Paper (PDF, working paper, September 2026) ·
Code (available soon) ·
Data (available soon)
Memory, Expectations, and Climate Adaptation
with Frank van Oort · Pre-registered field experiment (AEA RCT Registry) · Working paper, September 2026
Smallholder farmers face large and growing weather risk, yet take-up of insurance and other
protection remains low. The value farmers place on protection depends on the weather they
expect. Where records and forecasts are limited, those expectations are built from memories of
past weather, and distorted recall can leave demand for protection poorly aligned with the risks
farmers actually face. We measure how much experienced weather survives into perceived risk. In
a survey experiment with 686 farmers in eastern Uganda, we elicit remembered and expected
weather through visual token allocations and map geospatial rainfall and vegetation records onto
the same scale. We find that memory is strongly compressed, with about 90 percent of seasonal
weather variation lost in the baseline estimate. Yet remembered weather strongly predicts
expectations, so this compression carries into beliefs about future risk. By contrast, salient
extremes remain vivid. Farmers who recall an extreme are 19 percentage points more likely to
expect another and also plan substantially more adaptation. In a randomized experiment,
information on past local weather moves expectations toward the record and raises confidence.
Accessible historical information can therefore help farmers assess climate risk and inform
adaptation decisions in the face of climate extremes.
Paper (PDF, working paper with online appendix, September 2026) ·
Photos from the field ·
Code (available soon) ·
Data (available soon)