摘要:Analysis of rural vulnerability to weather shocks requires an accurate exposure record. Both self-reported and weather data records have been used with potential biases. We are the first to evaluate these biases by constructing two comparable records: self-reported weather shocks from a rural household panel dataset and measured weather shocks from weather data. Despite their positive relationship and similar average welfare effects, the two records display a large discrepancy. Being endogenous, the self-reported record introduces the attenuation bias in analysing non-farm labour allocation in response to shocks. We suggest to instrument the self-reported shock with the measured shock for better analytical quality..