Climate Change, Migration, and Inequality
Filiz Garip, Princeton University
Existing work presents mixed findings on the impact of weather events on international mobility. Relying on fine-grained data over 1980-2018 in the Mexico-U.S. setting, we turn to machine learning (ML) tools to first determine if weather events can predict migration choices of 140,000+ individuals. We use random-forest models which allow us to include a comprehensive list of weather indicators measured at various lags and to consider complex interactions among the inputs. These models rely on data-driven model selection, optimize predictive performance, but often produce ‘black-box’ results. In our case, the results show that weather indicators offer at best a modest improvement in migration predictions. We then attempt to open the black box and model the linkages between select weather indicators and migration choices. We find the combination of precipitation and temperature extremes and their sequencing to be crucial to predicting weather-driven migration responses out of Mexico. We also show heterogeneity in these responses by household wealth status. Specifically, we find that wealthier households in rural communities migrate in the immediate aftermath of a negative weather shock (relative to the ‘normal’ weather in their community), while poorer households need to experience consecutive and worsening shocks to migrate to the United States. This pattern suggests that migration as an adaptation strategy might be available to select households in the developing world.