Extended Data Fig. 6: Choice of, and sensitivity to, weather controls specification.
From: Disparate air pollution reductions during California’s COVID-19 economic shutdown
A Sensitivity of main results (Fig. 3) to different functional forms of climate variables (temperature, precipitation, and relative humidity). B Mean squared error (MSE) of models fit with different functional forms of weather controls and no other covariates, cross-validated, for both PM2.5 and NO2 models (see Supplementary Information text (‘Selection of weather controls’) for additional details.). Tables S9-S10 show the MSE statistics for each specification, and the number of observations dropped due to unique weather controls values. We selected interacted vigintile fixed effects as the optimal structure for this analysis based on the combination of lowest MSE for PM2.5 and very few dropped observations, even with full interactions of temperature, precipitation, and relative humidity bins.
