Fig. 1: Effect of the fraction of hydropower generation affected by drought (FHD) on PM2.5 concentration, dose-response model, placebo tests, and heterogeneity with respect to size and fuel type.
From: Droughts worsen air quality and health by shifting power generation
All results are based on the analysis sample (N = 79,022 plant-month observations), unless otherwise stated. a Point shows the β coefficient from Eq. (1), with error bars indicating 95% confidence intervals (CI) derived from standard errors clustered at the market level (19 clusters). b Distribution of the implied total PM2.5 concentrations, i.e., the marginal effect of the coefficient presented in (a) plus the predicted level of PM2.5 in the absence of droughts. c Regression coefficients (points) and 95% CI (error bars) from a dose-response model where the FHD variable is discretized in four groups. The reference group is FHD less than 0.25. d Distribution of the implied total PM2.5 concentrations, i.e., the marginal effect of each coefficient presented in (c) plus the predicted level of PM2.5 for the reference group. b, d Box plots indicate median (middle line), 25th, 75th percentile (box), and minimum and maximum (whiskers) as well as mean values (triangles). e Regression coefficients (points) and 95% CI (error bars) from two placebo exercises, i.e., the impact of FHD on PM2.5 around non-combustion power plants (N = 62,066 plant-month observations) and around combustion power plants before they are operational. Non-combustion power plants include wind, solar, geothermal, and nuclear. The placebos exclude plants with combustion power plants operating within a 50 km radius. f, g Regression coefficients (points) and 95% CI (error bars) of the impact of FHD on PM2.5 for each size and fuel type sub-group. f Coefficient for large power plants (≥30 MW) is statistically different from small plants (p-value = 0.053). g Coefficient for coal plants is statistically different from oil plants (p-value = 0.038). Reported p-values correspond to two-sided t-tests without adjustment for multiple comparisons. Source data are provided as a Source Data file (sourcedata.xlsx). The data and code used to obtain the estimates are available at https://www.openicpsr.org/openicpsr/project/217201.
