Fig. 1: Triply robust approach applied in this study.
From: Floods and cause-specific mortality in the United States applying a triply robust approach
We developed a triply robust approach: (1) County-month observations were assigned into treatment/control, and propensity scores of being treatment/control were estimated (named “propensity-score model”). (2) With inverse-probability weights based on propensity scores, age-standardized death counts were regressed on covariates in the control group (named “control-outcome model”). Using the fitted control-outcome model, counterfactual outcomes in the treatment group were estimated had floods not occurred. (3) In the treatment group, the residual mortality rates were regressed on the flood exposure (named “treatment-outcome model”), weighted by the inverse-probability weights calculated from propensity scores.
