kevjosey/causal-eco

Subpopulation-Specific Exposure Responses between Fine Particulate Matter and Mortality in the Medicare Population

Stratified Fine Particulate Matter Exposure Response Functions (ERFs)

Functions

• calibrate.R: Calibration function for estimating covariate balance weights with an entropy loss function.
• gam_ipw.R: Functions for estimating exposure response function with generalized additive models and inverse probability weights. Includes a sandwich variance estimator for the point-wise confidence intervals of the ERF, which incorporates/propagates uncertainty from the fitted calibration weights.
• gam_dr.R: A doubly-robust implementation of the GAM ERF estimate similar to the estimator in gam_ipw.R that requires an initial outcome (GAM) outcome model fit. Should be used in tandem with erf_fun.R.
• erf_fun.R: Implementation of the doubly-robust exposure response function specific to the ecological PM2.5 regression.
• srf_fun.R: Implementation of the stochastic intervention analysis (useful for obtaining excess events associated with different NAAQS policies).

Analysis

Data Cleaning and Descriptives

• data_process.R: Data processing script which aggregates binary responses into strata within ZIP-code years.
• desriptives.R: Descriptive statistics appearing in Tables 1 and 2 of the manuscript. Additional code for generating the covariate balance plots is also provided here.

Models

• region_srf: Scripts that fit the region-specific (and whole US) stochastic intervention curves using doubly-robust methods. We use entropy balancing to estimate the IPWs - think of entropy balancing as a type of method of moments estimator whereas the more traditional way of estimating IPWs is with a plug-in estimator.
• state_srf: A doubly-robust implementation to estimate the shift responses (stochastic interventions).

Experiments

• erf_sim.R: Simulation study to demonstrate the effectiveness of the exposure response estimator.
• srf_sim.R: Simulation study examining accuracy of the stochastic interventions.

Plots

• Plots/plot_region.R: Script for generating the main plots that appear in the manuscript.
• Plots/plot_state.R: Script for generating state-specific outcomes.
• Plots/plot_eco.R: Script for plotting sensitivity analysis results of the exposure response curve. Plots include a comparison of the overall ERF using an alternative exposure assessment.