Falsification using higher order influence functions for double machine learning estimators of causal effects

Introduction

Here we provide the code to reproduce the analysis described in:

Organization

• create_simulation_parameters_continuous_X.R — simulates training and oracle samples, fits nuisance parameters models in the training sample, and computes inverse Gram matrices for the simulation with continuous X.
• continuous_X_sim.R — simulates estimation samples, computes estimates of counterfactual means, and computes second order bias estimates for the simulation with continuous X.
• continuous_X_sim_choose_stopping_k.R — uses the approach described in the supplement to choose k for simulation with continuous X.
• create_simulation_parameters_binary_X.R — simulates training and oracle samples, fits nuisance parameters models in the training sample, and computes inverse Gram matrices for the simulation with binary X.
• binary_X_sim.R — simulates estimation samples, computes estimates of counterfactual means, and computes second order bias estimates for the simulation with binary X.
• compute_bias_all_sims.R — computes the conditional bias, and the Cauchy-Schwarz bias for simulations with binary and continuous X.
• create_simulation_parameters_CRC_GAN_data.R — simulates training and oracle samples, fits nuisance parameters models in the training sample, and computes inverse Gram matrices for the simulation with artificial data based on the National Cancer Database.
• CRC_GAN_sim.R — simulates estimation samples, computes estimates of counterfactual means, and computes second order bias estimates for the simulation with artificial data based on the National Cancer Database.
• create_parameters_CRC_real_data.R — sample splits, fits nuisance parameters models, and computes inverse Gram matrices for analysis of the National Cancer Database.
• CRC_real_data_analysis.R — computes cross-fit estimates of counterfactual means (and the risk difference), and estimates of the second order bias.
• CRC_real_data_analysis_choose_stopping_k.R — uses the approach described in the supplement to choose k for analysis of the National Cancer Database.
• process_outputs.R — R file which reproduces the tables and figures displayed in the main text, summarizing simulation results.
• process_outputs_supplement.R — R file which reproduces the tables and figures displayed in the supplement, summarizing simulation results.
• src — Folder containing scripts with functions and dependencies to be called by the above files.
• params — Folder should be created to which simulation parameters, which are outputs of create_simulation_parameters..., are saved.
• output — Folder to which simulation outputs are saved.
• figures — Folder containing figures, which are outputs of process_outputs.R and process_outputs_supplement.R.

Correspondence

If you have any questions, comments, or discover an error, please contact Kerollos Wanis at knwanis@gmail.com.