Replication code for "Doubly-Valid/Doubly-Sharp Sensitivity Analysis for Causal Inference with Unmeasured Confounding" by Jacob Dorn, Kevin Guo, and Nathan Kallus (https://arxiv.org/abs/2112.11449)
There are two separate tasks: one to replicate the results in Section 7.1 and one for Section 7.2. The two projects share the file DVDS.R and simplified_dvds.R, but otherwise can be run independently.
The file DVDS.R is a set of underlying functions used in estimation. The file simplified_dvds.R are wrapper DVDS and ZSB estimation functions for binary and continuous outcomes.
The first task is the simulations (Section 7.1). To reproduce the figures in the paper:
- First run
dvds_sim.R. This may take one or several hours, and will output data in the folder/Simulations/dvds_output/ - Then run
dvds_plotting.R. This is roughly instantaneous and will output plots in/Simulations/dvds_output/
The second task is to analyze the RHC data (Section 7.2). The file run_DVDS_RHCExploration.R runs the real data application using the data in /Real_data/rhc.csv. The script generates outputs in /Real_data/Results/. Another file called temp is generated but not substantive.
The results reported in the paper were run on an Apple MacBook Pro running R version 4.2.1. Last updated: March 21, 2024.