Implemention of conformal sensitivity analysis (CSA) and conformalized sharp sensitivity analysis (CSSA) described in Conformal Sensitivity Analysis for Individual Treatment Effects by Mingzhang Yin, Claudia Shi, Yixin Wang and David Blei
- Provides code to reproduce the simulation results in the paper
- Implements sensitivity analysis for the ITE on synthetic data and an observational study
- Requires the cfcausal package, which can be installed according to the author's instructions
library("devtools")
library("cfcausal")
devtools::load_all(".")
- The Conformal Sensitivity Analysis (CSA) algorithm is implemented by the R scripts in the folder
./R
- The Conformalized Sharp Sensitivity Analysis (CSSA) algorithm is implemented in
./exp-cf/synthetic_exp-sharp.R
- The folders
./exp-cf
,./exp-ite
,./exp-fish
are the example folders that contain the executable scripts for the simulations in the paper
- Each example folder has a directory
./bash
that contain the bash scripts. The bash scripts are used to submit jobs to a cluster, call the corresponding R scripts, and produce the results
cd ./bash
bash run_XXX.sh
- The python script in each example folder can be used to generate the plots in the paper
- Each example folder has a directory
./out
that contains our numerical results - The fish consumption and blood mercury data is from R package bootsens. The data is also contained in
./exp-fish/data