rct-obs-extrapolation

Code for paper, "Falsification before Extrapolation in Causal Effect Estimation"

This repo is currently under construction. Check back later for end-to-end notebooks recreating the results of our NeurIPS 2022 Paper, "Falsification before Extrapolation in Causal Effect Estimation"

Datasets

• ihdpFull.csv: Full covariate dataset with 985 observations and 103 variables
• ihdp.csv: The dataset with covariates used in (Hill, 2011) and three additional variables indicating the mom's ethnicity, ending with 985 observations and 29 variables (including 1 treatment variable). Note that part of the observations was further removed in (Hill, 2011) to create confounding, which is not done to this dataset.

Access to Women's Health Initiative Dataset

Request access to the WHI dataset used in this work by going to the following URL: https://biolincc.nhlbi.nih.gov/studies/whi_ctos/, and clicking "Request Data."