Replication Data and Code for “Bayesian-Bootstrap Sensitivity Analysis for Partially Identified Treatment Effects: Computation and Experimental Benchmarking”
Description
This replication package accompanies the study “Bayesian-Bootstrap Sensitivity Analysis for Partially Identified Treatment Effects: Computation and Experimental Benchmarking.” The study develops a computational Bayesian procedure for evaluating treatment-effect uncertainty when conditional exchangeability may fail. The method combines cross-fitted propensity-score estimation, Bayesian-bootstrap posterior simulation, a normalized odds-weight sensitivity model, and an exact threshold algorithm for calculating the endpoints of partially identified treatment-effect intervals. The empirical application evaluates the effect of participation in the National Supported Work Demonstration on real earnings in 1978. The analysis combines treated NSW participants with nonexperimental comparison observations from the Current Population Survey and Panel Study of Income Dynamics. The randomized NSW comparison group provides an external experimental benchmark. The package reproduces the experimental benchmark, descriptive statistics, overlap and covariate-balance diagnostics, effective sample sizes, Bayesian posterior sensitivity intervals, robustness comparisons, Monte Carlo simulations, tables, and figure reported in the manuscript. The original NSW, CPS, and PSID data are publicly distributed through the National Bureau of Economic Research data page associated with Dehejia and Wahba’s studies. Because the source page specifies attributable non-commercial use, the raw data are not redistributed in this repository. The README provides the source link, required citations, variable definitions, and instructions for obtaining and preparing the original files.
Files
Institutions
- Abbottabad University of Science and TechnologyKhyber Pakhtunkhwa, Havelian