CATE estimation on IHDP semi-synthetic benchmark CEVAE preprocessing (replications 1-5)
0.562sqrt(PEHE)Bayesian X-Learner
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Bayesian X-Learnervariant=default, base_learner=XGB-MSE2026.04 | 0.562 | 0.2 | 0.079 | |
| Huber-DRtype=point, posterior=no posterior2026.04 | 0.575 | 0.153 | 0.037 | |
| Causal BARTm=200, draws=1000 + 1000, chains=22026.04 | 0.597 | 0.239 | 0.14 | |
| Student-t T-BARTresiduals=Student-t, architecture=same as T-BART2026.04 | 0.683 | 0.231 | 0.278 | |
| S-Learnerbase_learner=HistGradientBoostingRegressor2026.04 | 0.72 | 0.363 | 0.091 | |
| Bayesian X-Learnervariant=robust, weights=overlap weights2026.04 | 0.761 | 0.193 | 0.047 | |
| T-Learnerbase_learner=HistGradientBoostingRegressor2026.04 | 0.788 | 0.049 | 0.04 | |
| X-LearnerBayesian layer=no, base_learner=HistGradientBoostingRegressor2026.04 | 0.936 | 0.361 | 0.028 | |
| BCFimplementation=our pymc_bart, total draws=10002026.04 | 1.038 | 0.619 | 0.077 | |
| EconML Causal Forest2026.04 | 1.056 | 0.536 | 0.315 | |
| Bayesian X-Learnercontamination_severity=CB-Huber, delta=1.3452026.04 | 1.232 | 0.457 | 0.739 | |
| Bayesian X-Learnercontamination_severity=CB-Huber, delta=0.52026.04 | 1.795 | 0.745 | 1.368 |