An Efficient Doubly-Robust Test for the Kernel Treatment Effect
About
The average treatment effect, which is the difference in expectation of the counterfactuals, is probably the most popular target effect in causal inference with binary treatments. However, treatments may have effects beyond the mean, for instance decreasing or increasing the variance. We propose a new kernel-based test for distributional effects of the treatment. It is, to the best of our knowledge, the first kernel-based, doubly-robust test with provably valid type-I error. Furthermore, our proposed algorithm is computationally efficient, avoiding the use of permutations.
Diego Martinez-Taboada, Aaditya Ramdas, Edward H. Kennedy• 2023
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Two-sample testing | Synthetic high-dimensional location-learning dataset n=3000 v1 | Null Rate0.07 | 7 | |
| Distributional treatment effect testing | IHDP Scenario III Hill (test) | TPR0.34 | 4 | |
| Distributional treatment effect testing | IHDP Scenario IV Hill (test) | True Positive Rate53 | 4 | |
| Distributional treatment effect testing | IHDP Scenario VI Hill (test) | TPR3 | 4 | |
| Distributional treatment effect testing | IHDP Scenario V Hill (test) | True Positive Rate0.99 | 4 | |
| Distributional treatment effect testing | IHDP Scenario II Hill (test) | TPR44 | 4 | |
| Hypothesis Testing | Synthetic Null setting | Empirical Rejection Rate (n=300)7.5 | 2 | |
| Hypothesis Testing | Synthetic Mean shift alternative | Empirical Rejection Rate (N=300)0.155 | 2 | |
| Hypothesis Testing | Synthetic Variance shift alternative | Empirical Rejection Rate (n=300)6.5 | 2 | |
| Hypothesis Testing | Synthetic Localized bump alternative | Empirical Rejection Rate (N=300)0.035 | 2 |
Showing 10 of 10 rows