Covariate Shift Corrected Conditional Randomization Test

Authors: Bowen Xu, Yiwen Huang, Chuan Hong, Shuangning Li, Molei Liu

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, through simulation studies, we demonstrate that our method not only maintains control over Type-I errors but also exhibits superior power, confirming its efficacy and practical utility in real-world scenarios where covariate shifts are prevalent. Finally, we apply our methodology to a real-world dataset to assess the impact of a COVID-19 treatment on the 90-day mortality rate among patients.
Researcher Affiliation Academia Bowen Xu Harvard University bowenxu@g.harvard.edu Yiwen Huang Department of Statistics Peking University 2000010773@stu.pku.edu.cn Chuan Hong Department of Biostatistics and Bioinformatics Duke University chuan.hong@duke.edu Shuangning Li Booth School of Business University of Chicago shuangning.li@chicagobooth.edu Molei Liu: Department of Biostatistics Columbia Mailman School of Public Health ml4890@cumc.columbia.edu
Pseudocode Yes Algorithm 1 Covariate Shift Corrected PCR (cs PCR) Test. Algorithm 2 Covariate Shift Corrected PCR Test with Power Enhancement.
Open Source Code Yes Replication code for our simulation studies is submitted as supplementary material. It will also be made publicly available on Git Hub once our paper is accepted.
Open Datasets No The paper mentions using a real-world dataset from Duke University Health System for the COVID-19 application, but it is not publicly available. For simulations, data is generated, not from a specific public dataset.
Dataset Splits Yes We generate 1000 unlabeled source and target samples to estimate the density ratio and generate 500 labeled source samples for testing.
Hardware Specification Yes All experiments run on a Macbook Pro 2022 M2.
Software Dependencies Yes We estimate the covariance matrix of the sequence of Wi s using the Monte Carlo method and use the momentchi2 package [1] for calculating the p-value.
Experiment Setup Yes Additionally, we empirically choose the best hyperparameter L 3 for all our experiments through additional experiments shown in Appendix B.2.