Optimal Transport for Reducing Bias in Causal Inference without Data Splitting
Keywords: Causal Effect Estimation, Optimal Transport
Abstract: Causal inference seeks to estimate the causal effect given a treatment such as a kind of medicine or the dosage of a medication. To address the issue of confounding bias caused by the non-randomized treatment assignment on samples, most existing methods reduce the covariate shift between subpopulations receiving different values of treatment. However, these methods split training samples into smaller groups, which cuts down the number of samples in each group, while precise distribution estimation and alignment highly rely on a sufficient number of training data. In this paper, we propose a distribution alignment paradigm that involves all the training samples without data splitting, which can be naturally applied in the settings of binary and continuous treatments. To this end, we characterize the distribution shift by considering different probability measures of the same set including all the training samples, and reduce the shift between the marginal covariate distribution and the conditional covariate distribution given a treatment value. By doing this, data reduction caused by splitting is avoided, and the outcome prediction model trained on samples receiving one treatment value can be generalized to the entire population. In specific, we exploit the optimal transport theory built on probability measures to analyze the confounding bias and the outcome estimation error, which motivates us to propose a balanced representation learning method for causal inference of binary and continuous treatments. The experimental results on both binary and continuous treatment settings demonstrate the effectiveness of the proposed method.
Primary Area: causal reasoning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 14088
Loading