Learning Instrumental Variable from Data Fusion for Treatment Effect Estimation

The advent of the big data era brought new opportunities and challenges to draw treatment effect in fusion, that is, a mixed dataset collected from multiple sources (each source with an independent assignment mechanism). Due possibly omitted labels unmeasured confounders, traditional methods cannot estimate individual probability infer effectively. Therefore, we propose reconstruct label model it as Group Instrumental Variable (GIV) implement IV-based Regression for estimation. In this paper, conceptualize line thought develop unified framework (Meta-EM) (1) map raw into representation space construct Linear Mixed Models assigned variable; (2) distribution differences GIV different mechanisms; (3) adopt alternating training strategy iteratively optimize representations joint IV regression. Empirical results demonstrate advantages our Meta-EM compared state-of-the-art methods. project page code Supplementary materials is available at https://github.com/causal-machine-learning-lab/meta-em.