A0290
Title: Correcting latent class confounder bias in observational studies
Authors: Abdul-Nasah Soale - Case Western Reserve University (United States) [presenting]
Emmanuel Tsyawo - Universite Mohammed VI Polytechnique (Morocco)
Abstract: Causal inference is one of the goals of most inferential statistical analysis, especially in business, economics, and health. However, this objective is often unattainable due to potential unmeasured confounding. Bias is addressed in estimating average treatment effects in settings involving latent class confounders induced by proxy and ill-defined categorical covariates, which are correlated with both the treatment and response. A two-step process for debiasing is proposed. In the first step, the latent classes are recovered using model-free sufficient dimension reduction and clustering techniques. The estimated classes are then incorporated into a mixed model to account for the group structure in the data. The proposed method requires minimal assumptions and yields efficient estimates. An extensive simulation study is included to demonstrate the performance of the proposed method on synthetic data compared to existing methods. Two real applications on medical insurance and energy efficiency are also provided to illustrate the utility of the proposed method in practice.
