A0393
Title: Inference in regression with latent clusters: A penalty-free approach
Authors: Abdul-Nasah Soale - Case Western Reserve University (United States) [presenting]
Abstract: Model misspecification is a common problem in estimating treatment effects in samples from heterogeneous populations. The problem of estimating treatment effects in regression involving latent clusters due to samples coming from populations with different means and latent clusters induced by the inclusion of ill-defined categorical predictors in the model is presented. A two-step procedure for model-based subgroup analysis involving k-means clustering and least squares regression is proposed for estimating proportional and cluster-varying treatment effects. The proposed method also provides a graphical check for endogeneity between the latent clusters and the observed predictors via sufficient summary plots. The performance of the method on synthetic and two real data applications on medical and heating costs are included. The theoretical justifications are also provided.
