Propensity Score Matching in Observational Studies with Multiple Time Points

Inferring causal effects from observational studies is desirable when the randomized experiments are not feasible or are unethical.Due to the lack of randomization in observational studies, treatment selection depends on various covariates that are associated with outcomes of interest.In multiple-time-point observational studies, in particular, temporal features may potentially affect subjects differently, resulting in time-varying confounders.Using conventional methods for cross-sectional observational studies is inappropriate, since they fail to compensate for the biases arising from the time component.This dissertation focuses on two types of multiple-timepoint study designs, and proposes a propensity score matching method for valid causal inference.Chapter 2 deals with the repeated cross-sectional observational study, where the intervention of primary interest is conducted repeatedly on different cohorts, and a second intervention occurs between two time points.However, the covariates may be imbalanced for subjects in different intervention groups and for subjects participating at different time points.We extend from Rubin's causal model, and establish a potential outcomes framework for this study design.The assumptions for identification of various causal effects are discussed.We propose two matching algorithms for multigroup matching, and compare them via simulation studies.We further provide two