Brief Intro to My Master Research Paper
Randomized experiments are generally considered as the most effective way for estimating treatment efficacy. However, sometimes they are either too slow or prohibitively expensive to conduct in practice. In many studies, observational data seems to be the best available data, however, making causal inference in this context is challenging. Unlike randomized experiments, investigators have no control over the covariate distributions of treatment and control populations. Therefore, it is a concern that a significant difference observed between treatment and control groups may arise from differences in other covariates between the two groups instead of the treatment assignment. Thus, many methods in causal inference were proposed to reduce bias resulting from imbalanced covariates, most of which have been based on propensity score (Rosenbaum and Robin, 1983). Adjusting for confounders using the propensity score in observational studies has gained a lot of popularity over the past decades. This essay provides a literature review on the central role of the propensity score in making causal inference in observational studies. We introduce the framework for causal inference in Chapter 1. Methods including matching, stratification, regression adjustment, and inverse probability weighting are described in Chapter 2. In Chapter 3 we introduce the generalized propensity score (GPS) as well as the associated estimation methods. A real data analysis is presented in Chapter 4 to illustrate the use of propensity score with matching. In Chapter 5, we discuss a few remaining challenges when using the propensity score in this field.
