ABSTRACT
This chapter explains how to control for potentially important covariates by including them in the statistical model with the goal of improving power and precision of estimates. It focuses on the classic ANCOVA in cross-sectional studies or longitudinal studies with two repeated measures (pre-treatment versus post-treatment), describes how ANCOVA is properly used in randomized studies, and discusses challenges and limitations of this approach in observational studies. The chapter also considers ANCOVA in the context of more complicated studies with repeated measures, and distinguishes between situations when covariates vary between or within individuals. It presents the propensity scoring approach for reduction of bias in observational studies and discusses regression adjustments, weighting methods, and matching. The chapter further illustrates data examples with the appropriate code and output included in the online materials. It finally provides references for alternative approaches better suited for causal inference.
