Propensity Score Uses and Misuses in Observational Research
*Ralph B. D'Agostino, Jr., Wake Forest University School of Medicine
Keywords: propensity scores, observational studies, causal effects
Propensity scores have recently been extensively employed in applied medical research, particularly in cardiovascular medicine. While this has led to more sophisticated statistical practices for handling bias in observational studies, it also means that more investigators are now potentially misusing propensity scores than ever before. We focus on a brief overview of propensity scores and related issues that often lead to misuse. In particular, attention will be given to: describing what criteria should be used to determine whether a propensity score model “worked”, examining whether the propensity score was fit on the right unit of analysis, and exploring whether appropriate data is being used to fit propensity scores. Finally, a discussion will examine how and when propensity score methods can be used in isolation or in union with other statistical techniques.