Balancing properties : a need for the application of propensity score methods in estimation of treatment effects

摘要

There has been recently a striking increase in the use of propensity score methods in health sciences research as a tool to adjust for selection bias in making causal inferences from observational controlled studies. However, reviews of published studies that use these techniques suggest that investigators often do not pay proper attention to thorough verification of appropriate fulfilment of propensity score adjusting properties. By using a case study in which balance is not achieved, we illustrate the need to systematically asses the accomplishment of the balancing property of the propensity score as a critical requirement for obtaining unbiased treatment effects estimates.