Estimating causal effects of public health education campaigns using propensity score methodology

Abstract

Many evaluations of public health education campaigns attempt to draw conclusions regarding the effect of messages on audiences' attitudes, beliefs, and behaviors based on observational data. To make causal inferences in these instances, it is necessary to adjust estimated campaign effects for possible selection bias due to systematic differences between respondents exposed to the campaign and those that were not. In particular, it is necessary to adjust for the impact of confounding variables that are likely to be determinants of both campaign exposure and outcomes. In comparison to other available methods for adjusting for selection bias such as multiple regression and instrumental variable methods, propensity scores offer a particularly simple way of adjusting estimates of campaign exposure effects for selection bias. This paper discusses the logic of this approach and illustrates its application to the evaluation of the National Youth Anti-Drug Media Campaign.