Title: Contrast Specific Propensity Scores

Abstract: Basic propensity score methodology is designed to balance multivariate pre-treatment covariates when comparing one active treatment with one control treatment. Practical settings often involve comparing more than two treatments, where more complicated contrasts than the basic treatment-control one,(1,-1), are relevant. Here, we propose the use of contrast-specific propensity scores (CSPS). CSPS allow the creation of treatment groups of units that are balanced with respect to bifurcations of the specified contrasts and the multivariate space spanned by them.