pscore2: Stata module to enforce balancing score property in each covariate dimension

Propensity score matching has become a popular empirical method because of its capability of reducing the dimensionality of finding comparable units to conditioning on a scalar quantity. The validity of this approach relies on the balancing property of the propensity score. In practice, this is verified by using statistical tests along with subclassification. Within Stata, this is implemented by the program pscore provided by Becker and Ichino (Stata J., 2002). However, pscore is not constructive regarding the correct specification of the propensity score model, nor does it facilitate the actual requirement of covariate balance. The command pscore2 overcomes these drawbacks. It determines a set of intervals on the respective scalar-dimensional support of the propensity score with respect to the criterion that within each interval statistical similarity of covariates for treated and control observations cannot be rejected for a user-specified probability of a type-I error. Therefore, pscore2 implements a grid-search algorithm that updates the testing interval until convergence to the largest subinterval where covariate balance holds is achieved. The provided options allow for testing higher-order equivalence of each of the marginal covariate distributions for treated and controls. Furthermore, pscore2 automatically distinguishes between continuous and binary regressors and can handle nonvarying covariates.