Comparing continuous treatment matching methods in policy evaluation

The paper evaluates the statistical properties of two different matching estimators in the case of continuous treatment, using a Montecarlo experiment. The traditional generalized propensity score matching estimator is compared whit a new two steps matching estimator for the continuous treatment case, recently developed [1]. It compares treatment and control units similar in terms of their observable characteristics in both selection processes (the participation decision and the treatment level assignment), where the generalized propensity score matching estimator collapses the two processes into one single step matching. The results show that the two steps estimator has better finite sample properties if some institutional rules define the level of treatment with respect to the characteristics of treated units.