Propensity score matching

The typical evaluation problem aims at quantifying the impact of a ÔtreatmentÕ (e.g. a training programme, a reform, or a medicine) on an outcome of interest (such as earnings, school attendance or illness indicators), where a group of units, the ÔtreatedÕ, receive the ÔtreatmentÕ, while a second group remains untreated. Statistical matching involves pairing to each treated unit a non-treated unit with the ÔsameÕ observable characteristics, so that (under some assumptions) the outcome experienced by the matched pool of non-treated may be taken as the outcome the treated units would have experienced had they not been treated. Alternatively, one can associate to each treated unit a matched outcome given by the average of the outcome of all the untreated units, where each of their contributions can be weighted according to their 'distance' to the treated unit under consideration. An interesting quantity which avoids the dimensionality problem is the Ôpropensity scoreÕ, the conditional probability of being treated. psmatch implements various types of propensity score matching estimators: from one-to-one matching with replacement (optionally within a caliper) to a number of smoothed versions (including nearest neighbours, kernel, local linear regression). Additionally, it allows to implement Mahalanobis metric matching on two or three variables. Other options include estimation of the propensity score, bootstrapping of the treatment effect, the creation of matching quality indicators for a set of specified variables and producing a smoothed outcome for the treated as well. The software (version 2.0) was revised in August 2001. The current version is psmatch2 of Leuven and Sianesi.