IPW estimation and related estimators for evaluation of active labor market policies in a dynamic setting
This paper considers treatment evaluation in a discrete time setting in which treatment could start at any point in time. A typical application is an active labor market policy program which could start after any elapsed unemployment duration. It is shown that various average effects on survival time are identified under unconfoundedness and no-anticipation and inverse probability weighting (IPW) estimators are provided for these effects. The estimators are applied to a Swedish work practice program. The IPW estimator is compared with related estimators. One conclusion is that the matching estimator proposed by Fredriksson and Johansson (2008) overlooks a selective censoring problem.
Stata software for dynamic IPW estimation (contact Johan Vikström)