Consider a government contemplating the implementation of a training program. The decision to implement the program depends on the assessment of its likely effectiveness. Often the policy maker has available data from a previous implementation of the same program to inform this decision. There are two steps involved in exploiting data from previous training programs in predicting the effectiveness of the new program. First is the problem of using the data from the previous program to evaluate the effectiveness of that specific program. The second step is to generalize the results obtained for the old program to the new program. The current proposal focusses on the second step. The investigator discusses how this problem is related to the standard evaluation problem, and how propensity score methods can be useful. He proposes developing extensions of these methods that were developed for the binary treatment case to allow for multivalued and continuous treatments. He also intends to apply these methods to two different data sets. In both cases randomized assignment was used so unbiased estimates of the true average effect are available to judge the performance of estimators based on the proposed methodology.
