Predicted outcomes in a national sample: A source of comparison for program evaluation

The lack of comparison groups is a common problem for researchers conducting evaluations of small, specialized programs. When conventional comparison groups are not available, the authors suggest a possible alternative approach which involves: (a) locating a sub-group in a national study for whom relevant variables have been measured; (b) generating a statistical model which predicts the outcome of interest in the national data; and (c) applying the predictive model to program data, in order to estimate outcomes in the absence of progam intervention. The authors illustrate this new application by presenting the results of an analysis of data from Project Talent, a national manpower study, and discuss its application to the evaluation of a medical training program in New York City.