Lowering the Costs of Program Evaluation

During this period of reductions in government social programs, there is great value in developing program evaluation methods that are low cost and timely. This article explores the use of past evaluation data to develop statistical models useful in predicting future program performance. Model data requirements are much lower than those of evaluations, and models can yield policy results much more quickly. The modeling approach is used to develop a low-cost monitoring scheme for the Bonneville Power Administration Residential Weatherization Program. Three years of evaluation data (1981, 1982, and 1983) were utilized. The results suggest that the modeling approach is most effective when program characteristics and actors, and the external environment are stable over time. In our application, modeling results changed between 1982 and 1983. The article concludes with a discussion about the defensibility ofpredictions yielded by streamlined evaluations.