Abstract
Statistical analysis and considerations of power in the programs evaluation through two-stage sam pling designs. Evaluation of prevention programs involves decision errors due to the difficulty of randomly assigning individuals to research conditions. The nature of the intervention and design of the study determine the choice of the appropriate unit of analysis in impact assessments. When units of assignment and units of observation differ, that is, when clusters of people rather individuals are assigned at random to treatments, the analyses conducted at lower levels of the study hierarchy provide inefficient parameter estimates, and often result in inappropriate significance tests. Therefore, the purpose of this paper is (a) to present an analytical method that permits the use of data at all levels of design without increasing Type I error rates, and (b) to determine the number of clusters and the sample size per group according to variability and cost.
