This monograph critiques the many nonexperimental impact estimation approaches that have been based on external comparison groups. It proposes an approach to evaluating employment and training (E&T) programs that calls for using the group of individuals who apply to a program but then choose not to participate in that program as an "internal" comparison group. The following are among the topics are discussed: methods used to evaluate E&T programs in the past; the case for applicant-based comparison groups (desired qualities in a nonexperimental estimation technique, potential nonparticipant comparison groups, the history of applicant-based comparison groups, and the regression discontinuity model for screen-out-based impact analysis); estimating program effects in the Aid to Families with Dependent Children Homemaker-Home Health Aide Demonstrations; testing alternative estimates for selection bias (classical and Bayesian approaches to model validation); and summary and recommendations (the rationale for applicant-based comparison groups, empirical estimates using alternative applicant-based comparison groups, and specifying validation tests). Appended is information about the following: construction of grouped data, regression procedures, tests for proven bias in nonexperimental impact estimates in the study of voluntary versus mandatory programs, and upper and lower bounds on the risk function. Contains 24 tables/figures and 141 references. (MN)