Partitioning a Control Group to Achieve Appropriate Comparison Subgroups for Assessing Program Impacts on Completers and Noncompleters.

In evaluation research studies, it often occurs that several program participants (experimentals) drop out of the program prior to completion. Since noncompleters generally differ substantially from completers in many respects, a control group which originally was representative of the participant group will most likely not be representative of either the completers or noncompleters considered separately. This paper presents a maximum likelihood procedure for partitioning a control group in such a situation into separate comparison subgroups for assessing program impacts on completers and noncompleters. The approach was used in evaluating the Mountain Plains Career Education Program. (Author)