When programs benefit some people more than others: tests of differential service effectiveness.

Practitioners need to know for whom evidence-based services are most or least effective, but few services research studies provide this information. Using data from a randomized controlled comparison of supported employment findings for two multi-service psychiatric rehabilitation programs, we illustrate and compare procedures for measuring program-by-client characteristic interactions depicting differential program effectiveness, and then illustrate how a significant program-by-client interaction can explain overall program differences in service effectiveness. Interaction analyses based on cluster analysis-identified sample subgroups appear to provide statistically powerful and meaningful hypothesis tests that can aid in the interpretation of main effect findings and help to refine program theory.