Treatment Implementation, Statistical Power, and Internal Validity

Frequently program participants receive different levels of exposure to a given treatment. Cook and Poole (1982) recently argued that data on treatment implementation should be included in the analysis of treatment effects to increase power. It is argued that basing inferences about treatment effects on analyses including level of treatment implementa tion has a serious potential problem: Such an analysis is likely to result in biased estimates of treatment effects. Analyses including the level of treatment implementation must be interpreted cautiously unless the causes of differential exposure to the treatment are completely understood.