Baseline correction in a two-way randomized blocks design

Randomized blocks designs are used in clinical psychopharmacology research to test the hypothesis that relative effectiveness of different drug treatments depends on the type of patient being treated. Monte Carlo methods were used to evaluate the adequacies of different approaches to statistical control over baseline differences in such a design with special concern for the treatments x blocks interaction effect. Given the usual assumptions, analysis of covariance (ANCOVA) was shown to provide adequate baseline correction with consequent unbiased tests for the treatments x blocks interaction, as well as the main effect for the randomized treatments. Tests relying on simple pre-post difference scores and percentage change scores evidenced seriously conservative or nonconservative bias in the test for treatments x blocks interaction effect, a bias that depended on the direction of the corresponding interaction effect observed in the baseline measurements alone. This is discussed as a serious matter because the interaction effect in a randomized blocks design is a common basis for the claim of specific indications of drug treatments for particular types of patients.