Use of analysis of covariance in clinical trials: a clarification

Criteria are discussed for using analysis of covariance (ANCOVA) and t-tests in analysis of clinical trials. Revisions are suggested in the guidelines given recently by Egger et al. (Controlled Clin Trials 6:12-24, 1985). Specifically, if one assumes that treatment groups have been successfully balanced by randomization, then ANCOVA is more powerful than the t-test under a broad range of conditions; this is true even in the model assumed by the t-test, and regardless of the presence of measurement error. If the balance cannot be assumed, then under certain conditions the t-test should be preferred over ANCOVA on grounds of validity.