Assessing differential drug effect.

The regression effect gives a misleading impression of the relationship between drug or treatment effect and baseline measurement. We propose a method of adjusting for the regression effect; we also suggest a corresponding test for the differential drug effect. The likelihood ratio test for no differential drug effect is equivalent to a test for equality of variances, suggested by Pitman (1939, Biometrika 31, 9-12) and Morgan (1939, Biometrika 31, 13-19). The proposed adjustment and test are applied to an example of blood pressure data.