This dataset was created to be similar to a real 305-patient two-treatment randomized clinical trial for treatment of depression, by perturbing the depression scales in the original raw data. The outcome scale is the Hamilton-D depression scale, which is also measured at baseline.

The trial and this perturbed dataset represent a case where there is a strong nonlinear relationship between baseline and follow-up scales. The actual trial, like so many others, used an erroneous change from baseline as the outcome variable. With a nonlinear effect of baseline, the change from baseline relates strongly to baseline and the overall mean change is not interpretable.

A proper reanalysis appears here.