Although structuring by period is the most obvious way of introducing structure, this can also be applied to treatments. In parallel group trials we have already met situations where we might wish to fit separate variances for each of the treatment groups. There is an exact analogy in the cross-over situation, where the variances may differ for some of the treatments. Additionally, there may be good reason to suspect that the results from certain pairs of treatments may be more highly correlated than others if they have a similar mode of action. Thus, a more complicated structure for treatments than the simple compound symmetry may be highly plausible. This type of approach has been found to be particularly useful in the analysis of bioequivalence trials where treatment differences in reproducibility are highly relevant; examples of its application are available on the FDA website (www.fda.gov). We also present an example in Section 8.15.
Was this article helpful?