In the examples considered so far in this chapter, the mixed models approach has fitted the patient effects as random. As we have seen earlier, this implies that the observations within one patient are all assumed to have the same correlation and variance. However, it could be argued that in trials with three or more periods the correlation may vary with different pairs of periods. In particular, periods which were closer together might be expected to show higher correlations. In this section we explore the situation where the covariance patterns used are 'structured'.
Was this article helpful?