The overall treatment effect estimates obtained from Models 1 and 6 are summarised in Table 6.2. The small differences arise from differences in the distribution of missing values between treatments. Thus in this dataset selecting the most appropriate covariance pattern has had little effect on the results. However, this is not always the case. In situations where the variance differs between the visits, standard errors can differ more markedly.
In the last row of Table 6.2, results obtained after omitting the treatment.time effects are given. The treatment effects are then calculated as weighted averages of the effects at each time point. This contrasts with the models fitting the treatment-time interaction, which causes treatment effects to be unweighted averages of the effects at each time point. There is little difference between the two sets of results, because there is only a relatively small amount of missing data. It is a matter of personal choice as to which approach is preferable. There are noticeable differences in the treatment effect estimates between the visits. However, there is an absence of any coherent pattern and the treatment-time interaction was not significant in any model.
Was this article helpful?