| This thesis explores the application of fully sequential methods for the analysis of longitudinal clinical trial data. A new nonparametric approach will be developed, using sequential ranks, for the comparison of several treatment groups. Sequential ranking is an alternative to ranking by the usual method. Although sequential ranks are more likely to suffer from information loss than regular ranks, they are preferred here for their independence.;The performance of the three monitoring procedures will be assessed via a Monte Carlo simulation study. In particular, we will compare power and average stopping time for various treatment differences, different numbers of treatment groups, and different response distributions. The procedure will then be applied to data arising from an orthodontic clinical trial.;We will develop three alternative monitoring procedures. The first two will be large-sample, continuous analogues of the Pocock and O'Brien-Fleming group sequential monitoring procedures. The third procedure, a small sample version, will make use of the sign function, and will be grounded in the theory of simple random walks. |
