|
Abstract:
|
Factorial designs have been widely used in clinical research, especially for screening multi-factor interventions in early phases of trials. In this talk new tests for identifying main effects in factorial designs are proposed where the test statistics are weighted combination of simple effects. In the presence of interactions, the weighted tests are shown to be uniformly more powerful than traditional tests based on the average of simple effects. Furthermore, unequal sample size allocations are investigated to derive optimal design that achieve adequate power for a predetermined subset of tests for main effects. The computational burden is low because the optimal choice of the sample sizes under the power constraints can be formulated as a simple convex optimization problem. The new methods are illustrated by some real examples.
|
