Abstract:
|
In a crossover design, including period-specific baselines as covariates in an analysis of covariance (ANCOVA) is known to increase the precision of the estimated treatment effect. The potential efficiency gain depends on the joint covariance structure of the baselines and post-treatment responses, as well as the metric used to incorporate the baselines. Here we examine improvements in power that can be achieved by choosing an optimal linear combination of baselines (LCB) that minimizes the variance of the ANCOVA-based estimate of the treatment effect. Our work is relevant to balanced designs with up to four periods, specifically the 2x2, 3x3 and 4x4; with a natural extension to incomplete block designs, such as the 2-period 3-treatment design. Since the optimal LCB is a function of the covariance structure, which in practice is unknown, we propose an adaptive method in which first a suitable covariance structure for the given dataset is selected via AICC values, and then the corresponding optimal LCB is used in the ANCOVA. Relative to previously published methods, the proposed method leads to sizable gains in power, while maintaining the nominal type I error rate.
|