ABSTRACT
During the past decade, the approach for sample size determination originating from Bayesian’s point of view has received much attention from academia, industry, and government. Although there are still debates between frequentist and Bayesian, Berger, Boukai and Wang (1997, 1999) have successfully reconciled the merits from both frequentist and Bayesian approaches. However, no specific discussions regarding sample size determination for clinical trials at the planning stage are provided from Bayesian’s point of view. Their work has stimulated research on sample size calculation using Bayesian’s approach thereafter. The increasing popularity of sample size calculation using Bayesian’s approach may be due to the following reason. The traditional sample size calculation based on the concept of frequencist assumes that the values of the true parameters under the alternative hypothesis are known. This is a strong assumption that can never be true in reality. In practice, these parameters are usually unknown and hence have to be estimated based on limited data from a pilot study. This raises an important question: how to control the uncertainty of the parameter from the pilot study (Wang, Chow, and Chen, 2005). Note that the relatively small pilot study may not be the only source of the parameter uncertainty. In some situations, the magnitude of the non-centrality parameter may be obtained simply from subjective clinical opinions (Spiegelhalter and Freedman, 1986). In such a situation, the true parameter specification uncertainty seems to be even severe. Some related works can be found in Joseph and Be´lisle (1997), Joseph, Wolfson and du Berger (1995), Lindley (1997), and Pham-Gia (1997).
