Extended Abstract. When a new treatment is being considered, trials are carried out to estimate the increase in performance which is likely to result if the new treatment were to replace the treatment in current use. Many authors have looked at this problem and many procedures have been introduced to solve it. An important feature of the analysis in this work is that account is taken of the fact that only if it turns out that the new treatment, in a statistical sense, is clearly better than the previous treatment will the number of subsequent users of the new treatment be high.

Traditional classical methods of selecting sample sizes are based on the required size, and the required power of the test for a specified treatment effect. The most frequently used sample size formulae arises from the relationship between the standard error of the estimator of the parameter of interest and the sample size.

Since the formulae can be highly sensitive to the choice of inputs, careful selection of the parameter estimates and target criteria are essential steps in determining the sample size. Classical or frequentist methods are unable to take into account uncertainty in point specifications. Bayesian methods are ideally suited for design since they provide a tool for specifying uncertainty, and how it changes in response to further information.

In this work we figure out the optimal size of...[To continue please click here]