Bayesian sample size calculations for hypothesis testing

A Bayesian approach to sample size calculation in hypothesis testing problems is developed. The sample size is chosen to make it a priori probable that the Bayes factor is greater than a given cut-off of prespecified size. Two methods for choosing a cut-off are given: an absolute criterion and a relative criterion. Calculations can be done using either exact algebraic manipulation or through simulation. The approach permits the propagation of uncertainty in quantities which are unknown and permits the computation of power and type I error rates either conditionally or unconditionally on particular values of the parameter of interest. A graphical tool is given for assessing the sensitivity of the predicted outcomes to model and sample size specification. The approach is illustrated for a one-sided and for a two-sided alternative hypothesis for continuous data with a normal prior.