ABSTRACT
Many experiments consist of a series of independent observations on subjects or units that, for one reason or another, enter the experiment for treatment sporadically. Often a fixed sample size is agreed on as the minimum necessary to make an inference or decision concerning the treatment. In a frequentist context, the agreed-on sample size will generally depend on the power and size of the test of the hypothesis. This results in the application of a test at a certain a level. In the Bayesian context, although a predetermined sample size is not a determining constituent for computing the posterior odds of one hypothesis versus another, planning for the costs and the administration of an experiment may lead to a determination of sample size prior to embarking on a trial. In any event where experimental procedures are costly, it is of great interest for the investigator to know whether to continue testing a new treatment, drug, or therapy after partially completing an experiment.
