Selecting an interim analysis procedure.

Motivations for undertaking interim analyses differ, as do the methods proposed by different authors. This article evaluates five interim analysis procedures with regard to different requirements. The five procedures differ with respect to concern for the presence or absence of a true treatment effect. One provides interim criteria only for accepting Ho (terminating due to insufficient evidence of a true treatment effect), two provide criteria only for rejecting H(o), and the others provide criteria for either accepting or rejecting H(o). A computer program was developed to simulate applications of the interim analyses to sampling data. Actual Type I error probabilities, power, probabilities of early termination, and expected sample sizes resulting from the different interim analysis procedures are compared. One-sided and two-sided tests, equal and unequal interim sample segments, and interim alterations in the sample size or research design are considered. Results should be helpful in selecting a method that satisfies particular interests.