SELECTING AN INTERIM ANALYSIS PROCEDURE

Motivations for undertaking interim analyses differ, as do the methods proposed by different authors. This article evaluates five interim an alysis procedures with regard to different requirements. The five proc edures differ with respect to concern for the presence or absence of a true treatment effect. One provides interim criteria only for accepti ng H(o) (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 pr ogram was developed to simulate applications of the interim analyses t o sampling data. Actual Type I error probabilities, power, probabiliti es 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 al terations in the sample size or research design are considered. Result s should be helpful in selecting a method that satisfies particular in terests.