542 – Nonresponse Issues
A Second Source of Conservatism in Fisher's Exact Test for Rare Events
Mark Von Tress
Alcon Laboratories
Fisher's exact test (FET) is often recommended for small numbers of events of concern in a comparison of proportions, or when the asymptotic assumptions of chi-square or other z tests are not met. This test controls the type I error conservatively since the size of the critical region under the hypothesis may be smaller than the nominal type I error. A second source of conservativism is discussed where the size of the critical region under the alternative in the final observed table may be substantially smaller than the nominal power. This affects the design of a single study because we observe only one 2x2 table instead of many. This paper recommends examining the probability of having less than the nominal power and the variance of the conditional power function prior to study initiation to determine if inverse sampling may provide an advantage over standard sampling.