ABSTRACT
Statistical significance tests investigate statistical significance. It is also important to investigate practical significance, which for medical data, is referred to as clinical significance. Edwards’ remarkably simple proof is the basis for a logical approach to contingency table analysis, based upon the dual concepts of measures of association and log-linear models. When developing appropriate statistical procedures, a general result for log-contrasts of multinomial cell frequencies is required. In the product binomial situation, the two-sided version of Fisher’s exact test has the best finite-sample power properties available, when compared with any other test procedure. Quite interestingly, Lehmann provides an expression permitting the calculation of the power function of Fisher’s exact test, when all marginal totals are indeed fixed, in terms of the cross-product ratio measure of association λ.
