Improvements in power obtained from stepwise testing methods and incorporating correlation structures are well known and documented (see e.g., Hochberg and Tamhane, 1987) but power gains from discreteness are less well known. Multiple comparisons of dependent proportions can be made more powerful by utilizing stepwise testing methods, incorporating specific discreteness characteristics of the exact tests, and incorporating dependence structures. Existing solutions have not taken advantage of recent developments in multiple testing: when multiple tests have been considered at all, they have typically used simple Bonferroni or Scheffé-style methods, or have not taken advantage of discreteness in the distributions ( Bhapkar and Somes, 1976, Rabinowitz and Betensky, 2000 Kitajima et al., 2009). 2005 Demšar, 2006).ĭespite the wealth of applications involving comparing multiple dependent proportions, it is surprising that the problem of multiple comparisons of dependent proportions has been so little studied in the literature. The test is also used to compare classification rates (sensitivity, specificity, and overall) among multiple predictive models, such as those for predicting prostate cancer from diagnostics tests and patient characteristics ( Leisenring, Alono and Pepe, 2000 Durkalski et al., 2003 Lyles et al. Applications related to McNemar's test abound in recent issues of Biometrics and elsewhere, for example in the evaluation of safety and efficacy data in clinical trials ( Klingenberg and Agresti, 2006 Klingenberg et al., 2009). McNemar's test is used to compare dependent proportions.

Xlstat multivariate comparison