New asymptotic inferences about the difference, ratio, and linear combination of two independent proportions

摘要

In statistics it is customary to realize asymptotic inferences about the difference d, the ratio R or a linear combination L of two independent proportions. In this article the authors evaluate ten inference methods, and conclude that for = 1% some of the new procedures behave better than the classical. In cases d, R, or L, the optimal method consists in adding 0.5, 0.5 or 1 to all the data, respectively, and then applying a modification of the arc-sine transformation (d or R) or the likelihood ratio test (L). A free program may be obtained at ULR http://www.ugr.es/local/bioest/Z_LINEAR_K.EXE.