Two-tailed unconditional inferences on the difference of two proportions in cross-sectional studies

Unconditional inferences about the difference d = p(2) - p(1) between two independent proportions are usually carried out under the double binomial model. However, when the results proceed from a cross-sectional study, it is more appropriate ( and exact) to carry them out under the multinomial model. These inferences have been developed in the case of tests and confidence intervals with one tail, but the case of two tails is customarily solved using the TOST ( two one-sided test) method. In this article, the exact and asymptotic two-tailed inferences are developed using two real two-tailed tests. The program for carrying them out may be obtained without charge from: http://www.ugr.es/local/bioest/EQUIV_ASO.EXE.