# Two tails or one tail McNemar's test in a binary classification problem. Which one should i choose?

I'm using McNemar's test for compare two designed models in a binary classification problem. As you know we have two kinds of this test. One tail and two tails. What is difference of these kinds and which one should i choose for calculating my p-value? I'm using Yates's correction for continuity for this test.

Thanks.

• I suggest considering the Wilcoxon signed ranks test instead of McNemars. See this paper. – Marc Claesen Oct 4 '14 at 1:27
• @Marc, in a dichotomous case Wilcoxon test = Sign test =McNemar's test. – ttnphns Oct 4 '14 at 6:38
• Oh that's embarassing. Thanks for that clarification @ttnphns. – Marc Claesen Oct 4 '14 at 7:53