I'm using McNemar's test. Basically this question is about best practices when reporting results using McNemar's test.

I want to report the effect size. What is a sensible effect size for McNemar's test? I've seen the odds ratio b/c and the proportions b/(b+c) and c/(b+c) both used in papers. If I say what b and c are then all possible effect sizes can be computed. However I haven't seen this, is it bad form?

Thanks!

**Contents**hide

#### Best Answer

In general, I think best practice for presenting measures of effect size depends on the question of interest and the usual practice in your field. There's little point reporting an effect measure that readers will be unfamiliar with.

Having said that, in this particular case if you want a *single* effect measure I think the (conditional) odds ratio *b* / *c* is the obvious choice. Giving both proportions *b* / (*b* + *c*) and *c* / (*b* + *c*) gives more information and clearly the odds ratio can be derived from these two. I wouldn't call two proportions an 'effect size' though.

I'd strongly encourage you to report *all four* cells of the 2×2 table concordancy table, however. That way nothing is hidden and anyone can calculate whatever statistics they wish. It's only four numbers, after all.

### Similar Posts:

- Solved – Can Cramér’s V be used an effect size measure for McNemar’s test
- Solved – McNemar’s test vs chi-squared test for before & after data with different people from the same population
- Solved – McNemar Test: Reporting Chi-Square
- Solved – McNemar Test: Reporting Chi-Square
- Solved – McNemar Test: Reporting Chi-Square