# Can Cochran's Q test take a 3 level nominal variable as the response (instead of 2)?

I conducted a study using within-subjects (repeated measure) design for collecting nominal data. I have been reading a lot closet I get for a suitable test is Cochran's Q test since I have repeated measures for nominal data from the 3 related-samples. However, my responses for the dependant variables has 3 categories, and Cochran's Q test takes only 2 (e.g. yes, no)!

McNemar-Bowker Test can take 3 or more categories for the dependent variable, but it is only for 2 related samples (i.e. before and after), and I have 3 related samples.

So, is there any test similar to Cochran's Q test (3+ related samples) but taking more than 2 nominal responses for these dependent variables?

• Yours is a good question, to my mind, but not yet formulated clearer. Less technically - more substantially. What do you want to test, actually? Do you want something like extension (on 3+ related samples) of McNemar-Bowker test or maybe of Cohen's kappa test (maybe with glance at a their comparison), or something else? – ttnphns Mar 17 '16 at 21:17
• Yes, I want something like extension (on 3+ related samples) of McNemar-Bowker test – Fatma S Mar 18 '16 at 15:30
• How do you define what are "symmetric cells" in a cubic table? McNemar-Bowker should test for symmetry. Perhaps what you really need is marginal homogeneity test for k related samples?? – ttnphns Mar 18 '16 at 17:36
• @ttnphns I want to know whether there is a significance difference between the 3+ related samples's responses on the repeated measure. Does marginal homogeneity test for k related samples differ from McNemar-Bowker test? and how can I use it please? – Fatma S Mar 23 '16 at 15:10
• When the number of categories (responses) c is >2, marginal homogeneity test is different from McNemar (Bowker) test. Such test for k=2 samples exist in Nonparametrics in SPSS. Unfortunately, for any k>2 it is absent in SPSS and I don't know where to seek it. – ttnphns Mar 24 '16 at 6:33