I want to measure whether the binary responses (0 and 1 - disagree or agree) are different among four different statements - i.e. whether certain statements are significantly more likely to be agreed with. It is the same people answering each question, but, due to the fact the statements are different each time, I am not sure if this qualifies them as being treated as "paired" (as opposed to seeing if their responses changed on the same statement).

If they are paired, then I thought that Cochran's Q test may be appropriate? If not, does anyone know which test may be? I keep getting pointed to Chi Squared but it is only an outcome I am measuring and not a factor.

I have four SPSS columns, an n of 80, hence each column has 80 0s or 1s. Thank you!


1 Answer 1


Yes, Cochran's Q is a correct test for this situation. You can use post-hoc McNemar tests to conduct multiple comparisons among questions.

The data are "paired" in the sense that the same person is rating each of the statements. Probably more useful is to realize that what you really want to look at the marginal differences among the statements. That is, to determine if Statement A is more popular than Statement B, you need to compare the count of (agree with A, disagree with B) to (disagree with A, agree with B). If someones agrees with both A and B, then their vote doesn't help to decide which is more popular. This is exactly what McNemar's test does.

I don't know if there is any kind of limit to the number of statements you could compare with Cochran's Q as an omnibus test.

The following link has an example of Cochran's Q test used in a similar circumstance. The example uses R, and the post-hoc separation of the practices is conducted with pairwise McNemar tests. R Handbook: Cochran’s Q Test for Paired Nominal Data.

  • $\begingroup$ Thank you! So much. I always envisaged paired data to be a measurement of individuals on the same outcome, so I am correct in thinking this is still paired, despite the outcomes being different (i.e. 4 different statements)? Finally, do you know how many different statements the test could handle? i.e. if I wanted to test whether respondents' answers were significantly different across 11 different statements could a Cochran's Q handle that? Really appreciate your help. $\endgroup$
    – carley
    Commented Jul 27, 2017 at 18:23
  • $\begingroup$ I'll try to add answers to these questions to the main answer. $\endgroup$ Commented Jul 27, 2017 at 18:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.