I'm doing an outcome study for a form of therapy in which I have pre-post data for 48 participants and pre, post and follow-up data for 10 ps. The outcomes are all measured on a yes/no basis so it's nominal data. I think I need to use the mcnemar test to analyse pre-post or post-FU but are there any assumptions about expected or observed frequencies in each contingency table cell that I could violate given the small sample size?
I ask because SPSS seems very sporadic in whether it will run the analysis correctly or not and I can't work out why - for some of them it seems like it's because the expected frequency counts are less than 2, but this isn't consistent so I'm really confused.
Any help greatly appreciated.
p.s. Any expertise on whether a Cochran's Q would be ok to run for a pre-post-FU analysis on the 10 ps would also be great!
Most sources state that the McNemar test (asymptotically - chi-square distribution) should not be employed with extremely small sample sizes. But sources do not agree on the minimum acceptable sample. When the sample size is small, the exact binomial probability for the data should be computed.
You can find the exact McNemar StatXact software, but I think, that it shouldn’t be a problem if you calculate it using asymptotic McNemar.