I have baseline and follow up group, they are same people (paired) but in each group I have three categories, so to test the changes I cannot apply MacNemar test since it works for 2x2
I presume you mean 3 x 3!
When we extend the McNemar test, there are two hypotheses we can test. The first is asymmetry - that the off-diagonal values are asymmetric. The second is marginal homogenity - a test that the frequencies along the table margins are similar.
In genetics, the test is called transmission/disequilibrium test (TDT) and is used to test the association between transmitted and nontransmitted parental marker alleles to an affected child. I don't know if SPSS does the test. For sure, Stata and R will do it.
I wonder if an ordinal logistic regression might be neater? You have paired data so the differences are -1 0 or 1 leading to an ordinal DV. You then have a categorical predictor with three levels that can be used to predict the outcome by coding them as dummy variables (for example).
I think you could also set this up as a multilevel ordinal model.
Thom has a point. It is critically dependent on your hypothesis, isn't it? There are a number of 'answers' that can be calculated from your data, but which one of them is really an answer depends on your question!
And, of course, a statistic that isn't the answer to a question is not a statistic, no more than a stick that doesn't come back is a boomerang.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics