I faced a problem with non-symmetric (non k x k) contingency tables in SPSS. I have categorical variables for a certain pathology (theoretically scored 0-3), and I want to compare the scores of region A with those of region B in a set of subjects. The samples are related (though it is not a repeated test) because A and B regions are present and scored for each subject. In theory, related-samples test for non-dichotomous categorical variables (i.e. larger than 2x2 contingency tables = k x k) can be done by the McNemar-Bowker test in SPSS (an extension of the McNemar test for 2x2 tables), which is fine. However, if e.g. a pathology is so frequently severe in region A that no 0 score (or even no 1) is given while region B has the complete spectrum of scores (0-3) then we face an asymmetric contingency table (e.g. 3x4 or 2x4) for which the McNemar-Bowker test fails and gives no result. Does anyone has a suggestion which test is appropriate in such scenario? Many thanks!
- Badges
- Science method