Based on your description, ordinal regression should work

Hello Krishna,

Besides ordinal regression, here are three other possibilties:

1. For an omnibus test of difference across the five occasions, consider the Friedman two-way anova by ranks.

2. If you wanted to know whether some pair of occasions (e.g., Visit 1, Visit 2) showed changes, a test like McNemar's method (a "statistics..." option in crosstabs within spss). Technically, this is Bowker's extension of the McNemar test.

What the test does is examine the instances in which the paired scores disagree, and evaluate whether the direction of disagreement is equal, or more likely to be in one direction than another.

3. If you want to compare scores from some pair of occasions for equality (as in #2), you could also use the Wilcoxon signed ranks sum test.

Good luck with your work.

The longitudinal aspect can be dealt with using multilevel software, and the ordinal aspect can be dealt with for that also (and it varies by software how). What are wanting to see about the changes across times? With just three categories there and each of the 10 comparisons (5 choose 20) is a square table, you can also look at the sociologist who have done a lot of work on transitions between discrete states. Other fields are also using this methods. I think with just three levels I would not use methods that treat ranks as the variable of interest, but this differs from David Morse 's advice, so likely both approaches will be acceptable to most.