Post-hoc in ordinal logistic fits

[#1]

Hi There,

I am running an ordinal logistic model to find differences in response variable giving three locations and positions within those locations (nested by position). The model compares by default positions and locations ‘3’ vs ‘2 and 1’. I am trying to figure out if 2 is different from 1 and I am using the 95% confidence intervals to do so. Is this the right approach or is there another way to do a post hoc on ordinal logistic fits?. Also, what does it means that there are no confidence intervals reported for some paired comparisons (see below).

Thank you!

 

[#2 · Accepted solution]

How about creating a null model, in which the levels that you suspect indifferent should have the same coefficient? To do that, you need to create a new table, in which desired levels should be coded the same, e.g. replace 2 by 1. Then, compare the null model and the alternative, using likelihood, number of parameters, sample size. You may construct either a Chi-square test or information criteria to compare two models.
For missing intervals, a long shot is that you don't have enough data.