Conditional change ordered logistic with different approach of same variable


New Member
I am working with two-wave panel data looking at outcomes of retirement on self-rated health (5 point scale, very good to very bad)). It is questionable to interpret this variable as interval/ratio, therefore I use ordered logistic regression (ologit in STATA). The results show significant cut-off points, indicating that this approach is appropriate. At the same time, because of the two waves, I employ a conditional change model, thus I include self-rated health at t1 (SRHt1) as a predictor for self-rated health t2 (SRHt2). Logic would then say that, since I argue the dependent (SRHt2) is not on interval-level, I should include SRHt1 as dummies. This is still not really a problem, but I also want to test for an interaction with SRHt1, and this is where it becomes cumbersome. I would have to include four (5 - 1 reference category) interactions, making for a large table and difficult interpretations. Rather, I'd use the interval version of SRHt1 for the interaction (the interaction is not significant, no matter the approach). I realize this is inconsistent, but I want to know whether mathematically this is permissible.