Kaigang Li posted on Thursday, July 03, 2008 - 7:16 am
Hello Prof. Muthen,
If I test a linear growth model for a categorical outcome with time-invariant and time-varying covariates, is there any way to test the parallel regression assumption/proportional odds assumption in Mplus?
No, there is not such a test in Mplus. You can read about this in the Sage (white series) book by Long (1997), Regression models for categorical and limited dependent variables, pp. 140-145. His discussion includes the simple checking of a similar slope for a set of binary regressions ("An informal test", p. 141).
Kaigang Li posted on Friday, July 04, 2008 - 1:45 pm
Thanks, I will read that book. Following that question, I have another one. It seems that the nominal (unordered) variable cannot be the outcome variables in the growth model when I read the CHAPTER 6 of Mplus manual, is that right?
So if the proportional odds assumption for ordinal outcomes is violated according to the Long's informal test, what can I do in terms of the growth model?
I have not seen growth modeling done with a nominal outcome - it doesn't seem like a natural fit. Perhaps you want to switch to latent transition analysis if you are concerned with the proportional odds specification.
Kaigang Li posted on Sunday, July 06, 2008 - 1:48 pm