0
$\begingroup$

I'm running model in which I analyze salary of recent graduates. People graduated from different majors and in different years. The dependent variable (salary) is measured using intervals, e.g., "less than 1000", "between 1000 - 2000; "between 2000 - 5000", not the actual amount that people make. Because of the fact that data is hierarchical and the DV is ordinal, I use the Cumulative Link Mixed Model (CLMM) in the ordinal package in R.

The important assumption here is the proportional odds assumption. Unfortunately, I didn't find information on how to test this assumption in multilevel models and from my reading, it is not possible to do in the ordinal package.

That's why I have the following questions:

1) Does it make sense to test the proportional odds assumption using cumulative link model (CLM) which does not take into account the hierarchical structure of the data I have?

2a) if this is not a correct approach, what would you recommend to use as an alternative?

2b) if such a testing makes sense (and the assumption is not met), what should be the alternative model? Is it the multinomial logistic regression model(with the random effects specification)? Could you recommend a package to do it for multilevel data?

Thank you!

$\endgroup$
0
$\begingroup$

1) Obviously it does not make any sense to test the proportional odds assumption using cumulative link model disregarding the hierarchical structure of the data.

2) One valuable option is the multinom function from nnet package. MCMCglm package is also good and easy to implement despite its Bayes in approach.

$\endgroup$
  • $\begingroup$ Thank you Paolo! In fact, this idea of testing the assumption of proportional odds without specifying the random effects have been suggested for example here rcompanion.org/handbook/G_12.html (section "Check model assumptions"). However, it doesn't "sound" right to me. Could you elaborate on this why you also think that this is not a correct approach? $\endgroup$ – Gabriela Nov 12 '18 at 22:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.