# Ordered logistic multilevel regression in imputed dataset

Is there any way to calculate an ordered logistic multilevel regression in an imputed dataset? I have tried Stata’s GLLAMM (Generalized Linear and Latent Mixed Models), but that won’t work with the mi estimate, post:command prefix. Should I “estimate anyway”, or is there a better way to do it, preferably R or Stata?

> mi estimate, post : gllamm q18 [...] , link(ologit) f(binom) i(country) eform

mi estimate: command not supported
gllamm is not officially supported by mi estimate; see mi estimation for a list of
Stata estimation commands that are supported by mi estimate.  You can use option
cmdok to allow estimation anyway.


Also, suppose my multilevel model has only two levels, i.e. workers nested in countries. IIRC, in the case of linear OLS regression, such multilevel model would be identical to a ordinary model that includes country dummies. If I use that workaround with ordered logistic regression, will my results be biased?

• can you please confirm that it worked? – StasK Dec 19 '12 at 15:28

I don't see any reason why cmdok should not work. Admittedly, gllamm is a complicated estimation command, but I don't see anything wrong with letting it run in the context of mi (provided, of course, that you've done your imputations correctly accounting for the multilevel nature of your data).
The dummy variable interpretation of the fixed effect regression model does way more harm than good. You are not the first one trying to push it to other models, but the analogy only works in linear models. There are two more distributions that allow the fixed effects conditional estimation, binomial (xtlogit) and Poisson (xtpoisson), and that's the end of the fixed effects world. You can introduce the country dummies, of course, but it won't give you a fixed effects model, just the model with each country having its own intercept in the linear prediction xb.