Multinomial repeated logistic regression

Welcome to the forum for MLwiN users. Feel free to post your question about MLwiN software here. The Centre for Multilevel Modelling take no responsibility for the accuracy of these posts, we are unable to monitor them closely. Do go ahead and post your question and thank you in advance if you find the time to post any answers!

Remember to check out our extensive software FAQs which may answer your question: http://www.bristol.ac.uk/cmm/software/s ... port-faqs/
Post Reply
shakespeare
Posts: 70
Joined: Thu Feb 14, 2013 11:12 pm

Multinomial repeated logistic regression

Post by shakespeare »

I have a longitudinal data set that is comprised of a set of physician prescription patterns (three unordered categories). Within each physician there are two assessments, pre- and post-test. At the pre- and post-test, there are 3-140 responses for each physician. So I have these measurements nested in times, and times nested in physicians. There are several educational groups that form an IV. It's not really the type of data I was expecting (normally one gets one measurement at each time period), but it's what I was given and I'm wondering if there's a way to use all of the data. Can this be set up as a 3-level longitudinal model in MLwiN? I realize that with only two time periods I can't do any growth curves, but I'm normally a SAS user, and it does not provide for repeated measures designs in the multinomial case. What do you think?
shakespeare
Posts: 70
Joined: Thu Feb 14, 2013 11:12 pm

Re: Multinomial repeated logistic regression

Post by shakespeare »

Thought about this some more. Seems like a repeated cross-sectional analysis would work. That makes this a three level model: individual measurements nested in time nested in physicians. If I include the grouping variable which is at level three, then I have an equation that looks like the following for a random intercept model that treats physicians as random:

yijk=B0+B1*Groupk+vok+uojk+eijk

where

vok is the random effect for physician
uojk is the random effect for time
eijk is the residual for each person

I'd like to free up the covariance structure for u so that it could be something like AR1 to account for the correlations across time. I guess I could also free up the covariance structure for e to account for the correlations within each time point. Does this make sense? I think I could figure out how to set this up in MLwiN.
Post Reply