Multinomial logistic regression - SPSS - multilevel data!!

#1
Hello all

Thanks in advance for any advice. I have data with a categorical dependent variable with 8 levels, and multiple (6-15) IVs, that are both categorical and continuous, depending on what I choose to include in the model. The clear contender for building a model is multinomial logistic regression.

The problem I have is that the data is clustered, in that one person can occur multiple times, in which case their IVs (e.g. gender) will be overrepresented in the sample. In Stata the issue can be solved (I believe) by using the robust option. However, there is not such an option in SPSS.

Firstly, I should say that I know this might be hard. I have read in many forums that SPSS is not great with multilevel data. However...is there ANY way I can adjust my model to account for multiple cases of the same person. I'm really just looking for a short term fix. In the medium term, I will buy Stata (feel free to recommend another program!).

Any help would be greatly appreciated.
Very best
Paul
 
#2
Hi Paul,

Yes, SPSS has options.

One is to run a GEE model (generalized estimating equations). These are really set up for repeated measures data, and it adjusts for correlations among the residuals for the same subject. It's under the Generalized Linear Models tab in the menus. The command is Genlin, I believe, if you want to look it up in the Command Syntax Reference.

The other is to run a generalized linear mixed model (GLMM). This is only available in SPSS 19, so if you have an earlier version, you're out of luck.

Both can run multinomial logistic regressions, and both are available in Stata as well (with more options), but I don't remember the command names. I don't know Stata as well.

Karen
 
#3
Wow - thanks so much Karen! So far I've only received negative reactions when talking about SPSS in conjunction with multilevel modelling (at least when it comes to categorical variables). You're the first to come up with a solution! Thank you!
C86


Hi Paul,

Yes, SPSS has options.

One is to run a GEE model (generalized estimating equations). These are really set up for repeated measures data, and it adjusts for correlations among the residuals for the same subject. It's under the Generalized Linear Models tab in the menus. The command is Genlin, I believe, if you want to look it up in the Command Syntax Reference.

The other is to run a generalized linear mixed model (GLMM). This is only available in SPSS 19, so if you have an earlier version, you're out of luck.

Both can run multinomial logistic regressions, and both are available in Stata as well (with more options), but I don't remember the command names. I don't know Stata as well.

Karen