As Steve said in his email, the HLM text book will give clear detail on what you are doing and how to accomplish this (in a fairly software ambivalent way). Until recently, the only way to do this using HLM/MLM was either with the HLM software itself or with MLwin. These software do not work in the same way as SPSS or SAS, nor are they as commonly available on a University campus or research firm. However, they still are the most well developed programs for this purpose.
Having said that, I noted that when version 19 came out, they added the ability to do generalized linear mixed models, which would mean that you should be able to do a multinomial logistic multilevel regression. Having said that, I've not found much in the way of guides, reviews, or even clear information on how they developed this advancement. As a result, I don't yet feel comfortable using it instead of my other programs, and I'm currently doing some "testing" to compare it with the results obtained with HLM. The problem I currently have is the only data I have to make tests of things like log link MLM's is so large that it challenges the SPSS program due to the matrix size. Because of how SPSS and even SAS handle these "mixed" models, the matrix created in a nested model can get huge, and program like HLM, which handle it differently, will often successfully run reasonable programs that SPSS and even SAS will not.
An example for the above, which has created problems for me: I have 250K students who took a drug use, family values, and extra-curricular activities survey. They take the surveys every other grade, and these are all taken within schools. The data has over 5000 schools. The Schools are nested within networks which work collaboratively in interventions to reduce use, and the programs run by each vary (with some not running much at all). These networks can sometimes be nested within districts too, meaning that one district might have multiple networks. You typically would not run a model with this many levels (4), but even if run with 3 levels, you often end up with a matrix that SPSS and SAS won't accept. Further, I find that the models act as if they are over-estimated even if I only add one random variable per level. This program runs perfectly in HLM, and has been checked by numerous experts as a perfectly reasonable model to answer the research questions. It lead!
s me to believe that the SPSS and SAS approach is problematic.
Matthew J Poes
Research Data Specialist
Center for Prevention Research and Development
University of Illinois
-----Original Message-----
From: SPSSX(r) Discussion [mailto:
[hidden email]] On Behalf Of Abu
Sent: Wednesday, December 21, 2011 7:58 AM
To:
[hidden email]
Subject: A multilevel multinomial logistic regression analysis in SPSS
Does any of you know where I can find guidance/instruction for doing multilevel multinomial logistic regression in SPSS? I have a categorical dependent variable (it has five categories). It is measured three times. I want to conduct a multilevel analysis (two levels): within individual (time 1, time 2, and time 3), and between individuals. Specifically, i want to
(1) partition the within individual and between individual variation and (2) model the patter of changes in the dependent variable over time. Thanks, Abu
--
View this message in context:
http://spssx-discussion.1045642.n5.nabble.com/A-multilevel-multinomial-logistic-regression-analysis-in-SPSS-tp5091985p5091985.htmlSent from the SPSSX Discussion mailing list archive at Nabble.com.
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Locked
