## multilevel logistic

Moderators: statman, Analyst Techy, andris, Fierce, GerineL, Smash

abby

### multilevel logistic

Greetings, this may seem like a silly question... but I know that mixed models is equal to multilevel modelling in spss... but if I want to run a logistic multilevel model do I use the multinomial logistic feature?
statman
Posts: 2757
Joined: Tue Jun 12, 2007 12:08 pm
Location: Florida, USA
Mixed models has a repeated measure aspect, whereas multinomial logistic avoids the 2 category restriction (DV) that logistic regression imposes so, no, I don't see them as the same.

BTY, no question is silly but sometimes our answers might be
See the note below

Statman
Statistical Services
Labor
Posts: 35
Joined: Thu May 08, 2008 10:58 am
I have used HLM 6.0 (Raudenbush & Bryck) for this purpose. You can give it a try, it is not a one size fits all package, but you can do multiple levels and define binary dependent variables.
sara b
Posts: 7
Joined: Fri Jan 09, 2009 2:23 pm

### multi level logistic? do I understand?

My data is nested (units within hospitals) and I am trying to do a linear mixed model analysis. My outcome variable is dichotomous: Medication errors (yes, no). The below syntax is for nursing care hours total (NCH_T) to med errors (ME) – a simple piece of my overall model, but a place to start.

Is this syntax correct or am I not on the right track? or do I have to define my outcome variable differently? Should I be using HLM 6.0 as in the previous post? I would appreciate any guidance or references. I am using SPSS 17

MIXED ME BY NCH_T
/CRITERIA=CIN(95) MXITER(100) MXSTEP(5) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0,
ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE)
/FIXED=| SSTYPE(3)
/METHOD=ML
/PRINT=DESCRIPTIVES SOLUTION TESTCOV
/EMMEANS=TABLES(OVERALL).
Labor
Posts: 35
Joined: Thu May 08, 2008 10:58 am
Yes, HLM is your package. There used to be a trial version on their website. Perhaps MlWin can do the same, but as far as I know, MIXED only handles linear models. That said, I'm not a guru, so I might be wrong. For testing purposes, you can do a linear regression / mixed models analysis, just to get an idea whether there might be some effects. The silly thing is that predicted values may be over 1 or under 0. It will not shine.

With HLM, unlike with SPSS, you'll need a L1 and a L2 dataset, and load them both in the program. Theoretically, any type (SPSS, EXCEL, SAS,...) works. In practice, much patience is needed. However, once it's done, you connect them both and you click your way into a model. Be tolerant with iterations and take the time to understand the way the output is produced.

### Who is online

Users browsing this forum: No registered users and 2 guests