Date:         Thu, 1 Dec 2011 12:13:59 -0500
Reply-To:     Gene Maguin <emaguin@buffalo.edu>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Gene Maguin <emaguin@buffalo.edu>
Subject:      Re: multilevel logistic regression over time
In-Reply-To:  <AFAB3BDD6FA022468ED594D954E35518194AC102EE@DCBL101VX.root.sutterhealth.org>
Content-Type: multipart/alternative;

Carol, look at Genlinmixed. I've never used it but



From: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] On Behalf Of
Parise, Carol A.
Sent: Thursday, December 01, 2011 12:10 PM
To: SPSSX-L@LISTSERV.UGA.EDU
Subject: Re: multilevel logistic regression over time



Is there a way to handle repeated measurements of a dichotomous DV in SPSS?
I looked briefly though the MIXED syntax and there doesn't appear to be a
way to choose the logit link. Is there a way to work around this?

  _____

From: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU] On Behalf Of R B
Sent: Wednesday, November 30, 2011 7:18 PM
To: SPSSX-L@LISTSERV.UGA.EDU
Subject: Re: multilevel logistic regression over time

Matt,

I'm not familiar with the model to which you are referring, but let's see if
I can help get you started. Suppose your study design consists of couples
who are assigned to one of two treatment conditions who are then repeatedly
measured over time (e.g., 4 time points) on a dichotomous response variable.
The data set is structured vertically as followed:

subj_ID couple_ID cond time y
1          1        1    1  0
2          1        1    1  1
1          1        1    2  1
2          1        1    2  1
1          1        1    3  1
2          1        1    3  0
1          1        1    4  0
2          1        1    4  0
1          2        2    1  0
2          2        2    1  1
1          2        2    2  0
2          2        2    2  1
1          2        2    3  1
2          2        2    3  1
1          2        2    4  1
2          2        2    4  0
1          3        1    1  1
2          3        1    1  1
1          3        1    2  1
2          3        1    2  0
1          3        1    3  1
2          3        1    3  0
1          3        1    4  1
2          3        1    4  1
.
.
.

Note that subject 1 from the first couple is NOT the same person as subject
1 from the second couple. That is, the dataset is structured such that
subject identification numbers are nested within couple identification
numbers (This point is important when it comes to construction of the second
RANDOM statement specified below).



The primary research question is whether or not there is a different
response between treatment conditions over time, taking into account nesting
of subjects in couples and repeated measures. With that stated, here's some
SAS code:

proc glimmix data=mydata method=laplace;
  class cond time subj_ID couple_ID;
  model y (event=last) = cond|time / s link=logit dist=binary;
  random = int / subject=couple_ID;
  random = int / subject=subj_ID(couple_ID);
run;

The GLIMMIX invocation statement specifies employment of a laplace
approximation method, although the adaptive quadrature method would be
preferred if computationally feasible. The CLASS statement specifies all
categorical predictor variables, including the subject variables. The MODEL
statements assumes a 2X4 fixed effects design matrix including the main
effects, condition and time, as well as the interaction effect,
condition-BY-time. The first RANDOM statement accounts for nesting of
subjects within couples to account for within-couple correlations. The
second RANDOM statement assumes a compound symmetric correlation among
within-subject measurements collected repeatedly over time.



Certainly, there are more sophisticated [and perhaps more appropriate random
effects design matrices to consider], but I think this is a reasonable
parameterization that has the chance of not being too computationally
intensive.



View this post as a starting point upon which you can modify to more
accurately fit your particular study.

HTH,

Ryan

On Wed, Nov 30, 2011 at 3:37 PM, Pirritano, Matthew <MPirritano@ochca.com>
wrote:
>
> Listers,
>
>
>
> Is it possible to run a longitudinal multilevel logistic regression? I'm
particularly interested in doing this with dyadic data a la Kenny's Actor
Partner Interdependence Model.
>
>
>
> And if so, are there any good guides for how to do this? And if not are
there any tutorials to run this type of analysis with other software,
specifically SAS? I've come across some for SAS and multilevel logistic with
dyadic data but I'm not sure how easy it will be to add the longitudinal
component to that.
>
>
>
> Thanks
>
> Matt
>
>
>
> Matthew Pirritano, Ph.D.
>
> Research Analyst IV
>
> Medical Services Initiative (MSI)
>
> Orange County Health Care Agency
>
> (714) 568-5648
>
>


[text/html]