Date: Mon, 26 Feb 2007 22:05:35 +0100
Reply-To: "Yesilkagit, A.K. (Kutsal)" <A.K.Yesilkagit@uu.nl>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: "Yesilkagit, A.K. (Kutsal)" <A.K.Yesilkagit@uu.nl>
Subject: Re: multilevel generalized linear model
Content-Type: text/plain; charset="iso-8859-1"
Thanks for all your advice, but I was a bit afraid that SPSS would not be able help me (yet). I'm working with SPSS v 15.0 and my first guess was that GEE would be the most proper procedure to try out.
Let me briefly describe the structure of my dataset. Maybe you'll come up with a workable suggestion (before I try out HLM).
I am comparing the establishment of independent public organizations across three countries between 1945 and 2000. In each country these bodies have been created in different years by successive cabinets across various policy areas.
So I have, at least I think I have, the following possible clusters: country, cabinet, policy areas and years. There is a hierarchy that runs from countries to cabinets/policy areas and years and my idea was to select each one of them successively as subject, nest the lower level group within the higher level one and run the analysis.
In other words, I was thinking of using subject definition with the possibility of nesting variables within each other.
Would this be a proper approximation of a multilevel analysis with GEE? More specifically, can I use the subject/nesting option in this way?
Van: SPSSX(r) Discussion namens Daniel Robertson
Verzonden: ma 2/26/2007 8:20
Onderwerp: Re: multilevel generalized linear model
Well, it depends on how the data are clustered. SPSS 15 includes
generalized estimating equations in the GENLIN procedure, which handles
observations clustered within subjects with a binary response variable
(e.g., repeated measures data). But there is currently no capacity in
SPSS for handling subjects clustered within groups w/ a binary RV. As
has been pointed out, HLM 6 does this quite well, and a 15-day tryout is
available for download from <estore.e-academy.com>.
> To fit an HLM model to dependent variables that are binary, multinomial, or
> ordinal, you will need a package like HLM 6. SPSS 15 does not handle these
> types of multilevel models (though one may hope that the next version will).
> Stephen Brand
> For personalized and professional consultation in statistics and research
> design, visit
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:SPSSX-L@LISTSERV.UGA.EDU]On Behalf Of
> Kutsal Yesilkagit
> Sent: Sunday, February 25, 2007 11:00 AM
> To: SPSSX-L@LISTSERV.UGA.EDU
> Subject: multilevel generalized linear model
> Is it possible to execute a multilevel generalized linear model for binary
> response variables with a dataset that contains clustered data (at the
> base level)?
> As far as I can see this is only possible in software that is specialized
> for mlm (HLM, MLWIN etc).
> kind regards
> Kutsal Yesilkagit
> assistant professor
> Utrecht School of Governance
> University of Utrecht
Sr. Research and Planning Associate
Institutional Research and Planning
440 Day Hall, Ithaca NY 14853-2801
ph:607.255.9642 / irp.cornell.edu