Date: Sat, 29 Jul 2006 03:56:26 +0200
Reply-To: Marc Halbrügge <email@example.com>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Marc Halbrügge <firstname.lastname@example.org>
Subject: Re: adjusting for non-independence?
Content-Type: text/plain; charset=ISO-8859-1
> I've got an analysis with a bivariate outcome (select or not selected)and a categorical IV (2-5 levels). Simple crosstabs and chisquare, right?? not really.
> Here's the deal...
> The unit of observation is the person. However, the people are in clusters from different locations (A,B,C). So, people from a location are more likely to respond similarly and this needs to be taken into account in the analysis.
> How do i do this? Would logistic regression controlling for location do it? Apparently in Stata, there is a 'cluster' option that can be used with any type of model so you'd specify 'location' as the cluster variable. Any analogous function in SPSS?
There are logistic models that would work. See Agresti, "Categorical
Data Analysis", chapter 12.1: "Random Effects Modeling of Clustered
Unfortunately, the current SPSS version doesn't do it yet (SPSS 15.0
will do, as stated some days ago on this list).
In a similar situation, I did the following: I used a permutation test,
but only permuted the sequences within the clusters. This way I kept the
effect of non-independence constantly (at least in my opinion).