-------- Original Message --------
Subject: Regression for clustered data
Date: Mon, 3 Oct 2011 09:05:27 -0400
From: Paul Sanfilippo <[email protected]>
To: Morphmet <[email protected]>
Dear Morphometricians,
I'm wondering whether someone can offer some advice on how to tackle a
statistical problem. I'm wanting to evaluate whether shape is associated
with several covariates (i.e. centroid size, sex, age, + a couple of
others), each in turn. My dataset is composed of a couple of thousand
twins, and I'm looking at an ocular phenotype (optic nerve shape). The
difficulty I'm having is that I'm not sure how to account for the
non-independence in my data as I effectively have 2 levels of
clustering; i.e. pairs of twins are genetically related (either 50% on
average or 100%) -> right and left eyes obviously come from the same person.
If I understand this right, typically when you have groups to consider
one could perform a MANCOVA and test whether regression slopes between
groups are the same as a means of determining whether an overall value
can be used. But this assumes independent groups right? So if I was to
try this using 4 groups (e.g. Twin 1 Right, Twin 1 Left, Twin 2 Right,
Twin 2 Left) for my data, results would be biased and invalid.
Additionally, in my case the groups in themselves aren't of any
interest, rather it's a nuisance factor that I just want to account for.
In other (non-shape) regressions I've performed for my thesis (with only
one dependent variable), I've used a Multilevel modelling approach with
each level successively nested i.e. eye -> individual -> twin pair. But
I can't find info (have had a look at R packages) as to whether this
type of modeling can be applied in the multivariate case (i.e. multiple
dependent variables as in shape data), at least xtmixed in Stata only
accepts one dependent variable.
The other option I guess is to perform separate regressions for each of
my 4 groups, but I don't think this is optimum.
Any thoughts/advice/etc?
Thanks a lot in advance,
Paul
--
___________________________________________________
Paul Sanfilippo
PhD Candidate
Centre for Eye Research Australia, The University of Melbourne
Department of Ophthalmology, Locked Bag 8, East Melbourne 3002
Victoria Australia
p +61 3 9929 8175 f +61 3 99298164 www.cera.org.au
