R help - Jun 2005 - user-defined spatial correlation structure in geeglm/geese

Dear all,

 

We have got data (response and predictor variables) for each country of the
world; I started by fitting standard GLM and tested for spatial correlation
using variogram models (geoR) fitted to the residuals of the GLM. Spatial
autocorrelation is significant. Therefore, I think about using general
estimation equations (geeglm or geese in geepack) allowing for residual
spatial correlation . Several questions:

 

1. Are GEE’s the right choice?

2. How can I specify a user-defined spatial (exponential) correlation
structure in geeglm or geese? I did not really get this from the help-sites.
I guess I need to set corstr = "userdefined" and zcor = Zcor, with
Zcor
being in some way the specified correlation matrix – but how? I earlier
specified spatial exponential correlation structures in glmmPQL, but it
seems that this does not work in the same way for geeglm or geese. Do I need
to use the function genZcor?

3. Makes it sense to use continents to specify clusters? Or is the number of
clusters (8) too low?

 

I am grate-full for any comments!

 

Thank you in advance, 
Swantje

 

 

Swantje Löbel

PhD student

 

Department of Plant Ecology

Evolutionary Biology Centre (EBC)

Uppsala University

Villavägen 14

SE-752 36 Uppsala

Sweden

 

Tel. +46 18 471 28 70

Fax +46 18 55 34 19

 

http://www.vaxtbio.uu.se/resfold/lobel.htm

 


	[[alternative HTML version deleted]]