Skip to content

Duttileul's method

2 messages · W. Daniel Kissling, Roger Bivand

#
Hi all,

I am currently looking for a possibility to calculate the geographically effective 
degrees of freedom of a (non-spatial ordinary least squares) regression model in the 
presence of spatial autocorrelation within a gridded dataset. My regression model has 
spatial autocorrelation in the residuals (indicated by Moran's I values) but I do not 
want to use a spatial model (e.g. errorsarlm in package spdep). Instead, I would like 
to calculate the significance tests of my (non-spatial) regression model using the 
reduced degrees of freedom. I know that there is the Duttileul's method (Duttileul 
1993) in other software packages which does exactly this. Is there anything 
comparable in R? Is there a function in the spdep (or any other) package which 
allows this to be computed?

I am very happy to get feedback.

Best regards
Daniel

--------------------------------------------------------------------
W. Daniel Kissling
Community & Macroecology Group
Institut f?r Zoologie, Abteilung ?kologie
Johannes Gutenberg-Universit?t Mainz
D-55099 Mainz, Germany

Tel. +49 (0)6131 3925200  
Fax +49 (0)6131 3923731 
E-mail: kissling at uni-mainz.de

Personal: http://www.danielkissling.de/
Research Group: http://www.oekologie.biologie.uni-mainz.de/commecol/main.html
Virtual Institute Macroecology: http://www.ufz.de/index.php?en=6302
Centre of Macroecology: http://www.macroecology.ku.dk/
#
On Wed, 4 Apr 2007, W. Daniel Kissling wrote:

            
There is no implementation of degrees of freedom adjustment in contributed 
packages that I am aware of, and RSiteSearch() also gives no hits, so this 
seems to be correct. Essentially, to adjust the df you may need to fit the 
spatial model anyway, to get an estimate of the loss of df caused by 
autocorrelation. There is also a recent paper by Griffith in AAAG, 
December 2005, pp. 740-760.

Roger