[R-sig-Geo] Why can you not calculate Moran's i for glm() residuals?

[Jenn.B]

Hi Tara,

I have not been able to find any additional information on an acceptable
method for examining spatial autocorrelation in the residuals of a glm model
(or any model besides an lm), nor have I been able to ascertain whether
visual inspection of Moran's i correlograms is the way to go in this case
(rather than testing for significance), or if a variogram is a better option
(if anyone else on this list can add to this, that would be much appreciated
- by both of us I'm sure)!

On a side note: In creating your correlograms, are you using the ncf or
pgrimess package (the latter of which is a wrapper for spdep's moran.test )?
I've found that even when run on the dependent variable (i.e., not the
residuals, but the Y variable, which in my case is a count or
presence/absence) the two functions produce very different correlograms,
particularly in the largest distance bins.

Cheers - and good luck,
Jenn
-- 
View this message in context: http://r-sig-geo.2731867.n2.nabble.com/Why-can-you-not-calculate-Moran-s-i-for-glm-residuals-tp5568942p5628276.html
Sent from the R-sig-geo mailing list archive at Nabble.com.