Skip to content

Moran's I negative binomial generalized linear model

3 messages · Roberto Patuelli, Wild Life, Roger Bivand

#
Dear Antonio,
You may want to look at Jacqmin-Gadda et al. (1997), Tests of Geographical 
Correlation with Adjustment for Explanatory Variables: An Application to 
Dyspnoea in the Elderly, Statistics in Medicine 16, 1283-97.
Cheers
Roberto

****************************

Dear list,
I'm interested in testing the spatial autocorrelation of the residuals
of an negative binomial generalized linear model (glm.nb) using
Moran's I test.
My question is about the best method to do this. Can I use the
function lm.morantest? Or is there another more appropriate way?

Regards,

Antonio Silva

********************
Roberto Patuelli, Ph.D.
Istituto Ricerche Economiche (IRE) (Institute for Economic Research)
Universit? della Svizzera Italiana (University of Lugano)
via Maderno 24, CP 4361
CH-6904 Lugano
Switzerland
Phone: +41-(0)58-666-4166
Fax: +39-02-700419665
Email: roberto.patuelli at usi.ch
Homepage: http://www.people.lu.unisi.ch/patuellr
2 days later
#
Yes I know the work of Jacqmin-Gadda et al. (1997) which proposes the
use of a new statistic T.
However, more recent work of Lin and Zhang (2007) uses the test of
Moran's I to access the autocorrelation of residuals.
I'm more interested in using this approach, which has been used by
several authors.
The question would be if I should calculate the test using a sparse
matrix (lm.morantest) since it's on residuals or if I should calculate
it by the normal way (moran.test) as suggested by Lin and Zhang
(2007).

Many thanks for the reply.

Regards,

Ant?nio Silva


On Tue, Apr 20, 2010 at 11:15 AM, Roberto Patuelli
<roberto.patuelli at usi.ch> wrote:
#
On Fri, 23 Apr 2010, Wild Life wrote:

            
If you examine the class of a model object fitted using glm(), you'll see 
that it also inherits from "lm". This means that you may use 
lm.morantest() at your own risk. This is not the same as moran.test(), and 
I do not believe that Lin and Zhang suggested its use. I have no idea why 
you think that sparse matrices are involved. In testing model residuals, 
it is essential to handle the presence of the X variables, which is what 
lm.morantest() does. Since no studies have been done to find out how 
accurately this test detects spatial autocorrelation in glm errors, it is 
premature to simply extend findings for lm objects. It is also unclear 
whether this is what Lin and Zhang did, as I am unaware of them having 
published their code.

Hope this helps,

Roger