[R] Help needed on calculation of Moran's I

[Geertje Van der Heijden]

Hi,

Thanks for your help, Thibaut. It was helpful to look at the graphs.

I think that the problem is that when the lower boundary is 0 and the upper boundary is arbitrarily high it takes all connections into account (which shouldn't be a problem right?). If I make the lower boundary slightly higher than 0 and keep the maximum on some random high value or decrease the upper boundary - i.e. not all possible links are taken into account - the lm.morantest does work, but the Moran's I value is still really close to the expected value.

I have 58 sites from across the Neotropics. I am looking at the relationship of environmental and biogeographical factors on species richness and would like to see if there is still a spatial pattern in my data when these variables are taken into account. As far as I understand I should take all possible interactions between the sites into account if I want to test the residuals for spatial autocorrelation, right? I am a novice in this kind of spatial analyses, but is there a possibility to take all possible links between one point and all other points in the dataset into account when dealing with the Moran's I test for spatial autocorrelation in your residuals or have I misinterpreted the working of Morna's I? Or is there a better way to calculate it then I have done so far?

Any help on this will be greatly appreciated!

Many thanks,
Geertje


~~~~
Geertje van der Heijden
PhD student
Tropical Ecology
School of Geography
University of Leeds
Leeds LS2 9JT

Tel: (+44)(0)113 3433345 
Email: g.m.f.vanderheijden04 at leeds.ac.uk


-----Original Message-----
From: Thibaut Jombart [mailto:jombart at biomserv.univ-lyon1.fr] 
Sent: 26 October 2007 15:49
To: Geertje Van der Heijden
Cc: r-help at stat.math.ethz.ch
Subject: Re: [R] Help needed on calculation of Moran's I

Geertje Van der Heijden wrote:

>Hi,
>
>I am trying to calculate Moran's I test for the residuals for a 
>regression equation, but I have trouble converting my  coordinates into 
>nb format.
>
>I have used the dnearneigh() funtion now with an arbitrarily high upper 
>distance to make it include all plots. However, when I do the
>lm.morantest() I get a Moran's I value which is the same as the 
>expected value and a P-value of 1.
>
>I have used the following code:
>  
>
>>attach(lowland)
>>coords <- as.matrix(cbind(long, $lat)) coord.nb <- dnearneigh(coords, 
>>0, 10000 longlat=TRUE) coord.list <- nb2listw(coord.nb, style="W") 
>>lianasp.lm <- lm(lianasprich ~ log(averdist) + dsl + lianadens +
>>    
>>
>wooddens)
>  
>
>>lm.morantest(lianasp.lm, coord.list, alternative="two.sided")
>>    
>>
>
>And get the following result
>
>        Global Moran's I for regression residuals
>
>data:  
>model: lm(formula = lianasprich ~ log(averdist) + dsl + lianadens +
>wooddens)
>weights: coord.list
> 
>Moran I statistic standard deviate = 0, p-value = 1 alternative 
>hypothesis: two.sided sample estimates:
>Observed Moran's I        Expectation           Variance 
>     -1.754386e-02      -1.754386e-02       2.125706e-16 
>
>Can anyone tell me where I went wrong or if there is another way to 
>generate a nb list?
>
>Many thanks,
>Geertje
>  
>
Hi,

did you have a look at the graph you produced (using plot.nb)? Maybe the maximum distance between any two neighbours is too high. What does
max(dist(coords)) say ?
Anyway, there are many different connection networks available in several R packages, some being more appropriate than others to model paticular spatial distributions.

The package adegenet proposes a small tool to choose among several graphs interactively:

library(adegenet)
xy = matrix(runif(100),ncol=2)
yourcn = chooseCN(xy)

Regards,

Thibaut.

--
######################################
Thibaut JOMBART
CNRS UMR 5558 - Laboratoire de Biométrie et Biologie Evolutive Universite Lyon 1
43 bd du 11 novembre 1918
69622 Villeurbanne Cedex
Tél. : 04.72.43.29.35
Fax : 04.72.43.13.88
jombart at biomserv.univ-lyon1.fr
http://lbbe.univ-lyon1.fr/-Jombart-Thibaut-.html?lang=en
http://pbil.univ-lyon1.fr/software/adegenet/