[STATSGRASS] spatial autocorrelation

[Roger Bivand]

On Mon, 21 Mar 2005, Trevor Wiens wrote:

> I've reviewed some of the documentation about accounting for spatial
> autocorrelation in GRASS and came across an interesting statement in the
> McCauley Spatial Autocorrelation GRASS Tutorial.
> 
> "If data are measured on a nominal scale, then the join count statistic
> is used."
> 
> I'm using logistic regression to model bird distributions. We reduced
> our count data to use / availability because there were insufficient
> data to model abundance. The data was transect gathered. I've avoided
> the use of a spatial autocorrelate and a trend surface, because this
> necessitates knowing where things are in order to predict them. If I
> understand the above statement, the use of Geary's C or Moran's I are
> thus inappropriate because they are interval / ratio based statistics.
> 
> In the point counts along transects, how does one measure a joint count
> statistic?

A join count statistic treating left and right observations as neighbours 
will work, but it might be worth looking at runs tests in some form, 
because you are working on a line - in R, the rle() function gives the 
runs. You are right that Moran's I is not appropriate. With the join count 
test, the edge effects can be strong, but for you with a transect this 
isn't a big issue.

Roger 

> 
> T
> 

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Breiviksveien 40, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 93 93
e-mail: Roger.Bivand at nhh.no