[STATSGRASS] Re: [GRASS5] r.moran

[Roger Bivand]

On Mon, 27 Mar 2006, Wolf Bergenheim wrote:

> On Fri, 24 Mar 2006, Roger Bivand wrote:
> 
> >> In ArcMap I got Moran's I as 0.35 and with R I get 0.54. ArcMap gave a Z
> >> value of 217.3 and R gave 52.2.
> >
> > That is useful. I've got another conversation going off-list on the R side
> > about the Arc Moran's I, there are issues with how the spatial weights are
> > specified. We've managed to reproduce some values, but not all. It will be
> > documented on the R-sig-geo archive (which is searchable through
> > RSiteSearch() from the R prompt) before long, but if you could add your
> > case, that would help a lot.
> >
> > What we need is the R commands used to create the weights plus the results
> > of running moran.test() on the data, and the equivalent (copy&paste from
> > the ArcGIS Python results popup window (the one with the progress bar at
> > the top with a good deal of meaningless text and number that are not
> > properly rounded)).
> >
> 
> I'll do that when I'm at the GIS lab again. Tomorrow or next week.
> 
> > So far we know that for a points shapefile, the default in ArcGIS is to
> > join all points to all and use inverse distance, not standardising the
> > weights. Also the flat distance threshold can be reproduced. We haven't
> > tried a polygon layer yet.
> >
> 
> In ArcGIS I was able to choose the weight, and IIRC I used the inverse of 
> the distance squared, but it is possible that I selected only inverse of 
> the distance. I'll double check and get back to you on that.
> 
> >
> > If you could contribute matched argument sets with the same output, that
> > would be great, but any documentation of what is going on will be
> > valuable, either to show that spdep/GeoDa and ArcGIS give the same results
> > or otherwise.
> >
> 
> Sure, but I think that I'm not allowed to share the data. I'll have to 
> check that. Or better yet do you have some standard data (with areas) as a 
> shapefile? I could run it in Arc, and give you back the results.

Preliminary investigations of the Arc Moran's I indicates a question about 
the estimated variance when row standardisation is chosen. I've put some 
simulated point data on:

http://spatial.nhh.no/R/etc/simdata.zip

which would avoid opening closed data. Using inverse distance for polygon 
data also strikes me as being unusual - most of the literature uses 
contiguity. The zip archive includes results from GeoDa, R, Stata, and 
ArcGIS, and all agree apart from row standardized ArcGIS.

Roger

> 
> > Thanks for your patience!
> 
> No, thank *you* for your help!
> 
> --Wolf
> 
> 

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no