[R] spatial statistics vs. spatial econometrics

Prof Brian Ripley ripley at stats.ox.ac.uk
Mon Aug 4 10:30:13 CEST 2003

I completely fail to recognise your description of V&R ch 14. That is (in
part) about *continuous* spatial fields, and does include pictures of
correlograms and variograms.  That's not the same problem as looking for
spatial dependence in areally-sampled data.

My 1981 book (is that the one you haven't read?) treats these different
problems in different chapters.  So do the other widely-cited references
in spatial statistics: geographers tend to ignore all the rest of the
spatial sciences, however.

On Thu, 31 Jul 2003, Michael Roberts wrote:

> Dear R users,
> I am putting together reading and resources lists for spatial statistics
> and spatial econometrics and am looking for some pointers from more
> experienced practitioners.
> In particular, I find two "camps" in spatial modelling, and am wondering
> which approach is better suitied to which situation.
> The first camp is along the lines of Venables and Ripley's Chapter 14
> (and presumably Ripley's book, but I don't have that yet)--spatial
> trends and kriging (e.g., the geoR package);  the second along the lines
> of Anselin's book--spatial lag and spatial-autocorrelation models (e.g.,
> the spdep package).
> As far as I can tell, these amount to the same thing (in princple).

Wrong.  Kriging is primarily interested in prediction.

> The first camp likes to use row-standardized "weight matricies" in
> building covariance structures (to ensure there isn't too much
> dependence?).  I find this very unappealing to many models.  This camp
> doesn't seem to look at variograms or correlegrams as often--they just
> fit the model, which I also find unappealing.  The covariance structures
> also tend to be very simple.  It looks like there is more flexibility in
> the second camp.
> Mixed model procedures also seem to have spatial covariance structures.
> Is there a reason why there appears to be so few cross references
> between these camps?  What makes each approach best for different kinds
> of problems?

They are different problems with different fields of applications.
There are quite a few other different problems you would probably 
mis-identify as `camps', including several approaches to spatial point 

Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

More information about the R-help mailing list