[Openspace] k-n-n spatial weights

[Luc Anselin]

k-nn weights do not work properly in the ML estimation
of the spatial lag and spatial error model. They work
fine in everything else. The reason is that they are
asymmetric and currently the ML routines in GeoDa
do not take that into account.
- if you have a small problem (N < 500) GeoDa uses
   eigenvalue calculation, but it does not take into account
   that the eigenvalues may be complex, which is the case
   for asymmetric weights --> GeoDa sometimes crashes
   and sometime seems to work, but in any case
   the result is incorrect. How incorrect depends.
- if you have a large problem (N > 500), GeoDa will
   crash
L.


On Sep 2, 2005, at 3:43 PM, Joshua Kincaid wrote:

> I've read that you can't use k-n-n spatial weighting in
> Geoda at this point; on the other hand, I have heard that
> you can, but the program may crash.  I have used this
> weighting in my spatial models and it seems to work fine.
> Therefore my question is: are the model estimates correct if
> Geoda is allowing me to use k-n-n weightings for spatial lag
> and error models or would they be erroneous?
>
> Josh Kincaid
> _______________________________________________
> Openspace mailing list
> Openspace at sal.uiuc.edu
> http://sal.uiuc.edu/mailman/listinfo/openspace
>