[R-sig-Geo] row-standardized spatial weighting matrix

[Roger Bivand]

On Thu, 29 Nov 2007, Christina Li wrote:

> Dear list members,
>
> I have a couple of questions:
>
> 1. I kept getting this warning "spatial weights matrix not
>    row-standardized" when I ran LM tests, although I already row
>    standardized the weights matrix.
>
> Here are the specific steps: I row standardized a square contiguity 
> matrix in MatLab, exported it as an ASCII file. I then imported the 
> ASCII file into R as a matrix (as.matrix). I then convert the matrix to 
> a weights list object (mat2listw). Then I ran LM tests.
>
> Could it because that the row standardization was done in MatLab, not in 
> R? I couldn't find out how to do it in R, so had to do it in MatLab.

The code shows that what is happening is that an attribute of the weights 
component of the object set by nb2listw() is being checked. If the 
attribute is not found, the warning is issued. When weights are converted 
from a matrix, the attribute is not set, and the style component of the 
listw object as a whole is set to "M". The usual solution is to do 
something like:

x <- mat2listw(xmat)
xW <- nb2listw(x$neighbours, style="W")

to get to a listw object for which we know that row standardisation holds. 
If the original specification was for general weights, use the glist= 
argument to pass x$weights.

lm.LMtests() could check itself, and not issue the warning, but checking 
would be costly, and it is better to do this once by hand rather than many 
times automatically. I'm pretty sure in addition that the ASCII file route 
introduced rounding errors, so even if it did check, row sums might miss 
unity by machine fuzz (say 1e-12).

>
> Can anyone tell me how to do row standardization in R directly? Thank 
> you!!!
>
> 2. I got significantly different LM test results for LMlag, when i used
>    the original contiguity matrix and the row-standardized one. Is there
>    supposed to be a big difference?
>

If you check this out on the example on the help page for lm.LMtests(), 
you'll see the difference between nb2listw(COL.nb, style = "B") and 
nb2listw(COL.nb, style = "W"). If I remember correctly, the test is based 
on row-standardised weights. In general, row-standardisation increases the 
influence of observations with few neighbours, while other styles may 
increase the influence of those with many neighbours.

Roger

> Any help is much appreciated!
>
> Christina Li
>
>
> ---------------------------------
>
> 	[[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>

-- 
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