[R-sig-Geo] Inverse distance weights and spatial error model-update

[Roger Bivand]

On Sun, 21 Feb 2010, Adam Boessen wrote:

> My apologies for the duplicate list, but I forgot to use plain text in 
> my last post, and thus there was a problem posting to archive.  Here is 
> the post now:

Thanks for cooperating about this, keeping to plain text really helps!

>
> Hi Everyone,
> I'm new to R, so thank you in advance for your patience.  I'm using US 
> census block groups in Buffalo New York to examine how neighborhood 
> characteristics affect crime.  I would like to use an inverse distance 
> weights (distance decay) of block group centroids that is banded at 1 
> kilometer.  In other words, I would like to create a one kilometer 
> buffer around each of the centroids, then use row standardized inverse 
> distance weights.  Finally, I would like to run a spatial error model 
> using these weights.
>
> Here is my code:
>> library(foreign)
>> library(spdep)
>>
>> buffalo <- read.dta("buffalo_blkgrps.dta")
>> attach(buffalo)
>> names(buffalo)
> [1] "bgidfp00"     "gage2900"     "gavghhinc00"  "gblack00"     "gcrwd00"
> [6] "gethhet00"    "ghhincsdln00" "glatino00"    "gocc00"       "gowner00"
> [11] "gpop00"       "gpov00"       "assaul"       "robber"       "burglr"
> [16] "motveh"       "murder"       "larcen"       "x"            "y"
>>
>> #distance based neighbors - making a neighbor list - bounded to 1 kilometer
>> coords <-(cbind(x,y))
>> neigh.nb <- dnearneigh(coords, 0, 1, longlat=TRUE)

I'm assuming that you know definitely that your coordinates are 
geographical not projected - I've seen users adding longlat=TRUE when 
not appropriate.

>> summary(neigh.nb)
> Neighbour list object:
> Number of regions: 409
> Number of nonzero links: 5298
> Percentage nonzero weights: 3.167126
> Average number of links: 12.95355
> 1 region with no links:
> 12
> Link number distribution:
>
> 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
> 1  1  3  3  4  5 14 25 28 23 19 26 32 44 29 24 29 25 24 13 15 12  3  4  1  1  1
> 1 least connected region:
> 3 with 1 link
> 1 most connected region:
> 382 with 26 links
>>
>> #making inverse distance weights
>> neigh.dist <- nbdists(neigh.nb, coords, longlat=TRUE)
>> inverse <- lapply(neigh.dist, function(x) (1/(x^2)))
>>
>> #creating row standardized spatial weights from list
>> neigh.listw <- nb2listw(inverse, style="W",zero.policy=TRUE)

No, use:

neigh.listw <- nb2listw(neigh.nb, glist=inverse, style="W",
   zero.policy=TRUE)

You pass the list of weights through the glist= argument, as in the 
example for col.w.d in ?nb2listw.

> Error in nb2listw(inverse, style = "W", zero.policy = TRUE) :
>  Not a neighbours list
>> traceback()
> 2: stop("Not a neighbours list")
> 1: nb2listw(inverse, style = "W", zero.policy = TRUE)
>>
>>
>> assault <- errorsarlm(assaul_r ~ gage2900 + gblack00 + gcrwd00 + gethhet00 + ghhincsdln00 + glatino00 + gocc00+ gowner00 + gpov00,
> + data=buffalo,listw=neigh.listw, zero.policy=TRUE)
> Error in eval(expr, envir, enclos) : object 'assaul_r' not found
>

Well, there is no assaul_r in buffalo, is there?

Hope this helps,

Roger

> The model will run when I do not include the code with the inverse weights and only interpoint distances, but it's unclear to me why I can't include the inverse distances when using nb2listw.  Any ideas on why this is occurring and/or help to alleviate this issue would be much appreciated.  The spatial error model from errorsarlm (package=spdep) will also run fine when I don't include the inverse weights.  Is there a better way to go about running this spatial error model using row standardized inverse distance weights?   Would spautolm from the spdep package be better suited for these data?  Any comments or suggestions are welcome.
> Thank you for your time!
>
> Adam
>
> Adam Boessen
> Doctoral Student
> Department of Criminology, Law and Society
> University of California, Irvine
> aboessen at uci.edu
>
> p.s. If it's helpful, here is my version of R:
>> version
>               _
> platform       i386-apple-darwin9.8.0
> arch           i386
> os             darwin9.8.0
> system         i386, darwin9.8.0
> status
> major          2
> minor          10.1
> year           2009
> month          12
> day            14
> svn rev        50720
> language       R
> version.string R version 2.10.1 (2009-12-14)
>
> _______________________________________________
> 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