[R-sig-Geo] Inverse distance weights and spatial error model-update
Adam Boessen
aboessen at uci.edu
Mon Feb 22 09:19:20 CET 2010
Hi Roger,
Thank you very much for your quick reply and for creating the spdep package. I tried the tweaks to the code as you note below, but I still get a warning message. I have reposted my code below with a few more questions:
> #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(neigh.nb, glist=inverse, style="W", zero.policy=TRUE)
Warning message:
In nb2listw(neigh.nb, glist = inverse, style = "W", zero.policy = TRUE) :
zero sum general weights
> warning()
Warning message:
There was nothing listed here. Is this warning a problem? I was hesitant about using glist since I didn't think it could be incorporated in the errorsarlm models (hence the error in the code below). Would using spautolm() help to get around my error below in the model?
> #making crime counts into rates
> assaul_r <- (assaul/gpop00)*1000
>
Good catch on this one. I meant to make my crime counts into rates. I haven't found a package where I can model crime counts as Poisson or negative binomial regressions. Any suggestions?
> #SAR MODEL
> ass05 <- errorsarlm(assaul_r ~ gage2900 + gblack00 + gcrwd00 + gethhet00 + ghhincsdln00 + glatino00 + gocc00+ gowner00 + gpov00,data=buffalo,listw=neigh.listw, zero.policy=TRUE)
Error in subset.listw(listw, subset, zero.policy = zero.policy) :
Not yet able to subset general weights lists
> traceback()
4: stop("Not yet able to subset general weights lists")
3: subset.listw(listw, subset, zero.policy = zero.policy)
2: subset(listw, subset, zero.policy = zero.policy)
1: errorsarlm(assaul_r ~ gage2900 + gblack00 + gcrwd00 + gethhet00 +
ghhincsdln00 + glatino00 + gocc00 + gowner00 + gpov00, data = buffalo,
listw = neigh.listw, zero.policy = TRUE)
Thanks again for your help!
Adam
Adam Boessen
Doctoral Student
Department of Criminology, Law and Society
University of California, Irvine
aboessen at uci.edu
On Feb 21, 2010, at 11:38 PM, Roger Bivand wrote:
> 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
>
More information about the R-sig-Geo
mailing list