Inverse distance weights and spatial error model-update

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:

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)

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)

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

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)

#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()

#making crime counts into rates
assaul_r <- (assaul/gpop00)*1000

#SAR MODEL
ass05 <- errorsarlm(assaul_r ~ gage2900 + gblack00 + gcrwd00 + gethhet00 + ghhincsdln00 + glatino00 + gocc00+ gowner00 + gpov00,data=buffalo,listw=neigh.listw, zero.policy=TRUE)

traceback()

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)

-- 
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

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:

From your listing of the neighbour object, one region has no neighbours at 

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)

--
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