[R-sig-Geo] about global getis g

[Roger Bivand]

On Sun, 21 Dec 2014, kun17 wrote:

> cold.zip <http://r-sig-geo.2731867.n2.nabble.com/file/n7587571/cold.zip>
>
> hello everyone,
>
> I have designed a random grid file and put a 'cold spots' in the center, I
> used globalG.test in spdep package to test the 'cold spots',but the results
> made me confused, below is my code:
>
> library(spdep)
> library(maptools)
> simu=readShapeSpatial("C:/GridCold.shp")
> dists=1:12
> n=length(dists)
> GG=EG=VG=ZG=pvalue=numeric(length=n)
>
> coord=coordinates(simu)
> for(i in 1:n){
> 	dnb=dnearneigh(coord,0,dists[i])
> 	simu.w=nb2listw(dnb,style="B",zero.policy=TRUE)
> 	G=globalG.test(simu$value,simu.w,zero.policy=TRUE,alternative="two.sided")
> 	GG[i]=G$estimate[1]
> 	EG[i]=G$estimate[2]
> 	VG[i]=G$estimate[3]
> 	ZG[i]=G$statistic
> 	pvalue[i]=G$p.value
> }
> cbind(dists,GG,EG,VG,ZG,pvalue)
>
> the results is:
>
>    dists         GG         EG           VG         ZG       pvalue
> [1,]     1 0.01326694 0.01307190 1.127119e-09  5.8097284 6.257426e-09
> [2,]     2 0.03791890 0.03762976 1.456950e-08  2.3954667 1.659922e-02
> [3,]     3 0.08282060 0.08294887 1.127367e-07 -0.3820144 7.024507e-01
> [4,]     4 0.13412895 0.13528451 3.728269e-07 -1.8925080 5.842333e-02
> [5,]     5 0.20749034 0.21073626 1.131177e-06 -3.0519145 2.273869e-03
> [6,]     6 0.27401114 0.27974817 2.170007e-06 -3.8945442 9.838356e-05
> [7,]     7 0.34154256 0.35049020 3.542196e-06 -4.7541397 1.992932e-06
> [8,]     8 0.42087567 0.43444829 5.557734e-06 -5.7572477 8.549642e-09
> [9,]     9 0.50377144 0.52100154 7.538721e-06 -6.2753643 3.488166e-10
> [10,]    10 0.58757397 0.60659362 8.890566e-06 -6.3787831 1.785007e-10
> [11,]    11 0.65976673 0.67685506 8.894732e-06 -5.7297151 1.005995e-08
> [12,]    12 0.72722625 0.74144560 8.057964e-06 -5.0091840 5.466130e-07
>
> the first two rows with distance 1 and 2 indicate significant 'hot spots'!
> when I use CrimeStat to calculate global G, when distance is 1 and 2,the
> results are both not significant.
>
> Could someone explain the difference?

The Crimestat IV values are:

> CS
     d        G       EG      SEG        ZG
V2  1 0.013267 0.013072 0.000262  0.745199
V3  2 0.037919 0.037630 0.000457  0.632980
V4  3 0.082821 0.082949 0.000736 -0.174370
V5  4 0.134129 0.135285 0.001036 -1.115844
V6  5 0.207490 0.210736 0.001491 -2.176953
V7  6 0.274011 0.279748 0.001903 -3.014358
V8  7 0.341543 0.350490 0.002316 -3.862926
V9  8 0.420876 0.434448 0.002797 -4.852804
V10 9 0.503771 0.521002 0.003203 -5.378934

which agree on G and E(G), but not beyond that. Ch 5 of the Crimestat 
manual has an error for B1, equation 5.21, which does not take into 
account the erratum published by the authors in 1993: "On page 195, the 
coefficient of W2 in B1, (just below center of the page) should be 6, not 
3.". However, this doesn't seem to be the only problem, as the spdep code 
when modified not to correct B1 doesn't reproduce the CrimeStat Standard 
error of "G" values.

Since CrimeStat is closed-source, and since we can show an error in their 
documentation, I suggest that you take this up with them, pointing out 
their not accommodating an erratum (there were more) published over twenty 
years ago, ask for a clarification, and report back here when they say 
what they are actually doing. The source to globalG.test() is readable, so 
they can readily check what it does.

Hope this helps,

Roger

>
> thank you!
>
> Kun
>
>
>
> --
> View this message in context: http://r-sig-geo.2731867.n2.nabble.com/about-global-getis-g-tp7587571.html
> Sent from the R-sig-geo mailing list archive at Nabble.com.
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>

-- 
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; fax +47 55 95 91 00
e-mail: Roger.Bivand at nhh.no