[R-sig-Geo] local R2

[Roger Bivand]

On Mon, 7 Dec 2009, Eda Laze wrote:

> Hello,
>
> Thank you very much Roger for your helpful answer. It worked pretty well.
>
> When I run gwr with polygon data I get:
>
> "Warning message:
> In gwr(formula = AFCCH07HA ~ ALDRO12MEA + ALDRO1MEAN + ALHSETTMEA +  :
>  data is Spatial* object, ignoring coords argument"
>
> However, I may provide now two results in one of combinations of variables I
> use. I use the same number of neighbors and variables in both cases,
> the only change is as follows:
>
> In case one I use longlat =TRUE and in case two I do not write longlat=TRUE.
>
> case one (longlat=TRUE): local R2 values vary from 0.1651 to 0.5328
>     gwr.e                pred                    localR2
>  Min.   :-465.4243   Min.   :-142.267   Min.   :0.1651
>  1st Qu.: -17.4693   1st Qu.:   5.429   1st Qu.:0.2461
>  Median :  -3.2120   Median :  16.735   Median :0.2788
>  Mean   :   0.3215   Mean   :  17.798   Mean   :0.2835
>  3rd Qu.:  16.0616   3rd Qu.:  29.726   3rd Qu.:0.3181
>  Max.   : 321.4507   Max.   : 162.569   Max.   :0.5328
>  sum.w
>  Min.   :33.40
>  1st Qu.:56.47
>  Median :62.22
>  Mean   :62.32
>  Max.   :91.10
>
> case two (without longlat=TRUE): local R2 values vary from 0.06066 to 0.70522
>
>     gwr.e                pred             localR2
>  Min.   :-323.5487   Min.   :-244.457   Min.   :0.06066
>  1st Qu.: -13.4717   1st Qu.:  -1.223   1st Qu.:0.32168
>  Median :  -1.2330   Median :  14.144   Median :0.38511
>  Mean   :   0.9375   Mean   :  17.181   Mean   :0.40192
>  3rd Qu.:   9.5695   3rd Qu.:  38.060   3rd Qu.:0.47538
>  Max.   : 320.2477   Max.   : 176.522   Max.   :0.70522
>
>          sum.w
>  Min.   : 36.83
>  1st Qu.: 50.90
>  Median : 56.49
>  Mean   : 57.56
>  3rd Qu.: 62.76
>  Max.   :120.35
>
> I thought that adapt=0.01 (number of neighbors) could decide on
> weights. Visualisation of map differs, map two is rather a "surface".
> Does it mean that weights differ from case one to two due to
> the fact that "longlat = TRUE uses distances on the ellipse with WGS84
> parameters" as described in the spgwr manual updated in June 2009.
>
> The polygon data are projected in:
> Projected Coordinate System:    WGS_1984_UTM_Zone_34N.
> I would welcome very much an accurate response, despite my efforts to
> get the answer these days by myself.

Since the data are projected, they are in UTM zone 34, use longlat=FALSE. 
As a sanity check, assign the projection to the object:

proj4string(obj) <- CRS("+proj=utm +zone=34 +datum=WGS84")

then inverse project to longlat:

library(rgdal)
obj_ll <- spTransform(obj, CRS("+proj=longlat +datum=WGS84"))

and rerun the analysis with longlat=TRUE (the bandwidth will be in km not 
m), and even do:

writeOGR(obj_ll, "obj.kml", "obj", driver="KML")

for the input object, displaying in Google Earth or similar.

Roger

>
> Thank you in advance
>
> ++++++++++++
> On Fri, Nov 27, 2009 at 1:12 PM, Roger Bivand <Roger.Bivand at nhh.no> wrote:
>> On Fri, 27 Nov 2009, Eda Laze wrote:
>>
>>> Hello,
>>>
>>> I am currectly using gwr in R to get local R2. I have a shape file -
>>> polygon data.
>>> When I write codes as below: case 1 and 2, I get two different Local
>>> R2 values and maps:
>>>
>>> case1.adpt<-gwr(FCHA~PAC1+PAC2+PAC6, data=filename,
>>> coords=cbind(filename$X, filename$Y), adapt= 0.01, longlat=TRUE).
>>> case2.adpt<-gwr(FCHA~PAC1+PAC2+PAC6, data=filename,
>>> coords=cbind(filename$X, filename$Y), adapt= 0.01, hatmatrix = TRUE,
>>> se.fit=TRUE). Local R2 have different values from case 1 and map is
>>> different as well.
>>
>> In the second case, you are not using longlat=TRUE, which will give rather
>> different weights. Please only give examples with available data sets, and
>> include say the first 5 local R2 values, so that helpers know that they are
>> looking at the same thing as you are. Always state the output of
>> sessionInfo().
>>
>> Hope this helps,
>>
>> Roger
>>
>>>
>>> Furthermore, when I visualize case 1: i.e., local R2 on map, each
>>> polygon has a certain value and responding color while in case 2: map
>>> shows clustering of Local R2 (values) similar to Georgia case study.
>>> However, I wonder why this happen and which is the right way to get
>>> local R2.
>>>
>>> I read spgwr manual and update. Though an accurate answer is very welcome.
>>>
>>> Thank you.
>>>
>>> _______________________________________________
>>> 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
>>
>>
>
> _______________________________________________
> 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