[R-sig-Geo] impacts in spdep does not work for me when there is only one predictor

Roger Bivand Roger.Bivand at nhh.no
Wed Sep 29 10:22:30 CEST 2010


On Tue, 28 Sep 2010, Roger Bivand wrote:

> On Tue, 28 Sep 2010, Eelke Folmer wrote:
>
>> Dear Roger and others,
>> 
>> Using the impacts function in spdep to calculate distributions (on the 
>> regression parameters from the spatial lag model) I encountered the 
>> following problem: when there is only one exogenous predictor in the model 
>> the impacts function won't give the output summary of the mcmc simulations. 
>> Given the time the function takes to produce the error I think that the 
>> simulations do run.
>
> Thanks for a clear report and reproducible example. I'll confirm when a fix 
> has been committed to R-forge - it is caused ny matrix dimensions being 
> dropped when there is only one row or column in some settings.

Fixes committed to the spdep project on R-Forge, could those interested 
please check before I release to CRAN? The packages on R-Forge should be 
revision 339, release 0.5-23, so for example Windows binaries for R 
patched will be available after midnight CET.

Roger

>
> Best wishes,
>
> Roger
>
>> 
>> I tried the same on the columbus data where it also happens
>> 
>> example(columbus)
>> listw <- nb2listw(col.gal.nb)
>> lobj  <- lagsarlm(CRIME ~ INC, columbus, listw)
>> summary(lobj)
>> imp   <- impacts(lobj, listw=birds.wlist, R=100)
>> summary(imp,  zstats=TRUE)
>> 
>> Error in dimnames(x) <- dn :
>>  length of 'dimnames' [2] not equal to array extent
>> 
>> lobj  <- lagsarlm(CRIME ~ INC + HOVAL, columbus, listw)
>> summary(lobj)
>> imp   <- impacts(lobj, listw=birds.wlist, R=100)
>> summary(imp,  zstats=TRUE)
>> 
>> works fine.
>> 
>> I hope I didn't miss anything obvious.
>> Thanks in advance,
>> Eelke
>> 
>
>

-- 
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