[R-sig-Geo] heteroskedasticity corrections for lagsarlm() in spdep

[Marco Helbich]

Dear Roger,

sorry for my last incomplete posting. Below is the output you need to 
know using the spdep example. Can you recommend other solutions to 
correct heteroskedasticity in a lagsarlm?

I appreciate your helping hand!
Best
Marco

 > data(oldcol)
 > example <- lagsarlm(CRIME ~ INC + HOVAL, data=COL.OLD, 
nb2listw(COL.nb, style="W"))
 > lm.target <- lm(example$tary ~ example$tarX - 1)
Fehler in model.frame.default(formula = example$tary ~ example$tarX - 1,  :
   ungültiger Typ (NULL) für die Variable 'example$tary'
 > traceback()
5: model.frame.default(formula = example$tary ~ example$tarX - 1,
        drop.unused.levels = TRUE)
4: model.frame(formula = example$tary ~ example$tarX - 1, 
drop.unused.levels = TRUE)
3: eval(expr, envir, enclos)
2: eval(mf, parent.frame())
1: lm(example$tary ~ example$tarX - 1)
 > sessionInfo()
R version 2.13.0 (2011-04-13)
Platform: x86_64-pc-mingw32/x64 (64-bit)

locale:
[1] LC_COLLATE=German_Austria.1252  LC_CTYPE=German_Austria.1252 
LC_MONETARY=German_Austria.1252
[4] LC_NUMERIC=C                    LC_TIME=German_Austria.1252

attached base packages:
[1] splines   stats     graphics  grDevices utils     datasets  methods 
   base

other attached packages:
  [1] robustbase_0.7-6   AER_1.1-8          strucchange_1.4-4 
sandwich_2.2-7     lmtest_0.9-28      zoo_1.7-4
  [7] Formula_1.0-0      car_2.0-10         survival_2.36-5 
nnet_7.3-1         spdep_0.5-35       coda_0.14-4
[13] deldir_0.0-13      nlme_3.1-101       MASS_7.3-12 
Matrix_0.9996875-3 boot_1.2-43        maptools_0.8-7
[19] lattice_0.19-23    sp_0.9-82          foreign_0.8-43

loaded via a namespace (and not attached):
[1] grid_2.13.0  tools_2.13.0
 >

Am 21.10.2011 17:31, schrieb Roger Bivand:
> On Fri, 21 Oct 2011, Marco Helbich wrote:
>
>> Dear list,
>>
>> I have estimate a spatial simultaneous autoregressive lag model by
>> means of the lagsarlm() function in the spdep package. The
>> Breusch-Pagan test for spatial models bptest.sarlm() shows that there
>> are some problems with heteroskedasticity.
>>
>> On p. 21 in the spdep manual states that "it is also technically
>> possible to make heteroskedasticity corrections to standard error
>> estimates by using the “lm.target” component of sarlm objects - using
>> functions in the lmtest and sandwich packages."
>>
>> First I thought the code below could be the solution, but it fails. I
>> have slightly adapted the code (lagsarlm instead of errorsarlm) on
>> page 21.
>>
>> data(oldcol)
>> example <- lagsarlm(CRIME ~ INC + HOVAL, data=COL.OLD,
>> nb2listw(COL.nb, style="W"))
>>
>> lm.target <- lm(example$tary ~ example$tarX - 1)
>> if (require(lmtest) && require(sandwich)) {
>> coeftest(lm.target, vcov=vcovHC(lm.target, type="HC0"), df=Inf)
>> }
>>
>> I am sure anyone has already done this. I would appreciate every hint
>> or piece of code (and sorry if this is nonsense).
>
> But your code works? Could you please be more specific about what
> doesn't work, the exact error message, the output of traceback() after
> the message, and of sessionInfo()? Of course, for the spatial lag model,
> this is really not OK, because of interaction between the covariance of
> rho and the betas. For the spatial error model, they do not interact,
> and the beta covariance matrix block is the same in OLS on the
> transformed covariates and ML.
>
> Roger
>
>>
>> Thank you in advance!
>> Best
>> Marco
>>
>>
>


-- 
Marco Helbich
Chair of GIScience
Department of Geography, University of Heidelberg

marco.helbich at geog.uni-heidelberg.de
http://giscience.uni-hd.de

Berliner Straße 48, D-69120 Heidelberg, Germany
fon: +49(0)6221 / 54 5572  //  fax: +49(0)6221 / 54 4529