[R-sig-Geo] Discrete choice spatial durbin or spatial general model

Roger Bivand Roger.Bivand at nhh.no
Sat Aug 1 09:56:58 CEST 2015


On Sat, 1 Aug 2015, Mueller,Drew wrote:

> I would like to know if it is possible to estimate either the spatial 
> durbin or general spatial model within spdep or sphet (or another 
> package) for discrete choice dependent variables.  I have attempted to 
> pass
>
> family = binomial(link = "probit")
>
> to spdep, hoping that spdep uses the glm()function and accepts "..." 
> arguments, but get an "unused argument" error.

Since there is nothing in the documentation or code to suggest anything of 
the sort, "hope" is wildly speculative and without foundation. The 
literature on spatial probit estimation has been reviewed in two recent 
articles - have you read them? One is:

http://journal.r-project.org/archive/2013-1/wilhelm-matos.pdf

another in Journal of Regional Science.

> If it is not possible in 
> spdep, is there another package that is able to estimate a model with 
> both autocorrelation and heteroskedasticity for binomial or multinomial 
> discrete responses?  I have successfully estimated the SAR and SEM using 
> packages McSpatial and spatialprobit, but wanted to know if it is 
> currently possible to estimate a model that includes both types of 
> spatial effects for discrete responses.  I suspect that both spatial 
> effects are present in my transportation mode-built environment model.

Multinomial response is not available anyhere AFAIK. McSpatial and 
spatialprobit are feasible for single parameter models, as is the use of 
the slm latent model in INLA. Durbin models are not a problem, as the WX 
can be created using create_WX() in spdep. Double parameter models are 
problematic everywhere, even though they are occasionally used, as the 
significance of a second spatial parameter (error and lag) usually 
indicates problems with entitation. This is when the spatial footprints of 
the processes for the variables differ from the units of observation.

Hope this helps,

Roger

>
> Thanks for your help,
> Drew
> _______________________________________________
> 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



More information about the R-sig-Geo mailing list