[R-sig-Geo] spatial model with several observations per spatial unit

[Tom Petersen]

I am not so familiar with this kind of models, but I have browsed through a
little on Brad Carlin's web page now. How big problems are realistic to model
with this approach? I have 945 spatial units and about 20,000 firms observed up
to 10 years each.
/Tom
============================
Tom Petersen
Transport- och lokaliseringsanalys
Skolan för arkitektur och samhällsbyggnad
Teknikringen 78 B
KTH, 100 44 Stockholm
Tfn 08-790 68 33, 070-424 00 75



Citerar Paul Hewson <paul.hewson at plymouth.ac.uk>:

> Hello,
> Can you use a multivariate CAR for this?
> Best
> Paul
> -=-=-==-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
> Paul Hewson
> Lecturer in Statistics
> University of Plymouth
> Drake Circus
> Plymouth PL4 8AA
> 
> tel ++44(0)1752 232778
> email paul.hewson at plymouth.ac.uk
> web http://www.plymouth.ac.uk/staff/phewson
> 
> ________________________________________
> From: r-sig-geo-bounces at stat.math.ethz.ch
> [r-sig-geo-bounces at stat.math.ethz.ch] On Behalf Of Tom Petersen
> [tomp at infra.kth.se]
> Sent: 11 February 2008 14:08
> To: r-sig-geo at stat.math.ethz.ch
> Subject: [R-sig-Geo] spatial model with several observations per spatial
> unit
> 
> Dear list,
> I wonder if anyone has experience of implementing a spatial model with a
> large
> panel dataset?
> Besides the spatial dimension, the data is grouped by firm and industry. So
> I
> have several cross-sectional and time period observations in each spatial
> unit
> (the industry effect is so far modelled with separate regression equations).
> Does anybody have a piece of advise on how to progress with this
> kind of data?
> One possibility might be to implement a spatial SUR with one equation per
> time
> period(?). I forgot to mention that there is also considerable time series
> correlation in the data as well, as is common with economic data.
> So far, I have used a non-spatial "two-step" GMM (Arellano-Bond type) with
> some
> success, but of course without the cross-sectional correlation. Maybe it
> would
> be possible to proceed with this method, but weighting the observations
> differently. In this case, the cross-sectional (spatial) correlation must be
> estimated somehow in the "first step", in order to get the regressor
> parameters
> consistently estimated in the "second step".
> As is described above, the spatial correlation can be divided in "within
> area"
> and "across areas" components, apart from the (autocorrelated) time series
> component for each firm (and across firms, e.g. in the same spatial unit).
> 
> /Tom
> ============================
> Tom Petersen
> Transport and Location Analysis
> Dept. of Transport and Economics
> School of Architecture and the Built Environment
> Teknikringen 78 B
> KTH, SE-100 44 Stockholm
> e: tomp at kth.se
> t: +46 8 790 68 33
> m: +46 70 424 00 75
> 
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>