# [R-sig-Geo] how to generate a trend surface in spdep

chenliang wang hi181904665 at msn.com
Fri Oct 21 05:49:58 CEST 2011

```

>  On Thu, 20 Oct 2011, chenliang wang wrote:
>
>>  Dear list:
>>  I am just a newcomer to spatial autoregression.I have a question that
>>  how to generate a trend surface by means of function lagsarlm() of
>>  spdep:
>>  I have a sample points file, some vector points of independent
>>  variables,some surfaces of independent variables.I fit a spatial
>>  autoregression model using function lagsarlm(),and then I get some
>>  coefficients. How to generate a surface accordling to the simulated
>>  model? Thank you.
>>
>
>  The spatial lag model you have chosen is a spatial econometrics model,
>  for which prediction and interpolation are not well defined in the
>  literature. Interest is rather on the regression coefficients. If you
>  need to predict or interpolate, why do you not prefer a geostatistical
>  approach, where the regression coefficients are unimportant, but the
>  prediction, possibly of a surface, is the main goal?
>
>  Hope this clarifies,
>
>  Roger
>
>>
>>
>>
>>  Chen-Liang Wang
>>
>>  _______________________________________________
>>  R-sig-Geo mailing list
>>  R-sig-Geo at r-project.org
>>  https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>>
>
Thanks for your quickly reply! Maybe I 'm not familiar with the spatial
autoregression model .
We use a surface modeling process that surface is consist of two
components: trend and random error.
In the previous modeling process , our colleague use ordinary multiple
linear regression to get the trend of data, and then interpolate the
random error by means of several methods including geostatistical
approach.We fit the trend to reduce uncertainty and we interpolate the
residual or error to simulate local detail.
The entire modeling : predict = trend (regression of geospatial pattern)
+ error (interpolation of local detail)
I find the trend simulated by ordinary regression is not accurate
sometimes ,while the spatial autogression methods can predict it
perfectly . So I just think use the spatial regression method to
substitute for previous regression. But maybe some spatial regression is
not suitable for surface modeling according to what you have said.