Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: RE: Testing for Spatial Autocorrelation in Residuals


From   Allan Joseph Medwick <medwick@dolphin.upenn.edu>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: RE: Testing for Spatial Autocorrelation in Residuals
Date   Mon, 10 Mar 2008 12:17:12 -0400

Hi, All,

Thank you, Nick, for your email.  I will try taking a closer look at
-spautoc- later.  I was planning on doing a Moran scatter plot and
mapping the residuals.

It seems like I might be able to use the "Lagrange Multiplier
diagnostics for spatial dependence in linear models" (lm.LMtests) in
-stdep- in R.  There is also a "Moran's I test for residual spatial
autocorrelation" (lm.morantest) in the same package.  I need to look
into these further.

How much sense does it make to look for spatial dependence in my
independent variables in addition to reporting the regular descriptive
statistics?  If it exists, does it mean anything?

Thanks,
Alllan

On Mon, Mar 10, 2008 at 8:50 AM, Nick Cox <n.j.cox@durham.ac.uk> wrote:
> Allan refers to -spautoc- (SSC) and -gwr- (STB), both user-written programs.
>
> In addition, a more substantial suite is Maurizio Pisati's stuff, also in
> the STB. -search pisati, author-.
>
> I was the author of -spautoc-. I never got round to implementing a regression
> residuals flavour of that, nor is it on my agenda. But you should find that starting from -spautoc-
> and adding code would be easier than starting from scratch.
>
> That said, my general impression is that you are on your own for the most of the things
> you ask for, i.e. you would need to write the code yourself. However, the import of
> your letter is that you want formal tests for autocorrelation. Often it's as or more useful
> to map residuals or to do a Moran scatter plot, i.e. the spatial equivalent of a scatter plot
> of residual vs L. residual. (-spautoc- calculates weighted means of neighbouring values given
> information on neighbours.)
>
> Nick
> n.j.cox@durham.ac.uk
>
> Allan Joseph Medwick
>
> I would like to examine the residuals from a Poisson and a Negative
> Binomial model for spatial autocorrelation.  I know that -spautoc-
> will calculate Moran and Geary measures of autocorrelation, but I have
> read that the "regular" Moran's I calculation is not appropriate for
> residuals.  Is there a way to check the deviance and Pearson residuals
> from a Poisson or Negative Binomial model in Stata for
> autocorrelation?  Can Anscombe residuals be used in the analysis?  Are
> there any additional tests besides the Moran and Geary measures that
> would be appropriate to check?  Any other suggestions?
>
> If there is spatial autocorrelation in the error term, is there a way
> to model this in Stata?  Would -gwr- be the best choice?
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>



-- 
Allan Joseph Medwick, MGA
Doctoral Student, Higher Education Management
University of Pennsylvania Graduate School of Education
Telephone: 267.872.0336

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index