[Openspace] Spatial autocorrelation of errors

[Julia Koschinsky]

Gid,

You need to look at the LM tests for lag and error: they indicate a lag model is most appropriate for your data.

For details, please review the GeoDa workbook:
http://www.sal.uiuc.edu/stuff/stuff-sum/pdf/geodaworkbook.pdf
chs. 22-25, especially pp. 198-

Julia

---- Original message ----
>Date: Tue, 11 Sep 2007 17:24:09 -0700
>From: "Mazinga, Gideon (LLU)" <gmazinga01g at llu.edu>  
>Subject: RE: [Openspace] Spatial autocorrelation of errors  
>To: "Franklin, Frank" <frfrankl at jhsph.edu>, <openspace at sal.uiuc.edu>
>
>
>Dear Frank,
>
>Thanks for the response.I really appreciate the link you forwarded, it is very helpful.To clarify my question I have inserted a sample of the spatial dependence diagnotistics below for reference:
>
>DIAGNOSTICS FOR SPATIAL DEPENDENCE 
>FOR WEIGHT MATRIX : satisf.GAL  (row-standardized weights)
>TEST                          MI/DF      VALUE          PROB 
>Moran's I (error)           0.427718     3.8590954      0.0001138
>Lagrange Multiplier (lag)       1       12.1664919      0.0004866
>Robust LM (lag)                 1        5.3151522      0.0211407
>Lagrange Multiplier (error)     1        9.3345459      0.0022487
>Robust LM (error)               1        2.4832061      0.1150675
>Lagrange Multiplier (SARMA)     2       14.6496981      0.0006590
>
>
>Does the value of 0.427718 refer to Moran's I of the values or for errors. I think I am confused with the word error in bracket. In order for me not to proceed and test other models, what should the Moran's I (error) value be. Can I still proceed and run other models like spatial lag and spatial error model if the MI/DF value indicate a positive figure like the one above(0.427718)? My challenge is how do I determine that I need to run spatial lag or spatial error model? I hope this clarifies my question.
>
>Regards,
>
>Gid
>
>
>
>
>
>-----Original Message-----
>From: Franklin, Frank [mailto:frfrankl at jhsph.edu]
>Sent: Mon 9/10/2007 8:17 PM
>To: Mazinga, Gideon (LLU)
>Subject: RE: [Openspace] Spatial autocorrelation of errors
> 
>Gid,
>
>Gid...not exactly clear on what you mean in terms of autocorrelations in
>the "errors"...but if you mean if your model demonstrates spatial
>dependence based on the Moran's I  either due to spatial lag or spatial
>error, you should determine the source of this correlation, either by
>lag or error...or both, and then adjust for these sources of error by
>running separate models adjusting for each; and then examine the
>diagnostics to see which model has the best fit.
>
>Hope this helps.  
>
>Also see " Exploring Spatial Data with GeoDA: A Workbook at 
>http://www.csiss.org/
>
>Frank
>
>-----Original Message-----
>From: openspace-bounces at sal.uiuc.edu
>[mailto:openspace-bounces at sal.uiuc.edu] On Behalf Of Mazinga, Gideon
>(LLU)
>Sent: Monday, September 10, 2007 7:51 PM
>To: openspace at sal.uiuc.edu
>Subject: [Openspace] Spatial autocorrelation of errors
>
>When running multiple spatial regression analysis in Geoda and if it
>happens that there is spatial autocorrelation of  the errors, for
>instance, Moran's I = 0.27. What should one do? Can one still report the
>regression results? It is my understanding that it is better for the
>errors not be correlated, for the spatial model to be a better fit.
> 
>Can someone help on this?
> 
>Gid
>
>_______________________________________________
>Openspace mailing list
>Openspace at sal.uiuc.edu
>http://sal.uiuc.edu/mailman/listinfo/openspace