[Openspace] GEODA - Interpretation of LISA results

[Luc Anselin]

I am replying to the list since others may have wondered about this as 
well.

(1) The initial setup for the randomization is to use 99 permutations 
and p = 0.05 as
      the significance level.  The results for 99 permutations are 
somewhat sensitive
      to the actual random numbers used. With 999 this is much less the 
case.
      So, a scenario could be that the first time a location was 
marginally significant
      (at 0.05) but later permutations did not replicate this. With 
permutations, the
      results are never "exactly" replicable unless one uses the same 
random
      numbers, which GeoDa doesn't do.
(2) There was indeed a bug in an earlier version, but it's been fixed 
for some time.
       I can't replicate what you describe, however. The earlier version 
had
       high-low and low-high switched around, but not high-high and 
low-high.
       There is a problem with the LISA maps when multiple variables are 
used
       and new randomizations are calculated. All maps pertain to the 
latest
       variable selected and earlier maps are replaced. We are looking 
into this. If this
       is not your case, please send some more details to 
anselin at uiuc.edu

Also, we are posting minor fixes to
http://sal.agecon.uiuc.edu/stuff/fixes/
Right now, the version posted is 0.9.3d. Please make sure you have the 
most
recent one. You can download the executable and replace your current 
GeoDa
executable, no need to reinstall.

L.


On Tuesday, August 19, 2003, at 09:49 AM, Chris Green wrote:

> Two quick questions regarding the LISA results generated by GEODA
>
> First, when initially generating LISA results with GEODA, what are the 
> default settings that are used for  a. Randomization   b.  
> Significance Filter.   In some my analyses, I am noticing that when I  
> initially run a LISA (or a multivariate LISA) analysis, certain 
> geographic areas are consistently identified as being significantly 
> clustered ;  however on regenerating the results with different 
> randomization settings (after generating my initial results), these 
> cluster areas never become identified as significant again.   How do I 
> interpret what is happening here?
>
> Secondly, I am assuming that the classification schema used in GEODA 
> for identifying the four types of clustering (High, Low Low, Low High, 
> and High Low) is directly related to the four quadrants in the Morans 
> Scatterplot which is generated when undertaking a LISA (or 
> mulitvariate LISA) analysis.  In some of my analyses, I am finding 
> that when I highlight (brush) observations in the top right quadrant 
> of the Morans scatterplot (i.e. High-High), observations that are 
> classified as Low-High on the LISA map are highlighted.  When I save 
> these observations to Table, these highlighted areas are indeed 
> classifed as Low-High (category 3).    How do I interpret what is 
> happening here?
>
> Thanks
>
> Chris Green
> University of Manitoba
> Winnipeg, Manitoba
> Canada
>
>
> _______________________________________________
> Openspace mailing list
> Openspace at sal.agecon.uiuc.edu
> http://sal.agecon.uiuc.edu/mailman/listinfo/openspace
>