Bivariate LISA interpretation

[tracy creighton]

 

Hi everyone,

I'm new to GeoDa and although the results I get for a bivariate LISA make sense to me, I just wanted confirmation that I am interpreting them correctly. I am having a lot of difficulty finding resources that assist in the interpretation of GeoDa results.

I have two variables: [1] socioeconomic status (SES), where higher values = higher levels of poverty, and [2] an access index value, where higher values indicate better access. We are interested in the areas where high poverty and low access are significantly correlated.

With SES on the x axis (FACTOR2) and ACCESS (SUM_R) on the y axis, I get a negative correlation (see attached as I could not embed it). So can someone please confirm that I am on the right track in concluding that the polygons represented by the HL area are indeed the ones that we are interested in (HL = high poverty, low access). The non-spatial relationship was confirmed with a regular scatterplot and indeed, increasing poverty means decreased access. My interpretation makes sense to me when I type it out but then when I try to confirm through various resources, I end up confusing myself. I just wanted to ensure that I am not oversimplifying things!!

I used the discussion here

http://geodacenter.asu.edu/openspace/2007-January/000953.html as an analogy.

Thanks in advance!

Cheers
Tracy

 

 

[Julia Koschinsky]

Hi Tracy,

A quick response:

The LISA statistic tests whether a value in a given location (X1i) is
more similar to that of the average of its neighbors (X1j) than would
be the case under spatial randomness. In the case of the bivariate
LISA, this is extended to two different variables, e.g. X1i (e.g.
poverty in one zip code) vs X2j (accessibility in neighboring zip
codes) or X2i vs X1j - I think you might be thinking of a bivariate
relationship for the same location, which is a different thing. Also
please note that your attachment is about the global Moran
scatterplot, not the LISA (local Moran).

Another thing to note is that the high and low categories are defined
relative to the mean of each variable. If the mean is based on a
skewed distribution (ie mean and median are eg far apart), then the
high and low categories will be based on this mean.

You might also be interested in this talk by Luc Anselin on LISAs on
our website:
http://geodacenter.org/eTalks/Day1Foundations_2_LSA/stream_Day1Foundations_2_LSA.html

Best,
Julia

> I used the discussion here *
> http://geodacenter.asu.edu/openspace/2007-January/000953.html*<http://geodacenter.asu.edu/openspace/2007-January/000953.html>as

> an analogy.
>
> Thanks in advance!
>
> Cheers
> Tracy
>

> --
> You received this message because you are subscribed to the Google Groups
> "Openspace List" group.
> To post to this group, send email to openspa...@googlegroups.com.
> To unsubscribe from this group, send email to
> openspace-lis...@googlegroups.com.
> For more options, visit this group at
> http://groups.google.com/group/openspace-list?hl=en.
>
>

--
************************
Julia Koschinsky, Ph.D.
Research Director
Arizona State University
GeoDa Center for Geospatial Analysis and Computation
julia.ko...@asu.edu

http://geodacenter.asu.edu
http://www.facebook.com/geodacenter
http://twitter.com/GeoDaCenter