[Openspace] Geographic clustering of
koschins at uiuc.edu
Thu Jul 8 21:52:39 CDT 2004
You can run uni- and multivariate LISA analyses in GeoDa,
which contain a cluster analysis option. LISA stands for
Local Indicators of Spatial Association (for technical
details, see Luc Anselin's 1995 LISA article - let me know
if you need a copy).
LISA maps identify four types of spatial autocorrelation
based on a weights matrix that defines your contiguity
structure: Spatial clusters (negative near negative values
and positive near positive values) and spatial outliers
(negative near positive values and positive near negative
values). You can export the cluster cores, p-values, and
LISA indices and import them in a GIS or statistical package
for further analysis (note that the LISA map only displays
the core of the cluster, not the surrounding neighbors).
The steps to create LISA maps in GeoDa are:
-open map using a unique ID field
-create weights matrix (e.g., rook, queen, distance, k-
-select univariate (or multivariate) LISA
This process is quick to implement.
>I hope this topic is acceptable for this listserv.
>I am interested in using cluster analysis to develop
>typologies. Ideally, one of the criteria for clustering
>adjacency. Does anyone know how this can be done? A
colleague of mine
>has been experimenting with BoudarySeer, but the method
>software uses to cluster isn't entirely clear.
>Flora Matheson, Ph.D.
>Inner City Health Research Unit
>St. Michael's Hospital
>Phone: (416) 864-6060 x. 2580
>Fax: (416) 864-5485
>Email: mathesonf at smh.toronto.on.ca
More information about the Openspace