A spatial scan statistic for case event data based on connected components

Full text
Author(s):	Cucala, Lionel ; Demattei, Christophe ; Lopes, Paulo ; Ribeiro, Andre Total Authors: 4
Document type:	Journal article
Source:	Computational Statistics; v. 28, n. 1, p. 13-pg., 2013-02-01.
Abstract
We propose a new spatial scan statistic based on graph theory as a method for detecting irregularly-shaped clusters of events over space. A graph-based method is proposed for identifying potential clusters in spatial point processes. It relies on linking the events closest than a given distance and thus defining a graph associated to the point process. The set of possible clusters is then restricted to windows including the connected components of the graph. The concentration in each of these possible clusters is measured through classical concentration indices based on likelihood ratio and also through a new concentration index which does not depend on any alternative hypothesis. These graph-based spatial scan tests seem to be very powerful against any arbitrarily-shaped cluster alternative, whatever the dimension of the data. These results have applications in various fields, such as the epidemiological study of rare diseases or the analysis of astrophysical data. (AU)

FAPESP's process:	06/04955-1 - Scaling relations and photometric properties of galaxy clusters
Grantee:	Paulo Afranio Augusto Lopes
Support Opportunities:	Research Grants - Young Investigators Grants