Title: Local Likelihood Bayesian Cluster Modeling for Small Area Health Data
Speaker: Andrew Lawson, University of South Carolina
Date: April 24, 2006 12:00 - 1:30 pm
Location: DIMACS Center, CoRE Bldg, Room 433*, Rutgers University, Busch Campus, Piscataway, NJ
* note room change
The analysis of clustering can be approached from a variety of directions. Often it is assumed that relative risk model can be used coupled with exceedence probabilities to assess unusual risk. This does not allow the explicit modeling of clusters and is limited to hot spot detection. Previously, random object models have been forwarded where hidden processes are estimated. Here a Bayesian Hierarchical modeling approach to clustering is proposed where local likelihood is defined and clusters without reference to a hidden process. A lasso parameter is defined within which events are accumulated. This parameter has a spatially-structured prior distribution. Comparison of this approach for count and case event data via simulation (Hossain and Lawson(2006)) displays its good recovery behavior under different clustering scenarios.
Hossain, M. and Lawson, A. B. (2006) Cluster Detection Diagnostics for small area health data: with particular reference to evaluation of local likelihood Models Statistics in Medicine (to appear) Hossain, M. and Lawson, A.B. (2005) Local Likelihood Disease Clustering Development and Evaluation, Environmental and Ecological Statistics,12,3,259-273
see: DIMACS Computational and Mathematical Epidemiology Seminar Series 2005 - 2006