Lance Waller
Department of Biostatistics
Emory University
Flexible Neighborhood Definitions in Hierarchical Models
for Disease Mapping
Disease atlases display geographic variation of disease rates
(incidence or prevalence) often through choropleth maps of rate
estimates for a set of administrative districts. Such maps
typically involve a tradeoff between geographic resolution
(smaller areas), and stability of the rate estimates (too few
persons at risk in each area). Hierarchical models provide
a tool allowing the analyst to stablize rates for small areas
by taking advantage of spatial similarity between rates from
neighboring regions. Past applications use a fixed definition
of ``neighbor'' based on shared boundaries. We consider
distance-based neighborhood definitions and allow the data to
inform on which regions are neighbors, and the amount of
spatial similarity. We illustrate the approach on two data
sets, namely lung cancer mortality from counties in Ohio for
1988, and lip cancer in counties of Scotland for 1975-1980.
Joint work with Erin Conlon, University of Minnesota.
Back to Colloquium Series