|
Activity Number:
|
476
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #309519 |
|
Title:
|
Loglinear Modeling for Point-Referenced Spatial Data
|
|
Author(s):
|
Eric Tassone*+ and Marie Lynn Miranda and Alan E. Gelfand
|
|
Companies:
|
Duke University and Duke University and Duke University
|
|
Address:
|
212 N Duke St, Durham, NC, 27701,
|
|
Keywords:
|
Bayesian disease mapping ; hierarchical/multilevel models ; spatial loglinear models
|
|
Abstract:
|
We develop spatial loglinear models for point-referenced data, extending nonspatial approaches by modeling a collection of spatially dependent trials of individual locations, rather than independent trials that ignore location or use only areal unit information. We use a spatial process of loglinear models, where the sampling model is a single multinomial trial at each of n (finite) locations. Despite only one observation at each location, we estimate parameters of the driving process model and predict the loglinear model at every location in the study region; the specification of the process model enables this via spatial smoothing of nearby loglinear model parameters. Further, the loglinear modeling allows point-referenced inference about all marginal and conditional probabilities associated with the model. We illustrate our approach with North Carolina Detailed Birth Record data.
|
