|
Activity Number:
|
18
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Sunday, July 29, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
IMS
|
| Abstract - #308566 |
|
Title:
|
Space-Time Latent Structure Modeling for Small-Area Health Data
|
|
Author(s):
|
Andrew B. Lawson*+
|
|
Companies:
|
University of South Carolina
|
|
Address:
|
Epidemiology Biostatistics, Columbia, SC, 29209,
|
|
Keywords:
|
space-time ; health ; basis-functions ; mixtures ; MVCAR ; latent
|
|
Abstract:
|
In this talk I will compare two basic approaches to space-time latent structure modeling. The first approach will be focused on the use of mixtures where a weighted linear combination of m temporal profiles is assumed. The temporal components have temporal dependence, flexibly defined by autoregressive priors. The weight components are scaled to yield probabilities over the components but are allowed to have cross-correlation and spatial correlation between sites. Prior comparison of these approaches is made to some county-level lung cancer mortastributions are assumed for both the temporal and spatial components. A second approach to this form of modeling is the use of linear combinations of basis functions where the regression parameters have spatially structured prior distributions. These models allow the use of zero-inflated mixture priors to let components drop from the model.
|
