|
Activity Number:
|
457
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #303685 |
|
Title:
|
Integrating Bayesian Time Series Methods Into Self-Modeling Regressions
|
|
Author(s):
|
Kert Viele*+ and Rhonda VanDyke
|
|
Companies:
|
University of Kentucky and Cincinnati Children's Hospital Medical Center
|
|
Address:
|
Department of Statistics, Lexington, KY, 40506-0027,
|
|
Keywords:
|
Self-Modeling Regressions ; Time Series ; BARS
|
|
Abstract:
|
Self-modeling regressions perform well at explaining large amounts of the variation present in synaptic transmission data. A self-modeling regression analysis of a set of functions involves postulating a common shape function g, with individual observed functions assumed to be affine shifts (in both the x and y axes) of the common shape g. Most of the prior work in this area involves assuming independent normal errors. In this talk we will discuss a Bayesian implementation which utilizes Bayesian Adaptive Regression Splines (BARS) to estimate the underlying shape function g and incorporates Bayesian time series methods to estimate both the time series order of the error and the time series coefficients. Fortunately, MCMC techniques allow us to be "modular" in our approach, which allows us to consider extensions of the self-modeling framework such as mixtures of self modeling regressions.
|
