A Bayesian Analysis of Data with Hierarchical Changepoint

• Activity Number: 403
• Type: Contributed
• Date/Time: Thursday, August 15, 2002 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Stat. Sciences*
• Author(s): Alix Gitelman*+
• Affiliation(s): Oregon State University
• Keywords: hierarchical model ; changepoint

Abstract

A Bayesian model with hierarchical changepoint is developed to allow for different changepoint values according to location and/or time differences in a collection of samples. This model extends work of Carlin, Gelfand, and Smith (1992), who present a hierarchical Bayesian analysis of changepoint problems wherein the models before and after the changepoint are hierarchical, but there is a single changepoint. The additional component addressed in this paper allows for a hierarchical changepoint process in addition to hierarchical components before and after the changepoint. The new model is fit, using Markov chain Monte Carlo methods, to data relating growing-degree days (a combined measure of time and temperature) to seed germination in Downy Brome, a common weed found across the North American West. The model estimates that different sites across the West have different growing-degree day values (i.e., changepoints) at which Downy Brome seeds begins to germinate, suggesting that time and temperature alone do not characterize the onset of germination.