next up previous index
Next: asa.chapters.02 Up: ASA Chapters (2) Previous: ASA Chapters (2)


Sponsoring Section/Society: ASA Council of Chapters

Session Slot: 8:30-10:20 Thursday

Estimated Audience Size: 80

AudioVisual Request: xxx

Session Title: MCMC Methods in Practice

Theme Session: Yes/No

Applied Session: Yes

Session Organizer: Dey, Dipak K. University of Connecticut

Address: Department of Statistics, Box U-120, Storrs, Ct 06269-3120

Phone: (860)486-4196

Fax: (860)486-4113


Session Timing: 110 minutes total (Sorry about format):

110 minutes total...please allocate Opening Remarks by Chair - 5 First Speaker - 30 minutes Second Speaker - 30 minutes Third Speaker - 30 minutes

Floor Discussion - 15 minutes

Session Chair: Dey, Dipak K. University of Connecticut

Address: Department of Statistics, Box U-120, Storrs, Ct 06269-3120

Phone: (860)486-4196

Fax: (860)486-4113


1. Assessing Quality of Care Following a Heart Attack Using Monte Carlo Markov Chain Methods

Normand, Sharon-Lise,   Harvard School of Public Health, Boston

Address: Department of Health Care Policy

Phone: (617)432-3260

Fax: (617)432-0173


Landrum, Mary Beth, Harvard Medical School

Abstract: Medical practice guidelines are increasingly utilized to understand variations in the delivery of many medical and surgical procedures. The most common approach to developing guidelines is to elicit judgements from a multi-disciplinary panel of experts regarding treatment efficacy within distinct clinical strata. In this talk, we employ Monte Carlo Markov Chain methods to develop a standard of care for patients who have suffered a heart attack and subsequently, profile the quality of care delivered to a sample of Medicare patients who were treated in US hospitals during 1994 - 1995.

2. Analyzing Real Estate Data Problems Using the Gibbs Sampler

Ghosh, Sujit K.,   North Carolina State University, Raleigh

Address: Department of Statistics, Raleigh, NC 27695

Phone: (919)515-1950

Fax: (919)515-7591


Gelfand, Alan E., University of Connecticut

Knight, John R., Eberhardt School of Business, University of Pacific

Sirmans, C.F., University of Connecticut

Abstract: Real estate data is often characterized by data irregularities, e.g., missing data, censoring or truncation, measurement error, etc. Practitioners often discard missing or censored data cases and ignore measurement error concerns. We argue here that we can remedy these irregularity problems through simulation based model fitting using the Gibbs sampler. We describe the Gibbs sampler in the context of these issues focusing primarily on the missing data problem. We illustrate using a sample of residential property sales from Baton Rouge, Louisiana. We show dramatic improvement in inference by retaining the partially observed data cases rather than deleting them. We also detail how the other problems can be handled using the Gibbs sampler. We conclude that, while canned software to implement a Gibbs sampler does not exist, for the problems at hand, development is straightforward with substantial reward anticipated for the effort.

3. Bayesian Inference for Estimating Hunting Success Rates Based on Survey Data

He, Zhuoqiong,   Missouri Department of Conservation

Address: 1110 S College Avenue, Columbia, MO 65201

Phone: (314)882-9880

Fax: (314)874-8849


Sun, Dongchu, University of Missouri, Columbia

Abstract: Although most post-season harvest surveys are conducted at state level, the effective management of wildlife population often wants to estimate hunting success rates, hunting pressure and harvest at sub-area (such as management unit, region, or county) level. The sample sizes for some sub-areas are often very small or even zero. It is well known that direct survey estimates for small sub-areas are often yield unacceptably large standard errors due to small sample sizes. In this article, a Bayesian hierarchical linear mixed model is used to estimate hunting success rates at sub-area level for post season harvest surveys. The model includes fixed geographic variables such as forest coverage, random geographic variables such as regional differences, longitudinal variables such as time effects, and spatial variables such as the correlation between neighboring sub-areas. Markov chain Monte Carlo methods such as Gibbs sampling and adaptive rejection sampling are used to compute the Bayesian estimators of the parameters. The method is illustrated using data from Missouri Turkey Hunting Survey in the Spring of 1996. The Bayesian estimates are close to the frequency estimates for the counties with large sample sizes and quite stable for those counties with small sample sizes. Bayesian model selection is used to show that the spatial correlation between counties does exist.

List of speakers who are nonmembers: None

next up previous index
Next: asa.chapters.02 Up: ASA Chapters (2) Previous: ASA Chapters (2)
David Scott