Bayesian Analysis and Applications in Risk Analysis

A workshop
to be held in conjunction with the

Society for Risk Analysis Annual Meeting

8am - 5pm
Sunday, 7 December 2003

Renaissance Harborplace Hotel
Baltimore, Maryland

This full-day workshop will introduce the methods of Bayesian analysis. It will review the use of WinBUGS, a software tool for Bayesian calculation.

Synopsis
Agenda
Presenters
Registration
Venue
More information
Related links

Synopsis

Bayesian statistical methods are enjoying a renaissance and are increasingly being applied in many fields, including risk assessment. The workshop will cover four main areas:

The main features and advantages/drawbacks of Bayesian statistics;
Specific applications in several fields of risk analysis; and
Solutions to implementing Bayesian calculations in different software environments; and
An introduction to Markov chain Monte Carlo methods and WINBugs software.

The derivation of Bayes theorem and its extensions to discrete and continuous distributions will be covered, as well as the relationship of Bayes theorem to other forms of logical inference. The fundamental concepts of prior probability, likelihood, and posterior probability will be illustrated using a series of examples of increasing complexity. These include (1) problems featuring discrete and continuous prior distributions, and (2) likelihood values obtained by inspection, mathematical functions, count data, and logical inference. Topics will include Bayesian methods for parameter estimation and categorical analysis, and are demonstrated and developed using examples from engineering, medicine, public health, and toxicology. Discussions will include similarities and differences between frequentist and Bayesian methods. The second half of the workshop focuses on the problems of implementing Bayesian computations. The computational aspects of Bayesian analysis are discussed further with special emphasis on implementing Bayesian calculations in probabilistic simulation models. In addition to a thorough discussion of conjugate distributions useful in risk analysis, a method of utilizing output from @Risk to do non-parametric Bayesian calculations will be demonstrated. By providing a practical computational solution to formerly intractable problems, Markov chain Monte Carlo methods have played a significant role in expanding the application of Bayesian methods. This important technique is introduced using WinBugs software to illustrate the simultaneous estimation of parameters to develop posterior distributions. In addition to the demonstrations and hands-on activities, we will discuss the advantages and drawbacks to this approach. Each attendee is asked to bring a laptop with Excel, @Risk, and WINBugs 1.4 installed (free downloads of @Risk and WINBugs 1.4 are available via the web). USDA will provide some laptops (very few wll be available). Handouts will include lecture notes, specific examples, and CD containing the applications and solutions developed and demonstrated in the workshop.

Agenda

8:00 Introduction

Brief history of Bayesian statistics
Derivation of Bayes Theorem
Discrete distributions
Continuous distributions
Prior probability distributions
Likelihood functions and values
Comparison of frequentist and Bayesian statistics

9:00 Applying Bayes Theorem (hands-on)

Discrete distribution (proportion)
Continuous distribution (proportion)
Categorical analysis
Monty Hall Problem

10:00 -10:15 break 10:15 Bayesian Applications in Specific Fields (lecture/demonstration/discussion)

Medical Risk Assessment and Decision-Making
Toxicology: Evaluating Chemical Behavior
Public Health
Reliability Engineering
Ecological Risk Assessment

12:00 Break for Lunch 1:00 Review of Bayes Theorem

Prior probability
Likelihood
Posterior probability

1:15 Computational Aspects of Bayesian Analysis (lecture, demonstration, hands-on)

Software Environment for Bayesian Analysis
Incorporating Bayesian analysis in Monte Carlo simulation models
Brief theory conjugate distributions
Using conjugate distributions
A catalog of conjugate distributions
Non-parametric Bayesian computation using @Risk output

3:00-3:15 break 3:15 Introduction to Markov chain Monte Carlo methods for multivariate Bayesian Analysis

Markov chain Monte Carlo Lite
[MC]2 The heart of modern Bayesian computation: what it does, advantages over single parameter Bayesian analysis
WINBugs software: explanation, demonstration, class exercises.
Application in public health: estimating prevalence of rare diseases.
Applications in reliability modeling

4:45 Resources for further study

Web resources for Bayesian theory, computation
Paper resources: books, journals, etc.

5:00 close

Presenters

Robert McDowell

Robert McDowell is Senior Staff Economist of the Risk Analysis Systems program at the Animal and Plant Health Inspection Service (APHIS), a part of the U.S. Department of Agriculture. Robert McDowell received an award for innovation and technical excellence in building the first computer model to quantitatively predict the risk of intestinal illness in humans from eating internally contaminated eggs.

Stan Kaplan

Stan Kaplan is a partner at Bayesian Systems, Inc. Kaplan is one of the early practitioners of the discipline now known as Quantitative Risk Assessment (QRA), and a major contributor to its theory, language, philosophy and methodology. His ideas and methods are widely used in the international risk and reliability community, and in regulatory agencies worldwide. His first rigorous training in risk thinking took place during his first job in industry, which was assembling pieces of uranium to determine the critical mass. Later he began formalizing, mathematizing, clarifying, simplifying, and quantifying the risk assessment process, beginning with risk analyses of nuclear plants and major construction projects. Others of his applications include risk assessments of portions of the space shuttle system, aircraft impact on nuclear plants, hazardous materials storage and transportation, a spent fuel pool, re-search reactors, offshore oil drilling (environmental risk), uranium mill tailings, underground oil storage, pipeline construction projects (surface and undersea), a coal mine, chemical plants, a dam, tarsands projects, a refinery demolition project, manufacturing processes, licensing of a new agricultural chemical, and regulatory rule changes relating to importation of animals and fruits. He has been instrumental in motivating and training the U.S. Dept. of Agriculture (along with other Government agencies) in the use of QRA and Bayesian methods. Kaplan is a member of the National Academy of Engineering, a Fellow of the Society for Risk Analysis and the author of a number of the seminal papers in this field.

Igor Linkov

Igor Linkov, a Senior Risk Assessor with ICF, has more than 12 years experience managing human health and environmental risk assessment, as well as expertise in communicating risk to public and public education. He received his PhD in environmental and occupational health from University of Pittsburgh, MS equivalent in Engineering and Public Policy from Carnegie Mellon University. His educational experience also includes Post-Doctoral Fellowship at Harvard University. Dr. Linkov's skills include project management, assessment of human health and environmental risks from chemical and biological agents, basic research on chemical carcinogenicity, toxicology and advanced modeling. He is also developing software for environmental modeling, risk assessment and policy support. He is currently supporting Army Natick Laboratory in a large-scale risk assessment of site-related contamination and communication site risks to local communities and stakeholders. His current interests include developing of risk-based protocols for developing reuse options for military sites. Dr Linkov has organized four international conferences on risk-related issues for the North Atlantic Treaty Organization and managed two research projects. Most recently, he was invited speaker at the NATO conference on social and psychological consequences of terrorism where he proposed the use of risk-based approaches as an efficient communication and evaluation tools.

Registration

The registration fee is $250, before 10 November, or $310 on site. You do not need to register for the Annual Meeting to attend the workshop. Registration will be handled by

Secretariat sra@burkinc.com
Society for Risk Analysis www.sra.org
1313 Dolley Madison Boulevard, Suite 402
McLean, Virginia 22101 USA
703-790-1745, fax 703-790-2672

Venue

The event will be held 8am - 5pm on Sunday, 7 December 2003, at

Renaissance Harborplace Hotel
202 East Pratt Street
Baltimore, Maryland 21202 USA
800-468-3571 (toll-free reservations)
410-547-1200 (direct to the hotel)

The room for the event has not yet been determined; check with the hotel concierge. To reserve a room at the hotel, call 800-468-3571 before 10 November 2003. Be sure to identify yourself as a SRA Annual Meeting attendee to receive the SRA group rate of $135 per night (single or double occupancy) plus 12.5% tax. Cancellations must be made at least 48 hours in advance. See a description of the hotel at http://www.marriott.com/dpp/PropertyPage.asp?MarshaCode=BWISH.

More information

More information can be obtained from Robert McDowell robert.m.mcdowell@usda.gov, telephone 301.734.5951, fax 301.734.5899.