Get More From Your Models - Use Sensitivity Analysis

 

Full-Day Workshop

 

Sunday, December 6, 2009

Renaissance Inner Harbor

Baltimore, MD

 

Register via the Society for Risk Analysis.

 

Description

 

This workshop will answer key questions faced by those who conduct, manage, or review probabilistic and sensitivity analysis of risk models. When should you perform sensitivity analysis? What are the typical simulation techniques and software packages? What are the roles of uncertainty and sensitivity analyses as value added techniques in risk assessment? How do you prepare a model to facilitate sensitivity analysis? What are key considerations in the development of scenarios that are the basis for sensitivity analysis? What are some typical sensitivity analysis methods and how can you select among them? How should particular sensitivity analysis methods be applied? How should the results of sensitivity analysis be presented and interpreted? This workshop will answer these questions.

 

The methods and case studies presented are based upon several years of research at NC State University (Dr. Frey) and RTI International regarding developing quantitative risk assessment models for environmental and microbial systems and research regarding transferring, applying, and adapting sensitivity analysis methods developed in other disciplines (e.g. complex engineering systems) to quantitative exposure and risk assessment models. Results included a guidance document regarding selection, application, and interpretation of sensitivity analysis methods applied to quantitative risk assessment models. This workshop helps practitioners select specific sensitivity analysis methods relevant to the particular case study and model characteristics. The workshop will also aid in interpreting results from a sensitivity analysis in response to a particular modeling objective. Participants will be provided with course notes, a copy of the guidance document, and a tutorial with examples for how to perform sensitivity analysis using common methods. The basic concepts of probabilistic risk assessment will be illustrated using software packages such as @Risk and Crystal Ball. The uncertainty and sensitivity analysis methods will also be illustrated with practical case studies. This workshop is aimed at practitioners, managers, or reviewers who wish to refine their knowledge regarding approaches in risk assessment and sensitivity analysis methods.

 

Draft Agenda/Syllabus:

 

Introduction of Presenters                                                                                     (Morning)

            Amir Mokhtari

            Chris Frey

 

Introduction and background                                                                               (Morning)

            Common risk assessment paradigms

            Differences between chemical and microbial risk assessment frameworks

Definitions of variability and uncertainty

            Understanding deterministic and stochastic processes

            Two-dimensional simulation models

            Designing and implementing a two-dimensional Monte Carlo simulation

                       

Advanced techniques in quantitative risk assessment: the concept of sensitivity analysis

 (Morning)

 Sensitivity analysis vs. importance

             Motivation of the use of sensitivity analysis in exposure or risk assessment

 Situations in which it is necessary to perform sensitivity analysis

 Preparation of existing models and new models for sensitivity analysis                    

 

How are Existing and New Models Prepared to Facilitate Sensitivity Analysis?      (Morning)

            Preparation of Existing Models for Sensitivity Analysis

                        Identification of Model Structure

                        Identification of Inputs

                        Selection of Model Outputs for Sensitivity Analysis

                        Probabilistic Simulation and Implication for Sensitivity Analysis

                        Modifications of the Model

            Preparation of New Models for Sensitivity Analysis

                        Modeling Environment

                        Characterizing Variability and Uncertainty in the Probabilistic Simulation

                        Modeling Strategies

                        Model Documentation

    

Defining the case study scenarios for sensitivity analysis                                         (Afternoon)

            Conceptual framework for defining a case study scenario

            Identification of susceptible sub-population

            Identification of pathways of exposure

            Identification of temporal and spatial dimensions of the case study

            Probabilistic approaches

            Scenario uncertainty

 

Available methods for sensitivity analysis and general application procedures    (Afternoon)

            Mathematical methods

                        Nominal sensitivity analysis

                        Differential sensitivity analysis

            Statistical methods

                        Correlation coefficients

                        Regression analysis

                        Analysis of variance (ANOVA)

                        Classification and regression trees (CART)

                        Fourier amplitude sensitivity test (FAST)

                        Sobol’s method

            Graphical methods

                        Scatter plots

                        Conditional sensitivity analysis

General procedures for applying the methods to exposure or risk assessment models

            Deterministic versus probabilistic

            Identifying nominal values for inputs

            Need for normalization

            Add-hoc versus integrated approach

Method demonstration: Step-by-step tutorial with examples

 

Selection of sensitivity analysis methods                                                                (Afternoon)

            Key questions for selection of sensitivity analysis methods

                        What are the objectives of sensitivity analysis?

Based upon the objectives what information is needed from sensitivity analysis?

What are the characteristics of the model that constrain or indicate preference regarding method selection?

How detailed is the analysis?

What are the characteristics of the software that may constrain the selection of methods?

What are the specifications of computing resources?

Can “push-button” methods adequately address the characteristics of interest in the analysis?

Is the implementation of the selected sensitivity analysis method post hoc?

            Decision framework to assist in selecting sensitivity analysis methods

           

Practical case studies                                                                                            (Afternoon)

Discussion                                                                                                           (Afternoon)

Wrap-up

 

Presenters:

 

Amir Mokhtari

RTI International

Dr. Mokhtari is currently a Research Environmental Scientist at RTI International and has extensive experience in environmental and microbial food safety exposure and risk assessment, quantification of variability and uncertainty, and sensitivity analysis of probabilistic and stochastic models.

 

Chris Frey

North Carolina State University

Dr. Frey is a Professor of Civil Engineering at North Carolina State University.  His research is primarily in the areas of:  (1) quantification of variability and uncertainty in exposure and risk assessment; (2) quantification of variability and uncertainty in emissions factors and inventories; (3) measurement and modeling of real world in-use vehicle emissions; and (4) modeling and evaluation of energy systems.