ENBIS-16: Sensitivity Analysis for Computer Models

15 September 2016; 09:00 – 12:30; University of Sheffield, School of Mathematics & Statistics, Hicks Building (Room TBA)

ENBIS-16 Half-Day Free Post-Conference Course on Thursday, September 15th, 2016

Course Description

Computer models are used in many disciplines for making predictions when conducting physical experiments would be too costly or impractical. Typically, there will be uncertainty as to what values should be used for some of the model inputs, which induces uncertainty in the model output predictions. This course is concerned with methods for investigating how individual uncertain inputs contribute to the output uncertainty, so that model users can determine where to concentrate their resources in reducing input uncertainty. The course will include both theory and practical computational tools, implemented in the software package R. The course will also include an introduction to methods for eliciting probability distributions from experts, for the purposes of quantifying uncertainty about model inputs.

The course is designed for users of computer models, and statisticians interested in computer model uncertainty. Participants are assumed to have knowledge of standard probability distributions and basic Monte Carlo methods.

The course should enable participants to:

  • understand how to quantify input importance using variance-based and decision-theoretic measures;
  • calculate these measures for computationally cheap computer models, using R
  • elicit a univariate probability distribution from an expert.

Please note there is a limited number of free participant slots for this course. Once the quota is full and registration no longer possible, please e-mail office@enbis.org to express your interest to be waitlisted.

 

Course Outline

The course will involve a mixture of lectures and computer practicals. Proposed timetable is as follows:

9am-10.30am. Sensitivity analysis theory

10.30am-11am. Eliciting prior distributions from experts

11am-12.30pm. Computational methods

 

Teaching Materials Provided to Course Participants

Presentation slides and R code.

 

Technical Prerequisites for Participants

Participants should bring their own laptops on which R has to be pre-installed, ideally with the following packages, including the R packages SHELF and mgcv.

 

Course Facilitators

Jeremy Oakley is a Professor of Statistics at the School of Mathematics and Statistics at the University of Sheffield. He is an Associate Editor for the SIAM/ASA Journal of Uncertainty Quantification. He has been researching statistical methods for computer model uncertainty for over 15 years, and has a number of both methodological and applied publications in the field.

Mark Strong is an academic public health doctor, and a statistician, at the School of Health and Related Research at the University of Sheffield. He is currently has a National Institute for Health Research postdoctoral fellowship, working on health economic decision model uncertainty.