In an increasingly data-driven world, data and its use aren't always all it's cracked up to be. This course aims to address the critical lack of any or appropriate data in many areas where complex decisions need to be made.
For instance, how can you predict volcano activity when no eruptions have been recorded over a long period of time? Or how can you predict how many people will be resistant to antibiotics in a country where there is no available data at national level? Or how about estimating the time needed to evacuate people in flood risk areas?
In situations like these, expert opinions are needed to address complex decision-making problems. This course, aimed at researchers and professionals from any academic background, will show you how expert opinion can be used for uncertainty quantification in a rigorous manner.
Various techniques are used in practice. They vary from the informal and undocumented opinion of one expert to a fully documented and formal elicitation of a panel of experts, whose uncertainty assessments can be aggregated to provide support for complex decision making.
In this course you will be introduced to state-of-the-art expert judgment methods, particularly the Classical Model (CM) or Cooke's method, which is arguably the most rigorous method for performing Structured Expert Judgment.
CM, developed at TU Delft by Roger Cooke, has been successfully applied for over 30 years in areas as diverse as climate change, disaster management, epidemiology, public and global health, ecology, aeronautics/aerospace, nuclear safety, environment and ecology, engineering and many others.
What you'll learn
By the end of the course all learners will be able to:
- Recognize and advise on when and in which settings to use the Classical Model (CM) for performing Structured Expert Judgment
- Account for uncertainty assessments in complex decision-making context when data pose issues
- Use the CM to analyze expert data and obtain answers to questions of interest
- Participate in an optional IDEA Protocol module, which uses a different method of performing Structured Expert Judgment.
Verified learners will have the added benefit of being able to:
- Get an in-depth perspective on the CM method
- Analyze expert data to apply Structured Expert Judgment methods to real world scenarios
- Participate in optional modules about dependence elicitation and eliciting probabilities.
WEEK 1: Why and when to use SEJ?
Consider why and when to use Structured Expert Judgment (SEJ) and the Classical Model (CM), and then apply the model to applicable scenarios.
WEEK 2: Statistical accuracy (calibration) and information score
Learn how to use two key performance measures within the CM effectively.
WEEK 3: Performance-based weights and the Decision Maker
Learn how to aggregate expert opinion based on performance-based weights. Review other weighting schemes and evaluate them with respect to in-sample and out-of-sample validation techniques.
WEEK 4: Data analysis using Excalibur
The learners receive expert data that they will use to:
i) Compute the statistical accuracy and information scores for each expert,
ii) Aggregate their assessments with various weights, and
iii) Comment on the performance of the resulting Decision Makers.
WEEK 5: Applications of CM
Learn about real CM studies using an available TU Delft SEJ dataset and discuss particulars of the different studies provided.
WEEK 6: Practical matters (biases, experts, elicitation)
Consider the practical matters that are necessary for running the elicitation. Special attention will be given to biases and how to train experts to assess uncertainties.
Optional modules about another SEJ approach (the IDEA protocol) will be provided for learners who are keen on learning about an alternative method. Modules on dependence elicitation and eliciting probabilities will be provided to verified learners who want to learn about other contexts for which SEJ methods are appropriate.
Additionally, a more advanced course will be available for learners who are keen to apply the model to a project in their own situation to a problem of interest.
Unless otherwise specified, the Course Materials of this course are Copyright Delft University of Technology and are licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
This is a Massive Open Online Course (MOOC) that runs on edX.
Basic concepts in Probability Theory and Statistics. Links to videos introducing the concepts will be provided.