Thesis
On taking a moment to learn from experts
- Creator
- Rights statement
- Awarding institution
- University of Strathclyde
- Date of award
- 2012
- Thesis identifier
- T13276
- Qualification Level
- Qualification Name
- Department, School or Faculty
- Abstract
- Subject matter experts have become increasingly important as sources of valuable information in the support of decision making for the Dutch Defence. Yet, the Defence methodology toolbox is lacking a methodology for dealing with quantitative subject matter expert judgements. In this thesis we evaluate a methodology that reflects the discrete character of quantitative expert judgements and is flexible in the amount of detail that can both be specified by the experts and is needed for the decision problem at hand: the Bates linear methodology. This entails that the methodology can be applied within a relatively short time frame, leading to a short response time. The methodology evaluated in this thesis also provides a vehicle to gradually switch from expert judgement to actually observed data when this becomes available. To date little guidance is available as to how to obtain the assessments from experts necessary to poulate a Bayes linear model. In this thesis we have evaluated (a bivariate extension of) the extended Pearson-Tukey method for the derivation of the second order moment assessment needed to quantify a Bayes linear model, by evaluating its performance for a wide variety of bivariate distributions. We found this method to perform very well when variables are not strongly skewed. By means of simulation studies we show that the Bayes linear adjustment of moments can be inaccurate for not joint Normally distributed variables. Yet, we find that the use of higher order moment information can greatly increase the accuracy. For the distributions considered in this thesis the increase is between five and eleven orders of magnitude when third and fourth ordermoment information is used as well in the adjustment. For distribution with a poor performance of the regular adjustment of moments this increase in accuracy is sustained when this higher order moment information is to be obtained from expert assessments, leading to increased accuracy between one and two orders of magnitude. Finally we develop a performance based method to combine sets of (product) moment assessments from different experts into one set of assessments that represents a rational consensus of the experts' assessments, so that multiple experts can be consulted for a Bayes linear study. Based on the results presented in this thesis we strongly advise to complement the Defence methodology toolbox with the Bayes linear methodology.
- Resource Type
- DOI
- Date Created
- 2012
- Former identifier
- 948369
关系
项目
缩略图 | 标题 | 上传日期 | 公开度 | 行动 |
---|---|---|---|---|
|
PDF of thesis T13276 | 2021-07-02 | 公开 | 下载 |