Thesis

Mathematical aggregation of probabilistic expert judgements

Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2017
Thesis identifier
  • T14766
Person Identifier (Local)
  • 201064195
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • Mathematical aggregation of the probabilistic expert’s judgements in a structured expert judgement analysis is said to be relevant and critical. Due to absence of data, the judgements of the experts are used to perform forecasting and risk analysis. However, there is a gap in the literature where there may exist correlation in such judgements. This research is concerned with the situation where multiple experts are providing their numerical probability assessment for multiple quantities of interest. For each quantity of interest, there is a need of linear optimal weight on the basis of the experts’ judgement. This optimality is achieved by minimising the mean squared error (MSE) between the unbiased judgements provided by the experts for a quantity of interest, whose true value is unknown. Further, it has been assumed that the judgements of the experts is dependent on the sets of multiple quantities of interests, while, their errors presented in the judgements are correlated. This thesis presents two novel mathematical methods towards aggregating expert’s judgement through linear pooling. The first method is based on the empirical Bayes parametric formulation, and the second method is non-parametric. Both the chosen methods are compared using a stimulation study. This is to examine the performance of a given dependency structure which is further illustrated using a case study. In this context, a highly positive correlated expert gets the least weight when compared to an independent or negatively correlated experts. As stated in literature and reaffirmed through the simulation studies in this thesis, asymptotically the non-parametric approach has a slower error rate convergence, where the error is defined in terms of the MSE in comparison to the parametric empirical Bayes method. Based on the simulation study and the case study results, it is found that the empirical Bayes method outperforms the non-parametric method.
Advisor / supervisor
  • Quigley, John
  • Walls, Lesley
Resource Type
Note
  • Previously held under moratorium from 21st November 2017 until 21st November 2022.
DOI
Date Created
  • 2017
Former identifier
  • 9912573192802996
Funder

Beziehungen

Objekte