Simplified models for multi-criteria decision analysis under uncertainty

Includes abstract.Includes bibliographical references.When facilitating decisions in which some performance evaluations are uncertain, a decision must be taken about how this uncertainty is to be modelled. This involves, in part, choosing an uncertainty format {a way of representing the possible outcomes that may occur. It seems reasonable to suggest {and is an aim of the thesis to show {that the choice of how uncertain quantities are represented will exert some influence over the decision-making process and the final decision taken. Many models exist for multi-criteria decision analysis (MCDA) under conditions of uncertainty; perhaps the most well-known are those based on multi-attribute utility theory [MAUT, e.g. 147], which uses probability distributions to represent uncertainty. The great strength of MAUT is its axiomatic foundation, but even in its simplest form its practical implementation is formidable, and although there are several practical applications of MAUT reported in the literature [e.g. 39, 270] the number is small relative to its theoretical standing. Practical applications often use simpler decision models to aid decision making under uncertainty, based on uncertainty formats that `simplify' the full probability distributions (e.g. using expected values, variances, quantiles, etc). The aim of this thesis is to identify decision models associated with these `simplified' uncertainty formats and to evaluate the potential usefulness of these models as decision aids for problems involving uncertainty. It is hoped that doing so provides some guidance to practitioners about the types of models that may be used for uncertain decision making. The performance of simplified models is evaluated using three distinct methodological approaches {computer simulation, `laboratory' choice experiments, and real-world applications of decision analysis {in the hope of providing an integrated assessment. Chapter 3 generates a number of hypothetical decision problems by simulation, and within each problem simulates the hypothetical application of MAUT and various simplified decision models. The findings allow one to assess how the simplification of MAUT models might impact results, but do not provide any general conclusions because they are based on hypothetical decision problems and cannot evaluate practical issues like ease-of-use or the ability to generate insight that are critical to good decision aid. Chapter 4 addresses some of these limitations by reporting an experimental study consisting of choice tasks presented to numerate but unfacilitated participants. Tasks involved subjects selecting one from a set of five alternatives with uncertain attribute evaluations, with the format used to represent uncertainty and the number of objectives for the choice varied as part of the experimental design. The study is limited by the focus on descriptive rather than real prescriptive decision making, but has implications for prescriptive decision making practice in that natural tendencies are identified which may need to be overcome in the course of a prescriptive analysis