An expected utility theory that matches human performance

Student thesis: Doctoral ThesisDoctor of Philosophy (PhD)


Maximising expected utility has long been accepted as a valid model of rational behaviour, however, it "has limited descriptive accuracy sim- ply because, in practice, people do not always behave in the prescribed way. This is considered evidence that either people are not rational, expected utility is not an appropriate characterisation of rationality, or combination of these. This thesis proposes that a modified form of expected utility hypothesis is normative, suggesting how people ought to behave and descriptive of how they actually do behave, provided that: a) most utility has no meaning unless it is in the presence of potential competitors; b) there is uncertainty in the nature of com- petitors; c) statements of probability are associated with uncertainty; d) utility is marginalised over uncertainty, with framing effects pro- viding constraints; and that e) utility is sensitive to risk, which, taken with reward and uncertainty suggests a three dimensional representa- tion. The first part of the thesis investigates the nature of reward in four experiments and proposes that a three dimensional reward struc- ture (reward, risk, and uncertainty) provides a better description of utility than reward alone. It also proposes that the semantic differ- ential, a well researched psychological instrument, is a representation or description of the reward structure. The second part of the thesis provides a mathematical model of a value function and a probabil- ity weighting function, testing them together against extant problem cases for decision making. It is concluded that utility, perhaps more accurately described as advantage in the present case, when construed as three dimensions and the result of a competition, provides a good explanation of many of the problem cases that are documented in the decision making literature.
Date of Award2012
Original languageEnglish
SupervisorRoland Baddeley (Supervisor)

Cite this