Abstract:
Weighted utility logitmodel assumes a preference structure that is compatible with weighted utility, a class of generalizations of expected utility. This weighted utility logit model distinguishes between two types of uncertainty: uncertainty due to modeler's imperfect information and measurement errors, which are accounted by the traditional error term of the logit model, and the inherent uncertainty in the choice situation, when the decision maker knows only the distribution of travel times.
The traditional mean value utility model, where travel time is presented as a sure attribute of an alternative, is a special case of both expected utility and weighted utility model, with a risk parameter alpha = 0. It is shown that one cannot derive the general form of weighted utility model from the general form of expected utility.
The weighted utility logit model is estimated for a continuous, normally distributed uncertain variable, and an uncertain variable with a discrete distribution. The parameter restrictions for risk aversion and first order dominance are specified for the uncertain variable with a discrete distribution. Both weighted utility logit models and the traditional mean value utility logit model are evaluated in a Monte Carlo simulation. The weighted utility models accurately identify correct parameter values in a wide range of plausible risk parameter (alpha) values, including the neighborhood of zero. If a true weighted utility model is misspecified and estimated as a traditional mean value model, the results are biased.
tr71.pdf