A Bayesian framework was developed to model the decision of a fully rational traveller under provision of travel information. According to this framework, utilities are measured in terms of expected utility theory while beliefs and perceptions about travel time are modelled as probability distributions. The traveller is assumed to be capable of making decisions under conditions of uncertainty about travel time and capable of updating his/her beliefs about the traffic environment based on experiencing the real travel times. The decision problem is represented by a decision tree. It is debatable however whether travel decisions are fully rational in the sense of expected utility theory and principles of Bayesian perception updating for different persons. To capture such possible limited rationality and underlying heterogeneity in risk attitude, decision protocols and preferences among travellers, in the present paper a discrete choice model is developed to allow for differences in decision style and learning in the choice between travel alternatives. In this paper, individual travellers are classified into several categories regarding their decision style and learning rules. In particular, a distinction is made between risk aversion, risk taking and fully rational decision styles and between recency, primacy and Bayesian learning rules. A risk aversive person will evaluate choice alternatives on the worst-case outcome, a risk taking person on the most likely outcome and a neutral person on the expected value across outcomes. At the same time, a recency learner will only make a decision based on his last experience, while a primacy learner will only make decisions based on the first experience. For a Bayesian learner, beliefs are updated with a Bayesian updating rule. The perceived utility of a choice alternative is conceptualized by the combination of these factors with their corresponding weights in utility functions of travel alternatives. Hence, our previous developed decision tree representation can be generalized to predict decisions also when the traveller is not risk neutral and not a Bayesian learner. We designed two experiments to estimate models; one concerning a repeated single route choice decision and one concerning an activity-scheduling task. In the first experiment, respondents repeatedly make choices within alternative routes of a trip for a given activity and after each choice the respondent receives feedback about the true travel time. In the second experiment, respondents make choices among combinations of two activities. The choice data allow us to estimate the weights of alternative perception-updating rules and decision styles of the respondent simultaneously. For the covering abstract see ITRD E137145.
Samenvatting