Shortcomings and Limitations of Prospect Theory in Modelling Travel Behaviour

As evidence from empirical studies on PT are emerging, in both general and travel behaviour contexts, its shortcoming and limitations are also increasingly being identified.

While a robust finding in many SP studies is the gain/loss asymmetry, other features of PT are less explored in a travel behaviour context. Interestingly, findings reported by Hess et al. (2008) do not support the exact supposition put forward by PT that the value function for losses should be convex and relatively steep whilst the value function for gains losses would be expected to be concave and not quite so steep. They found different relationships between gains and losses for different attributes.

Avineri and Bovy (2008) described some of the major challenges modellers are faced with when applying PT to travel behaviour context. PT is not a full-fledged theory of decision-making, and there is an on-going debate about its applicability in some economics contexts. In addition to the general complexities in estimating CPT parameter values, transport researchers are faced with additional difficulties and challenges. Due to the different contexts travel journeys are made in, different modes and trip purposes, it is difficult to estimate a set of parameter values that represent a typical decision-maker. Therefore the incorporation of heterogeneity terms associated with perceived values of PT parameters might become a rather complex task. Moreover, due to the lack of definite and in-consensus reference points in travel contexts, modellers may have difficulties in understanding and modelling travel choice outcomes as 'gains' and 'losses', as observed and framed by the travellers.

In addition, some open questions remain about the suitability of PT to describe and model a dynamic environment of travel choice, and the effect of feedback and learning on repeated choice. PT was originally proposed in order to capture description-based decisions in one-shot tasks. Most features of PT, such as reference dependency and gain/loss asymmetry, have been exhibited in a range of behaviours tested mainly in static ('one-shot') settings, with no (or limited) feedback or incentives. However, some recent studies have questioned this paradigm, mainly in dynamic settings (repeated choice) that provides feedback (and sometimes reward) related to the choice, and features a learning process (see Barron & Erev, 2003; Ert & Erev, 2008; and in a travel behaviour context Avineri & Prashker, 2003; Ben-Elia, Erev, & Shiftan, 2008).

Many travel choices may be considered to be a routine (e.g. commuting) and can be therefore described as a repeated choice situation, where feedback (resulted travel time) is involved. There is a relation between the travel utilities experienced by the travellers in past time periods and the current attractiveness of travel choices. The travel time distribution is not likely to be completely known to travellers at the time of choice, and their perception of travel time may be dynamically updated, based on accumulating experience. The effect of this dynamic update of attributes and their distribution, combined with the effect of learning (which by itself is characterised by bounded rationality) might be stronger than loss aversion and other features of PT. Evidence to support this argument are presented in Avineri and Prashker (2003) and Ben-Elia et al. (2008). Avineri and Prashker (2005) showed that CPT fails to predict route-choice feedback-based decisions (although a dynamic generalisation of the CPT static model may provide valid predictions of travel choices). Ben-Elia and Shiftan (2010) also did not find choice behaviour fully consistent with PT (while only the mean travel time was tested as a reference point, absolute travel time differences were found significant).

Another open question associated with the dynamic nature of travel behaviour is how the reference point is updated through time as a function of the outcomes in past decisions (Arkes, Hirshleifer, Jiang, & Lim, 2008) – so far this issue has not been addressed in the travel behaviour context.