Applicability of an Extended Prospect Theory to Travel Behaviour Research: A Meta‐Analysis

Abstract

In transport research, choice behaviour is commonly investigated by using discrete choice models that comply with the utility theory (UT) paradigm. The objective of this article is to assess the extent to which the assumptions of an extended version of prospect theory (PT) allow for a better description of travellers’ choice behaviour than UT. To that aim the travel behaviour literature was reviewed for information about observed choice behaviour suitable for such a comparison. The review yielded 106 studies that covered almost the whole range of domains and contexts of travel behaviour. Compared to the corresponding assumptions of UT, a meta‐analysis showed a better descriptive performance of reference‐dependent framing connected with loss‐aversive valuation, and of interpersonal heterogeneity in choice behaviour strategies. Diminishing sensitivity approached the observed behaviour better than the sign‐independent utility function of UT in several studies where a comparison was feasible. In choice under risk or uncertainty, the non‐linear weighting of probabilities in connection with loss aversion appeared to offer a better description than expected UT. It is concluded that the joint application of the assumptions of an extended PT might provide a better understanding of human choice behaviour over the whole range of travel‐related contexts.

Acknowledgements

This article benefited from reviews of draft versions by Bert van Wee and five anonymous referees.

Notes

1. PT was proposed as a descriptive theory for choice between alternatives with probabilistic outcomes. Later, Tversky and Kahneman (1991) elaborated their theory for choice under certainty, which may also be encountered as reference‐dependent theory (e.g. De Borger and Fosgerau, 2008), and proposed CPT to accommodate a wider range of uncertain choice situations (Tversky and Kahneman, 1992).

2. In most publications, Kahneman and Tversky use the term ‘reference point’. For the evaluation of multi‐attribute options, they used ‘reference state’ (Kahneman and Tversky, 1984, p. 346) which is followed here.

3. The terms used for the internet query might yield an excess of studies that aimed to establish PT or one of its features. However, PT was not considered in the design of the choice data acquisition process in 89 out of the 106 studies. Fourteen studies were more or less deliberately designed to allow a comparison of the corresponding premises of UT and PT, but no signs were found that in any of these the researchers manipulated the design to arrive at a better performance of PT compared to UT.

4. Carrying out a classic meta‐analysis or a data‐mining approach by comparing all three corresponding assumptions of each assumption set against the choices of all considered individuals was considered. This was not feasible as the required choice data at the individual level were only rarely available.

5. If convincing or plausible evidence is found for a better descriptive ability of one of the EPT assumptions compared to its UT counterpart, this does not rule out that the UT paradigm offers a fair description and understanding of the choice behaviour observed in the context being considered. In fact, if this had not been true, by far most re‐examinations that supported an EPT assumption would have been impossible because they drew on the results of UT‐conformable parameter estimation. However, when any of the EPT assumptions might offer a better explanation of choice behaviour in a particular context this indicates that one should be very careful when extrapolating the results found in agreement with the UT paradigm to other contexts.

6. That is, fictitious personal circumstances which a participant is asked to adopt in performing a choice task.

7. ‘Risk’ is commonly used in the economic literature as synonymous to ‘loss‐neutral probability’. A 0.1% probability to get ill and a 0.1% probability to win a lottery is considered as the same risk level.

8. If the relative descriptive ability of the assumptions of EPT compared to UT does not differ between countries, this does not rule out systematic intercultural differences in adopted choice behaviour strategies. See Wang and Fischbeck (2004) for an illustration of intercultural differences in the framing of the reference state in a health insurance choice setting.

9. Strategic choice behaviour concerns the long‐term choice processes like residence location, employment issues or car ownership that may last from days to months and in which affective valuation often plays a decisive role. Operational choice behaviour refers to the mental processes that precede and govern concrete everyday actions like departure time postponement or lane switching that may last for minutes or a split‐second. Tactical choice has an intermediate position between strategic and recurrent operational choices. It is considered to establish mental scripts or habits that govern the patterns of operational activity planning, like commuters’ ‘default’ home departure times. It also accommodates constructs like willingness‐to‐pay or value‐of‐travel‐time.