A rank-dependent utility approach to model intra- and inter-individual heterogeneity in risky choice behaviours

ABSTRACT

Using a stated preferences survey, the objective of this paper is to investigate the intra- and inter-individual heterogeneity of mode choice, when travel time is subject to variability. By‘inter-individual heterogeneity’ is meant that people are different in terms of attitude to risk and have different utility functions. By ‘intra-individual heterogeneity’ is meant that the behaviour may be different even when performed by the same individual when faced with a different mode of transport. Based on Rank-Dependent Utility Theory, the paper shows that the occurrence of delays associated with train trips is overestimated whereas they are underestimated for car trips. A latent-class logit model offers a somewhat different perspective: if, overall, car users are more likely to perceive possible delays for train trips than for car trips, train users tend to consider the objective occurrence of delays as they are presented in the survey and adopt a risk neutral choice behaviour.

KEYWORDS:

• Discrete choice model
• Mode choice
• Latent classes
• Rank-dependent utility

JEL CLASSIFICATION:

• C25
• D81
• D90

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ For appraisal of transport projects, the former is the only relevant option since it relies on statistical measures of variability (for instance, standard deviation). However, the microeconomic foundations of how travel time variability generates costs are hidden. In contrast, the scheduling model is microeconomically sound.

² Two focus groups of 13 persons and 11 individual interviews.

³ In prospect theory, formulas ‘turned out not to be sound because they imply violations of stochastic dominance in preferring less money to more money in manners that are not only normatively but also descriptively unwarranted’ (de Palma et al. 2008, 5).

⁴ The Prelec functional form (Prelec 1998) has also been tested but it provided inconsistent results, possibly due to the two parameters which have to be estimated with this function. Further research is needed to solve this issue. These three parametric functions – Power, T&K (or Quiggin) and Prelec – are also the ones considered in de Palma et al. (2008).

⁵ We are aware that it is a assumption to be further discussed. Panel structure will be accounted for in future applications.

⁶ The Bayesian Information Criterion (BIC) is not reported here as it yields the same ranking.

⁷ Based on theoretical assumptions, Beaud, Blayac, and Stéphan (2016) consider risk-averse decision makers with a concave value function. They therefore derive the opposite result: a VOT which increases with travel time.

⁸ The value of the threshold is obtained by resolving the equation $w(p_j)=p_j$, for $0<p_j<1$.

Additional information

Funding

This work was supported by the Rhône-Alpes Region (France).