Accommodating Risk Attitudes in Freight Transport Behaviour Research

Abstract

Behavioural choice modelling is growing in interest as a framework to understand the decision-making of shippers, freight forwarders and other freight agents. Since the 1980s, we have witnessed an increasing number of freight behaviour studies, primarily the freight mode choice, where the roles of one or more freight decision-makers has been addressed, which was neglected in the traditional aggregate approach to freight demand modelling. Stated preference techniques have become a main approach to establishing the role of attributes that define the key drivers in the freight distribution chain. The underlying theory of choice is based on the neoclassical economic assumption that a decision-maker, in choosing, acts as if they are a utility maximizer (working to deliver a profit maximization or cost minimization outcome for the freight business), and this utility maximizing behaviour engenders a population theory of Random Utility Maximization (RUM). Despite the continuing appeal of the RUM framework in applied travel choice studies, a number of specific application assumptions have been questioned by studies in psychology and behavioural economics, arguing that the decisions made by agents are often conditioned on a number of underlying psychological components, one of which is risk attitude. This paper reviews recent freight behaviour studies established on RUM, and presents a major limitation of adopting a risk-neutral assumption through its linear utility specification. Using an existing freight stated choice data set, a nonlinear utility model is estimated which reveals risk-taking attitudes of transporters and shippers. An alternative behavioural paradigm, Rank-Dependent Utility Theory (RDUT), is introduced and incorporated to better accommodate trip time variability, a feature of growing importance in transport systems (passenger and freight). The proposed attribute-specific extended RDUT framework that accommodates the attitude towards risk and preference in freight transport behaviour modelling requires new data, which we detail.

Acknowledgement

The comments of two referees are greatly appreciated.

Notes

RP data is also used in some recent freight behaviour studies (de Jong and Johnson, 2009; Rich et al., 2009; Satar and Peoples, 2010).

Hensher et al. (2005) have suggested that a stand alone RP model can be used for forecasting as long as the constants are calibrated to RP shares. Another approach is to estimate a joint SP/RP model. Compared to its application in forecasting, SP is more often applied to obtain empirical estimates of willingness to pay (e.g. value of travel time savings).

Almost all reviewed studies in existing literature reviews were published before 2007.

“End-shipper is used to describe a shipper that hires carriers for all of their shipments” (Patterson et al., 2010, p. 3).

The practice of carrying trailers, semi-trailers or containers in a train atop a flatcar (intermodal transport) is referred to as “piggybacking”.

One Australian dollar (Au$) was equivalent to 0.76 US dollar in 2005.

Normally defined as the percentage of transport services arriving on time.

Masiero and Hensher (2010) applied the piecewise linear approximation technique to investigate “quasi-nonlinearity” which is introduced in the punctuality attribute identifying two decrease and two increase levels with respect to the reference point, while still maintaining the utility function linear in the parameter. Another commonly used approach is to assume a level of nonlinearity, e.g. squared attribute (x ² ).

Although a normal distribution can also be constrained and a lognormal distribution can produce all positive or negative individual parameters, they have some serious problems when estimating models (Cherchi, 2009). The form of the triangular distribution imposes the assumption that the means for the random parameters were restricted to be the same as the standard deviations.

In the broader literature, Rank-Dependent Utility Theory is not new, which was originally proposed in the early 1980s.

Outcomes can also be ranked from best to worst (Diecidue and Wakker, 2001).

We acknowledge that a recent Dutch passenger and freight value of time and reliability survey presented five equi-probable transport times and arrival times within the same alternative in a binary choice context (de Jong et al., 2007; Tseng et al., 2009).

We use three trip times as an example, and would suggest far more points on the distribution than three.