Accounting for Supply Chain Structures in Modelling Freight Mode Choice Behaviour
K Arunotayanun, J Polak, Imperial College London, UK
This paper addresses shortcomings in modelling the behavioural responses of freight suppliers, in particular those associated with accommodating the complexities of supply chains, the insufficiency of data, and the use of inappropriate model forms.
The study of choice behaviour in freight transport faces a number of challenges which set it part from the study of choice behaviour in passenger transport. One principal difference is the far greater complexity of freight transport systems, which results from the enormous diversity of commodity and firm characteristics. Moreover, in recent years, developments in the freight transport industry have lead to a substantial transition of freight agents? operations from a simple, peer-to-peer, stand-alone structure to one that often involves complicated interactions amongst many different agents within complex supply chain structures. However, existing approaches to the empirical analysis of freight demand have largely ignored the influence of supply chain structures. Against this background, this paper presents a new theoretical approach to accommodating supply chain structure in models of freight mode choice and illustrates the application of this method using data from a recent commodity flow survey.
The paper is divided into a number of sections. The first section presents a brief overview of the existing freight demand modelling literature, focusing in particular on mode choice decisions. In the second section, a number of alternative theoretical approaches to accommodating supply chain structure within discrete choice models of freight demand are also presented. These approaches were drawn from passenger transport studies concerning the interdependency among memberships in the household and from freight transport studies of which only strategic frameworks accounting for joint choice decision across agents and supply chain networks have been suggested but yet to be empirically explored. The advantages and disadvantages of each of these approaches are discussed.
The third section provided a brief overview of the data used in the study. These data come from the 2004 French shipper survey (ECHO). A unique feature of the ECHO dataset compared to other national freight transport data sources is that the shipments were individually monitored from their origins to final destinations, of which all related agents were interviewed. That means it includes the information of physical and flow characteristics of individual shipments, characteristics of shippers and transport operators related and especially organisational supply chain structures. These data provide a unique opportunity for some of the theoretical models of supply chain influence to be implemented and tested.
The fourth section reports the results, which focus on modelling shippers? choice of modes using multinomial logit (MNL), nested logit (NL) and cross-nested logit (CNL) models, of which the last two were used to account for possible patterns of correlation in the unobserved utilities among alternatives. To explore the influence of supply chain structures, two separate NL model structures were then developed; one is to account for the heightened correlation between transport modes sharing the same supply chain structure and another is to account for the correlation between supply chain structures sharing the same mode. The CNL model was also used to account jointly for the correlation along these two choice-dimensions. The results indicate that some gains in model performance can be obtained in the nesting by supply chain structure. The use of the CNL model also leads to some further gains in model fit. Moreover, through each stage of our analysis, the effects of different model specifications and of the use of models accounting for inter-alternative correlations were revealed via the changes in the trade-off between cost and time and via the estimations of market elasticities. The resultant analysis of the cross-elasticities of choice performed under the CNL models also allows us to gain much further insight into shippers? choice behaviour with respect to transport modes and supply chains.
The final section presents overall conclusions and highlights future research directions.
Association for European Transport