Theoretical Issues in the Development of a RUM of Inter-agency Choice, with an Application to City Logistics
G Paglione, R Batley, J N Ibáñez, Institute for Transport Studies, University of Leeds, UK
We develop a new theoretical framework for representing inter-agency choice within a discrete choice model, demonstrating its consistency with the paradigm of RUM. We illustrate this framework by means of a worked example based on an UDC scheme
Whilst politicians have shown considerable enthusiasm for urban goods transport - recognising the stimulus it offers to the development of cities (the so-called ?economies of urbanization?) - there is growing acceptance that this must be offset against less desirable implications for the environment, congestion, the incidence of accidents and energy consumption. The motivation for this paper arises from the proposition that many of the latter implications are symptomatic of an economic failure: that is, a failure associated with disequilibrium among the various agents comprising the supply chain. More specifically, there is an evident lack of co-ordination between the relevant stakeholders (producers, shippers, carriers, retailers, transport operators), and this manifests in inefficiency which is amplified throughout the supply chain. Hence the principal contribution of this paper is to propose a model that represents the preferences of each agent within the supply chain, as well as the interaction between the preferences of these agents. We develop this representation within the paradigm of the Random Utility Model (RUM), thereby facilitating practical application to valuation, demand forecasting and economic appraisal.
Review of methods
Having defined the context for our work, the paper presents a review of methods. The literature of game theory is vast, and our review therefore focuses on a particular interest in applying game theoretic concepts to discrete choice models. Transport analysis is of course well-versed in the use of discrete choice models, given their amenability to data on the repeated choices of an individual traveller (as in stated preference analysis) or data on a single choice by a sample of individual travellers (as in revealed preference analysis). Following perhaps the suggestion of Manski [Manski, 2000], there has been an increasing trend to include social variables, and in particular ?influence? variables, in discrete choice models [e.g. Brock and Durlauf, 2001; Soetevent, 2006; Kooreman and Soetevent, 2002 ; Aribarg et al., 2002; Dellaert et al., 1998; Dosman and Adamowicz, 2002]. Influence variables represent the relative weightings of different agents when their preferences exhibit interaction. Furthermore, there has been a small number of papers that seek to more explicitly integrate game theoretic concepts within discrete choice models [e.g. Bresnahan and Reiss, 1991; Kooreman, 1994]. Our review devotes particular attention to the corpus of work by Hensher and colleagues [Hensher and Chow, 1998; Brewer and Hensher, 2000; Hensher, 2002; Hensher and Rose, 2004; Hensher and Puckett, 2004; Hensher and Puckett, 2005/a; Hensher and Puckett, 2005/b], since this proposes and applies game-theoretic discrete choice models to a specific interest in urban goods distribution.
Critique of methods
Arising from our review is a methodological critique. A particular focus of this critique is our aspiration to ensure the validity of game-theoretic discrete choice models in terms of microeconomic theory. This is not an abstract concern, since microeconomic theory equips us with the apparatus to apply such models to our practical interests in valuation, demand forecasting and economic appraisal. The critique identifies an unwelcome - and potentially damning - property common to many of the models cited above. That is to say, the interest in merging game theoretic concepts with discrete choice models often engenders the so-called ?Mother Logit? property [McFadden et al., 1977]. According to this property, which derives from an attempt to accommodate similarities between discrete alternatives, the utility of an alternative is the function not only of its own attributes, but also the attributes of other alternatives. An unfortunate implication is that Mother Logit is not consistent with RUM [McFadden and Train, 2000].
A new theoretical framework, with application
Guided by our critique, the paper then develops a new theoretical framework for representing inter-agency choice within a discrete choice model, whilst ensuring consistency with microeconomic theory (and RUM). We suggest that this can be achieved through exploiting the vehicle of the Network GEV model [Daly and Bierlaire, 2006]. Daly and Bierlaire show that this structure is consistent with McFadden?s GEV requirements and is therefore consistent with RUM. Whilst conceived originally as a representation of individual discrete choice, we demonstrate that, through creative application, Network GEV might be re-constituted as a model of inter-agency choice. Accepting this proposition, Network GEV brings a potential for developing our interest in modelling interaction between decision-makers, whilst adhering to the RUM paradigm. Having developed our framework, and articulated its theoretical credentials, we illustrate by means of a worked example based on a conceptual urban distribution centre scheme.
Association for European Transport