Multidimensional Travel Decision-Making: Descriptive Behavioural Theory and Agent-Based Models

Chenfeng Xiong, Xiqun Chen and Lei Zhang

Abstract

Purpose – This chapter explores a descriptive theory of multidimensional travel behaviour, estimation of quantitative models, and demonstration in an agent- based microsimulation.

Theory – A descriptive theory on multidimensional travel behaviour is conceptualised. It theorizes multidimensional knowledge updating, search start/ stopping criteria, and search/decision heuristics. These components are formulated or empirically modelled and integrated in a unified and coherent approach.

Findings – The theory is supported by empirical observations and the derived quantitative models are tested by an agent-based simulation on a demonstration network.

Originality and value – Based on artificially intelligent agents, learning and search theory, and bounded rationality, this chapter makes an effort to embed a sound theoretical foundation for the computational process approach and agent-based microsimulations. A pertinent new theory is proposed with experimental observations and estimations to demonstrate agents with systematic deviations from the rationality paradigm. Procedural and multidimensional decision-making are modelled. The numerical experiment highlights the capabilities of the proposed theory in estimating rich behavioural dynamics.

Keywords: Learning; search; imperfect knowledge; decision rules; hidden Markov chain

Background

The study of travel demand estimation, forecasting and adjustment has long been a vital topic in the field of transportation planning. Being an induced demand, travel demand is often regarded as the product of other activities. Individuals commute to work, drop-off family members, travel for leisure, fly to customers/suppliers, visit relatives/friends and so forth. While these activities are often differentiated by locations and time, how these spatial/temporal details can be accounted for becomes an essential question for transportation planners and researchers. Moreover, these activities encompass interrelated travel decisions including destination, mode, departure time and route. Therefore, the complexity arising from the mutual effects of these multidimensional decisions upon each other and from their decision timing needs to be represented.

Traditional travel demand modelling structure distinguishes four decision dimensions: deciding the frequency of travel, choosing a destination, selecting a travel mode and travelling via a route. These decision dimensions are assumed to follow a predefined sequential manner of trip generation, trip distribution, modal split and traffic assignment, as known as the Tour-Step' method. Travel behaviour research gradually moved from aggregate demand models to more disaggregate individual- level and activity-based models (Bhat & Koppelman, 1999; Bowman & Ben-Akiva, 2001; Yovsha, Donnelly, & Gupta, 2008; Xiong & Zhang, 2013b; Zhang, Southworth, Xiong, & Sonnenberg, 2012). While the majority of interest focuses on advancing single-dimensional (single-facet) choices and more advanced representation of activity pattern such as scheduling (Bowman & Ben-Akiva, 2001; Golledge, Kwan, & Garling, 1994), land use influence (Salvini & Miller, 2005) and location choices (Bhat & Guo, 2007), the linkages among different travel behavioural dimensions are largely ignored (Pinjari, Pendyala, Bhat, & Waddell, 2011) and individuals' embedded behavioural processes that influence them to change certain dimension(s) of their travel behaviour remain unexploited.

Besides the rigid sequential assumption, travel demand models also rely on other simple and sometimes unrealistic behavioural assumptions in order to keep themselves analytically tractable. Perfect rationality theory is one of the well-known assumptions that individuals are fully rational, have perfect information and always maximise utility (Savage, 1954; von Neumann & Morgenstern, 1947). Being an approach with rich results, mathematical rigour and interesting applications, perfect rationality and utility maximisation allow structural insights and explain similarities and differences in travel behaviour. However, if using this theory to calculate how certain variations in the situation are predicted to affect travel behaviour, 4these calculations obviously do not reflect or usefully model the adaptive process by which subjects have themselves arrived at the decision rules they use' (Lucas, 1986).

The opposite holds true for the computational process models, a group of new methods that depart from rationality assumptions and implement learning, adaptations, information acquisition and decision-making efficiently by taking the advantages of computer power. These models are microsimulations relying on heuristic arguments and imitation of human behaviour. A large number of real-world or benchmark problems can be analysed by applying these models to simulate numerical results in different set-ups. Examples on the rapidly growing list include Pendyala, Kitamura, and Prasuna Reddy (1998), Arentze and Timmermans (2004), Wahba and Shalaby (2011), and Auld and Mohammadian (2012). On the one hand, these models introduce more complex learning, adaptation and behavioural rules instead of utility maximisation. But on the other hand, multiagent simulation cannot prove but only suggests a certain feature of travel pattern and still assumes sequential decision process. Thus it requires additional theories to conceptualise the more rigorous behavioural foundation and better explain behaviour adjustments along multiple choice dimensions (see Arentze, Hofman, & Timmermans, 2004; Pinjari et al., 2011).

Urged by the above-mentioned theoretical and modelling issues, this chapter describes an alternative framework to modelling multidimensional aspects of travel behaviour. Descriptive theory and models are built upon economics and travel behaviour research on learning (Arentze & Timmermans, 2005; Golledge et al., 1994), search theory (Stigler, 1961) and bounded rationality (Mahmassani & Chang, 1987; Simon, 1955). The theory recognises that there are inconveniences and risks associated with each behaviour adjustment dimension, which is conceptualised as a search cost unique to each individual and each behaviour dimension. On the other hand, an individual, based on his/her spatial knowledge, personal travel experiences and beliefs, forms subjective expectations on potential gains (search gain) from behavioural adjustments along each behavioural dimension. It is the interplay of these search gains and search costs along all feasible behavioural adjustment dimensions that collectively determine when individuals start seeking behaviour changes, how they initially change behaviour, how they switch behaviour adjustment dimensions, and when they are satisfied and stop changing behaviour. The theorisation of multidimensional knowledge updating, search model and behaviour process becomes a unified and coherent approach that models the activity and travel decision-making with a consistent behavioural foundation and increased rigour. The theory is supported by empirical observations and the derived quantitative models are tested by agent-based simulation on a demonstration network. For improved readability, emphasis is given to modelling knowledge and multidimensional search. Certain details about search rules and decision rules are excluded. Readers interested in these topics are referred to appropriate references.