An agent-based framework for modelling social influence on travel behaviour

Recent travel forecasting models have focussed upon the fact that travel is derived from the activities in which people participate, such as work, school, shopping, sport, leisure, and social events. Non-discretionary activities such as work and school may be explained by the traveller's sociodemographic characteristics and generalised travel costs, as well as long-term decisions such as a decision to move to a particular town. Participation in social and leisure activities is determined by one's friends and the groups that one is a member of, i.e., their household, their workplace/school, sporting groups, voluntary organisations and clubs. These acquaintances form part of an individual's social network: a representation of the people one interacts with. This demonstrates a shift in the activity-travel modelling field from “where” to “what” and now towards “who with”. On top of this, our changing use of ICT is influencing our activity and travel patterns, as some activities can now be replaced by online activities, and online activities can lead to actual travel. Some researchers are already looking beyond households to the influence of social networks. However, we are not aware of any agent-based urban models considering activity-travel choice of individuals. Existing work is in the conceptual or early implementation phases. The aim of this project is to develop and validate a model combining social (“who with”) and spatial (“where”) networks for investigating and predicting social activities. In this paper, we describe the design of our model. Agent-based modelling is a good fit for our model. Our system consists of different people, their relationships and interactions with each other, and their activities in and possible movement around the environment. The topology is not homogeneous and clusters may form. We have used a combination of the metamodels found in mature agent-oriented software engineering methodologies to design our model, focussing on system goals, the environment, acquaintances, roles, and services. The design successfully caters for the description of the environment, the nature of activities, and the dynamics of individuals and their networks. The individuals in our model each have an agenda, and interact and negotiate with others to schedule social activities, in particular negotiating about the nature of the activity, participants, time, and location. Existing models do not capture the actual joint decision making process behind activity scheduling, and although some work on joint decisions has been undertaken, these models focus on outcomes of interactions within households and have not considered personal social networks at large. We use existing multi-issue negotiation theory to describe an interaction design, which is shown to satisfy a number of basic properties, such as termination, liveness, and safety. Due to the current interest in predicting social activities and the changing nature of social activities due to our use of ICT, this type of model is of increasing importance to planners who need to be able to predict social activities and travel. The model is currently being implemented in Java, and will be validated using an extensive dataset of people's activity participation and personal networks, collected in Eindhoven, Netherlands. Future work includes more empirical experimentation with the protocols and implementation of and experimentation with the entire model.