University of Surrey Institute of Advanced Studies

Modelling Urban Social Dynamics

Workshop Report

Policymakers and academic researchers have a longstanding interest in understanding how cities evolve and change. However, conventional methods of modelling urban development have not been able to capture the complexities and historical particularities of urban social dynamics. Recently, new forms of computational modelling, including agent-based modelling, have begun to be used, but as yet there is no agreement about how such models should be constructed and evaluated, and the kinds of task for which they are best suited. This inter-disciplinary workshop provided a forum for current work in this area, bringing together the leading researchers to discuss these developments.

The workshop was held at the University of Surrey, UK, on Thursday and Friday, 7-8 April. There were 23 participants, from China (1), Columbia (1), Germany (2), Hungary (1), Ireland (1), Israel (1), New Zealand (1), The Netherlands (6), United Kingdom (5) and the USA (4). Twelve papers were presented, summarised below.

The papers

In the first paper presented at the workshop, O'Sullivan and MacGill motivated their research with the observation that neighbourhoods are key points for policy intervention. For example, neighbourhood revitalization projects are often followed by changing perceptions of the 'image' of the neighbourhood, which can lead to further investment or gentrification. However, neighbourhood boundaries may change over time and residents do not always share the same perceptions of neighbourhood boundaries. David O'Sullivan described a computational model of Auckland, New Zealand in which an attempt was made to use 'fuzzy' neighbourhood boundaries and a hierarchy of local and regional neighbourhoods.

Benenson described models to explain residential mobility and segregation in Israeli cities. His models rely on unique census data that included both household locations and residential demographics (e.g., income, ethnicity, etc.). Inspection of the data using a GIS revealed a great deal of ethnic and economic heterogeneity within neighbourhoods. This finding is contrary to expectations based on traditional models of neighbourhood change (such as the residential tipping model). The location patterns could not be explained by correlating factors such income, number of children in the household, building age or type of house. Much greater explanatory power was obtained using a Schelling-type simulation model in which there were asymmetric relationships (one ethnic group is more tolerant than the other) and modifying the model's implementation to a 'take the better' migration choice algorithm.

Waddell's paper discussed the problem of endogeneity in simulation modelling. Many urban processes (e.g., residential choice, the choice of firms about where to locate, and the availability of public goods and services) interact dynamically. Thus, it is difficult to isolate the effects of different factors on one-another. Waddell described an evolving effort to represent key endogenous processes in the simulation model UrbanSim. This model is currently being implemented in Paris.

Bruch reviewed the Schelling model and noted that the original model made assumptions about the form of the household decision function. She demonstrated that clustering would only emerge for discontinuous functional forms, while data for surveys suggested that people's actual decision functions for race were continuous. This left a puzzle: if residents' decision functions are continuous, why do we observe segregation? She then showed that using income instead of race as the sorting factor also did not lead to clustering, but if it were assumed that both race and income were significant, segregation would appear.

Pancs and Vriend used game theory to examine the robustness of Schelling's results to alternative assumptions about individual behaviour. They showed that even if all agents have a strict preference for perfect mixing, best response dynamics will lead to clustering. They derived analytically all 'myopic' Nash equilibria for a selection of decision functions, but showed that the best response for several of the decision functions did not converge at any myopic Nash equilibrium.

In contrast to the more theoretical bent of the first set of papers, the second set typically applied simulation techniques to the analysis of a real-world social problem or policy question.

O'Donoghue, Ballas and Lennon used a spatial microsimulation model to predict migration decisions between and within Irish District Electoral Divisions (DEDs) and the impacts of economic development and policy changes in rural areas. The model relies on data from five successive censuses (19981 - 2002). They use the model to identify a range of variables which are correlated with the growth of DEDs.

Haase, Holzkämper, and Seppelt used a GIS-based landscape simulation model to study urban shrinkage and demolition patterns in Leipzig, East Germany. While urban growth has been a longstanding area of interest in urban studies, the paper extends previous work to examine processes of urban shrinkage and building demolition, and relates this phenomenon to associated demographic changes (e.g., the declining birth rate). Using their microsimulation model, the authors run a scenario analysis to evaluate the implications of alternative policies proposed in Leipzig.

Eskinasi et al. relied on dynamic modelling techniques to adjudicate among competing arguments about how to make affordable housing available to the residents of the Haaglanden region in the Netherlands. The authors use system dynamics to model the relationships among housing allocation, new housing construction, the destruction of deteriorating buildings, and the regional housing market. The results from the model were used to inform housing policy in Haaglanden.

Makowsky used an agent-based model to examine the relationships among urban crime, mortality, and exogenous population shocks. The model demonstrates how expectations about life expectancy affect decisions about criminal behaviour. Using this model, the author simulated how a 'shock' to the population affects criminal behaviour in later years. The author's results suggest that the effect of morbidity shocks on life expectancy (and consequentially, criminal behaviour) can persist even 100 simulated years after the shock occurred.

Hulse et al. used a spatially explicit simulation tool to model how alternative policies about resource conservation and development will affect future land use and natural resources in the area surrounding the junction of two rivers in Oregon, U.S.A. The authors conclude that, in contrast to the current policy goals of Oregon land use planning, it is possible to implement conservation policies that can preserve vegetation for fish and other wildlife or agriculture, but not both.

Gil Quijano et al. described an agent-based model to study residential mobility in Bogotá, Columbia. The paper's main goal was to incorporate information at multiple levels of analysis. At the micro-level, the model includes households and housing units. At the meso-level, these micro-level units are organized into groups based on their identifiable characteristics. The population (both households and housing units) evolves according to general rules of evolution. The paper demonstrates that in certain circumstances the use of micro-level agents is not necessary; dynamic behaviour can be adequately captured by interactions among groups at the meso level.

Finally, Dugundji and Gulyas used agent-based modelling to study choice outcomes when individuals' behaviour is interdependent. Their model first examines outcomes that occur assuming all agents respond to the behaviour of all other agents. Next, the authors introduce two formulations of networks (a “random” graph and a “small world” graph) to introduce heterogeneity in the interactions/influence among agents. The authors demonstrate that the inclusion of the agent itself in sparse networks can dramatically affect the results. Moreover, the authors show that the estimation process to determine sets of coefficients can be highly sensitive to small variations in network instantiation.

Workshop themes

The papers could be divided into those that took a theoretical model (e.g., Schelling's model of residential tipping) as their starting point and studied its implications and extensions in order to better understand some form of behaviour (i.e., residential or transportation choice or the decision to engage in criminal behaviour); and those that took an urban policy related problem as the starting point, and applied an agent-based or microsimulation model to understand better the factors affecting location decisions. Papers in the former category were: O'Sullivan and MacGill, Beneson, Bruch, Vriend, Makowsky, and Dugundji and Gulyas; and in the latter category were: Waddell; O'Donoghue et al., Haase et al., Eskinasi et al, and Quijano et al..

A number of themes also emerged in the course of the discussion following each presentation. These themes included:

  • The influence of social networks
  • o The use of dynamic models to study spatial inequality
  • Urban transformation (not just growth, but also shrinkage)
  • Under what circumstances agent-based or microsimulation models are appropriate
  • How to achieve robustness in model results
  • Model validation
  • The aggregate implications of individual residential choice functions and algorithms
  • The definition and identification of 'neighbourhoods'
  • Modelling inter- and intra- urban changes

Next steps

There was general enthusiasm for continuing the discussion initiated at the workshop in another forum:

  1. A follow-up workshop at the EGOV05 conference in Copenhagen (22-26 August, 2005) had been mooted, although some thought that this followed the Surrey workshop too quickly to allow for substantial work in the interim.
  2. A couple of the participants suggested that they might be interested in organizing or contributing to organizing another workshop in 12 or 18 months' time.
  3. There was a discussion of the benefits of attempting to secure publication of the papers either in an edited book or as a special issue of a journal. Journal publication was strongly favoured over a book by those present. The convenors of the workshop agreed to negotiate with the editors of appropriate journals and report back to authors by the end of June.
  4. Meanwhile, the presenters agreed to revise their papers in the light of the discussions at the workshop in time for submission to a potential special issue, again with a deadline of the end of June 2005.

Acknowledgements

The organizers and participants are very grateful to:

  • The Institute of Advanced Studies of the University of Surrey and the European Commission's Network of Excellence for Complex Systems, EXYSTENCE, for their financial support;
  • The programme committee
    • Itzhak Benenson (Tel Aviv)
    • Elizabeth Bruch (UCLA)
    • Nigel Gilbert (Surrey)
    • David O'Sullivan (Auckland)
    • Nicolaas Vriend (Queen Mary College, London)
    • Paul Waddell (Washington)
  • The Institute's administrator, Madeline Brownett, for her superb organisation

Centre for Social Simulation, School of Human Sciences,
University of Surrey, Guildford GU2 7XH, UK
Tel: +44 1483 300800
http://cress.soc.surrey.ac.uk/