Modelling approach

The advantages of the Individual Based Models (IBMs) to model system changes from the level of individuals have been described in Chapter 2. Two different approaches, ABM and MSM, are used in this model. Based on the discussion of each of their strength and limitations previously in chapter 2, the MSM and ABM approaches are found to be complementary to each other in both theoretical and practical aspects. Details of how these methods are implemented in the model are described in the sections below.

3.3.3.1 MSM

As a static MSM only represents the population at a given time (Pudney 1994), where no characteristics of the micro units are changed during the process, a dynamic MSM is used in this study. This is mainly because a dynamic MSM cannot only update the characteristics of the micro units caused by the stimulation of endogenous factors, but can also project them over time to include demographic processes and social economic transitions, such as ageing, mortality, fertility or social and geographical mobility (O’Donoghue, 2001). On the other hand, small area differences play an indispensable role in modelling the population changes in this study and they also pose a substantial impact on various planning applications. Therefore a spatial MSM is used to model the individual changes with the local context. Although a dynamic spatial MSM is used for modelling the population and its dynamics to capture details at the individual level, the model structure parallels the macro multi-state cohort-component projection model. Populations are therefore simulated at the individuals (MSM) within small areas of ward (spatial) through various demographic processes, where characteristics of each of them are updated on the basis of transitional probabilities each step of the simulation (dynamic).

Being a widely applied instrument in studying and predicting the evolution of population, MSM is as important to the analysis of event histories as macrosimulation is to traditional aggregated demographic analysis. However, there is a gap between the micro-demographic theory and

demographic microdata and modelling techniques (Billari et al., 2002). Quite often, there is an unavailability of appropriate microdata for the modelling a specific important demographic transition in an MSM, due to its high standard of requirements of the data at the individual level. Also in a traditional demography model, there is a limitation of precision in theoretical constructions and often lacks theory at the basis of the applications of statistical models and data collection at the level of individuals (Billari et al., 2002).

3.3.3.2 ABM

Agent-based modelling (ABM) is an alternative approach that can model individual behaviours through multiple agents. In an ABM, each agent follows their built-in rules and acts/reacts according to such rules and the knowledge gained through interactions with each other and the environment that they live in. Through such interactions, simple and predictable local interactions can generate familiar but unpredictable global patterns, e.g. the formation of norm in an artificial society. This is called the emergent property of the ABM (Russell and Norvig, 1995). With such features, ABM provides theoretical leverage to model a complex social system where the global patterns of interest are more than the aggregation of individual attributes.

Although there are limited examples of using ABM approach in demographic models, Billari et al. (2002) consider ABM as a promising approach to help improve our understanding of demographic behaviours by studying demographic processes as the outcome of interacting agents. By focusing on dynamics of the population instead of equilibria, ABM is better suited for modelling specific processes. Migration is a complex demographic process where interactions and behaviours play an important role (Champion et al., 2002). Using ABM, individual activities and diversity of migration decisions leading to complex migration patterns can be simulated in detail. Espindola (2006) analysed the rural–urban migration using ABM, where the migration of workers is modelled as a process of social learning by imitation. As emergent properties of the model,

transitional dynamics are observed with continuous growth of the urban fraction of overall population towards equilibrium. While Loibl and Toetzer (2003) studied urban sprawl patterns through modelling suburban migration and residential area occupation. Distinctive migration behaviours of households with varying socio-economic status have been simulated in an ABM. Makowsky et al. (2006) build an ABM to simulate crisis-driven migration of agents within a multi-ethnic population. This study reveals that cultural networks temper an agent’s security calculus, with strong social ties dampening the human security dilemma.

However, compared to a MSM, an ABM is often built without the validation ambition and the individual rule-driven simulation can slow down the simulation when applied to a large population of individuals.

3.3.3.3 A hybrid approach

Based on the above discussion of the strength and limitation of ABMs and MSMs, a hybrid modelling approach is proposed in this section to bring the strength of both approaches together and address the limitations discussed. As described in Chapter 2, the origins of the hybrid approach can be traced from the time when the name of ABM had not been invented. Although mainly recognised for his pioneering work in spatial microsimulation of innovation diffusion and the conceptual framework for the analysis of spatial dynamics in a micro level time-space frame (Hägerstrand,1953; Holm et al. 2006), Hägerstrand has also brought the ABM concepts into the spatial MSM in order to explore the relationship between social contact and migration (Hägerstrand, 1957). In Hägerstrand’s spatial MSM, population and vacancies are evenly distributed in “migration fields” that are divided into cells. Then “active migrants” are distinguished from the “passive migrants”. The active migrants can randomly select and move to a destination in an adjacent cell, whereas the passive migrants are stimulated by the active migrants in a way that a passive migrant chooses a new cell where an earlier migrant from the same origin is staying and the attractiveness of all earlier migrants are equal. Although not a computer

based model, the migrants in this model demonstrate the basic “agents” characteristics of being able to interact with others and the environment. Since then, more discussions of using ABM in various types of models have been published and more modern attempts of the hybrid approaches have been made which confirm the potential of integrating the two approaches in modelling complex social systems (Conte et al., 1998; Axelrod, 2005; Boman and Holm, 2004).

In the current model discussed in the thesis, the MSM provide the theoretical basis with its roots of using real data and to be used for real application, providing valuable guidance for the projections. The statistical nature ensures the similarity between what it predicts and what is actually observed in the gathered data. Practically, its list processing power allows the model to simulate the detailed changes in large number of individuals with rich attributes. Most MSMs are also built with the validation ambitions and the alignment exercises have been well studied and widely practised in recent MSM research.

With great flexibility of the rule-driven simulations, the ABM provides a way to bridge the knowledge gap and data limitations when we study individual movements, interactions and behaviours. This model uses an ABM to model student migration where suitable microdata are not available and distinctive migration patterns have been found from the rest of the population. As agents can also carry their personal history and personal history sometimes can have an important impact on demographic changes, we also use ABM to explore the impact of personal migration history on mortality projections.

3.4 Data selection