Creating Realistic Synthetic Populations at Varying Spatial Scales: A Comparative Critique of Population Synthesis Techniques

Recent years have seen a rise in the number of methods and applications which require realistic individual-level data/synthetic populations. This trend can be attributed to a number of factors including increases in computational power and storage, a wealth of individual level data (for example, the British Household Panel Survey) and the development of new computational paradigms, such as cellular automata and agent-based modelling (ABM). Static spatial microsimulation samples a synthetic population (a population built from anonymous survey data at the individual level) which realistically matches the observed population in a geographical zone for a given set of criteria. There is a diverse set of research and policy applications that use synthetic populations in a spatial setting, including: health (Smith et al, 2009, Tomintz and Clarke, 2008, Brown and Harding, 2002), transportation (see, for example, McFadden et al, 1977; Beckman et al, 1996) and water demand estimation (Williamson and Clarke, 1996). ABM can also use synthesised data as a base population. There has been a rapid uptake in the use of ABM in Geography with applications ranging from simulating the movement of burglars (Malleson et al, 2009) to replicating dynamics in spatial retail markets (Heppenstall et al, 2006). Although the construction of an ABM does not require a complete individual data set, creating an agent population from a realistic synthesised individual dataset can only improve the realism of these models. There are several established methodologies for generating synthetic populations. The focus of this paper will be on deterministic reweighting (Smith et al, 2009), conditional probability (Monte Carlo simulation) (Birkin and Clarke, 1988, 1989) and simulated annealing (combinatorial optimisation) (Openshaw, 1995; Williamson, Birkin and Rees 1998; Voas and Williamson, 2000, 2001). These methods were selected due to their common application in geography. Many recent spatial microsimulation studies including Anderson (2007), Ballas et al. (2005), Voas and Williamson (2000, 2001), Tomintz et al. (2008) Smith et al. (2009) and Morrissey et al. (2008) have adopted a variation on at least one of the three approaches examined here. GeoComputation 2011