The enhancement of spatial microsimulation models using geodemographics

Abstract

The generation of synthetic population estimates through spatial microsimulation has been a popular technique in recent years, with applications to research and policy problems in many areas of social science. Estimation techniques typically involve cloning or matching households in surveys with small-area census data. When model estimates are benchmarked against real-world data, the models are typically well behaved and very robust, but they can struggle to capture the diversity of spatial variations shown by observed data. We argue in this paper that this is the result of 3 potential problems in spatial microsimulation estimation techniques. The first issue results from the matching process in the estimation techniques, and the second problem relates to the variations of household types in the surveys being reweighted. Third, similar household types may show different behaviours or have different attributes depending on geographical factors not contained in surveys (such as the proximity of service or job locations). The aim of this paper is to demonstrate and measure the loss of accuracy and intensity induced by spatial microsimulation in the context of real individual data. It will be argued in particular that while the first two problems have begun to be addressed in the literature, the third issue is still largely unreported. The paper will thus suggest a solution framework which involves linking spatial microsimulation models with geodemographics and demonstrates the promise of this technique with real numerical experiments.