Traditionally social science models have been either static (in the sense that they replicate or predict a single time slice of a social system), or have dynamics that work with a ‘fixed rudder’ (in the sense that they are initiated and then run over time to replicate or predict some end-point, with no opportunity for adjusting the running model). This is actually rather unusual for geographical models as both these model pathways have considerable error issues. In the case of static models, there is almost no way to appropriately check the error involved in new predictions. In the case of ‘fixed rudder’ models, the errors involved in modelling non-linear systems ‘explode’ as the model runs, making most predictions extremely dubious (the potential error variance is generally far larger than the potential range of model results).
As an example: based on qualitative hotspot analysis there is a strong suspicion within crime-prevention organisations in Leeds (UK) that burglary shifts during the day from the city centre to the suburbs, however, it is unclear whether this is driven by shifts in occupiers (the patterns of homeowners over the working day), school children (juveniles moving to schools and then truanting), or policing patterns. In addition although the hotspots involved are large and clear, it is far from clear whether there is a direct and negative auto-correlation between them spatio-temporally; whether the volumes of crime are directly relatable; or whether multiple alternative crime locations of lower significance are also being exploited. Finally, it is not clear whether the same individuals are involved, or whether there is some division within the criminal community.
In most communities modelling real-world systems these issues have been addressed through data assimilation: the adjustment of the running model with dynamically arriving real-world data. For example, in meteorological models new air pressure, temperature, and other field data are integrated into running models, and the current errors used to recalibrate the running model. Social science models, however, have generally not taken this step, in part because of limited access to continuous datasets. Concentration on datasets such as decadal censuses, and the continual development of new collection schemes for quasi-longitudinal datasets, has made data assimilation to all intents impossible. However, the last five years has seen an explosion in the availability of continuous and spatially-linked datasets, both formal (for example, economic time series) and informal (for example crowd-sourced geolocated twitter feeds). Social science is now in a position to better constrain model errors using dynamic data assimilation, indeed, it is essential if socio-economic modelling is to fulfil its considerable potential. However, social scientists are now so tied into a small number of static modelling traditions that this is often hard to see.
This PhD will utilise methodologies found in other modelling communities to enhance socio-economic agent-based models (for a review of the techniques, see Evans, 2011). It will additionally develop new techniques to visualise the evolution over time of error surfaces, and investigate the unique opportunities agent-based modelling can bring to bear on the issue of error propagation in non-linear systems (most notably the accurate replication of those relationships in human society that tend to dampen the propagation of instabilities – social negotiation, compromise, group decision making). The example system used will depend on student interests, however, could be in the areas of retail pricing, crime, or the housing market.
The PhD would be suitable for anyone with a computational, mathematical, or physics-centred background, or anyone with a geographical, geological, meteorological, environmental, or sociological background and a strong interest in modelling human or physical environments.
For information on funding opportunities click here
For project related enquiries please contact the supervisors.
For application enquiries please contact Jacqui Manton