Analysing and visualising areal crime data. A case study of residential burglary in San Francisco, USA

Dissertation submitted in partial fulfillment of the requirements for the Degree of Master of Science in Geospatial Technologies.

Methods to visualise and analyse areal social data are limited. A traditional approach is Choropleth mapping. However, the rates on which these maps are based can be unreliable in sparsely populated areas, and there may be visual bias when areas are irregularly sized. Another common method is to perform point interpolation at the centroids of the areas. This approach may only be valid when areas are regularly shaped and small. This thesis explores how Area-to-Area and Area-to-Point kriging can be applied to analysing and visualising residential burglary rates in San Francisco, United States. Results are compared to the traditional methods used to analyse areal data. Additionally, the study investigates burglary hotspots and the relationship between socio-economic variables and burglary in the study area by conducting spatial and non-spatial regression analyses. The study concludes that Area-to-Area and Area-to-Point Poisson kriging methods may improve on existing approaches to interpolating areal crime data. The visualisation of areal data is improved through the smoothing of rates based on small denominators, and visual bias may be decreased by using Area-to-Point kriging. Using the kriging estimates of these techniques as inputs into hotspot and regression analyses provides a useful way in which to explore relationships at different scales. However, caution should be exercised when utilising these methods due to their limitations.