Improving Geocoding Rates in Preparation for Crime Data Analysis

Problem-oriented policing requires quality analyses of patterns and trends in crime incidences. A common form of analysis is the identification of geographical clusters or ‘hot spots’. For such analyses, crime incident records must first be geocoded, that is, address-matched so as to have geographical co-ordinates attached to each record. The address fields in crime databases typically have omissions and inaccuracies whilst a good proportion of crimes occur at non-address locations. Consequently, geocoding can have an unacceptably low hit rate. We present and test an improved automated and consistent approach to batch geocoding of crime records that raises the hit rate by an additional 65 per cent to an overall rate of 91 per cent. This is based on an actual implementation for a UK police force. Kernel density surfaces used to visualise the results of the test show that the additional geocoded records have distinct spatial patterning. This would indicate that without the improved hit rate, geocoded crime records are likely to be spatially biased and that ‘hot spots' of crime tend also to be ‘hot spots' of otherwise un-geocoded data.