Testing the Spatial Accuracy of Address Based Geocoding for Gun Shot Locations

We assess the positional accuracy of address based geocoding of shooting incidents relative to the location recorded via acoustic gun-shot detection technology. This provides a test of the accuracy of typical address based geocoding methods used in crime analysis, as well as provides evidence for how much accuracy one gains when using sensors. Examining over 1,000 shooting incidents in Wilmington, North Carolina, we find that the majority of address-based incidents are quite accurate, on average within 60 feet of the actual location (using a street centerline geocoder), or within 90 feet (using Google rooftop geocoding). However, based on the incident narrative we identify a subset of transcription errors in over 10% of the cases that increases the distance between the true shooting location and that geocoded using address data. This suggests mechanisms to prevent human errors may be more frugal than those relying on sensors in geocoding shooting incidents. Data to replicate the analysis can be downloaded from https://www.dropbox.com/sh/bceyldwgj84ztlw/AABdQBnjKGdO3GUxWM0ZMd3Ya?dl=0.