A multifaceted comparison of ArcGIS and mapmarker for automated geocoding

Abstract

Geocoding is to appropriate geographic units. We evaluated two commonly used geocoding software packages, ArcGIS and MapMarker, for automated geocoding of a large number of residential addresses from health administrative data in New York State, USA to better understand their features, performance and limitations. The comparison was based on three metrics of evaluation: completeness (or match rate), geocode similarity and positional accuracy. Of the 551,798 input addresses, 318,302 (57.7%) were geocoded by MapMarker and 420,813 (76.3%) by the ArcGIS composite address locator. High similarity between the geocodes assigned by the two methods was found, especially in suburban and urban areas. Among addresses with a distance of greater than 100 m between the geocodes assigned by the two packages, the point assigned by ArcGIS was closer to the associated parcel centroid ("true" location) compared with that assigned by MapMarker. In addition, the composite address locator in ArcGIS allows users to fully utilise available reference data, which consequently results in better geocoding results. However, the positional differences found were minimal, and a large majority of addresses were placed on the same locations by both geocoding packages. Using both methods and combining the results can maximise match rates and save the time needed for manual geocoding.