Geographic disparity of positional errors and matching rate of residential addresses among geocoding solutions. Issue 1 (2nd January 2016)

• Record Type: Journal Article
• Authors: Chow, T. Edwin; Dede-Bamfo, Nathaniel; Dahal, Khila R.
• Is Part Of: Annals of GIS. Volume 22:Issue 1(2016)
• Page Start: 29
• Page End: 42
• Publication Date: 2016-01-02
• Subjects: Desktop and Internet geocoding -- address-matching -- urban/rural disparity -- quality assessment -- crowdsourcing; Geographic information systems -- Periodicals; Periodicals; 910.285
• DOI: 10.1080/19475683.2015.1085437
• Languages: English
• ISSNs: 1947-5683

Abstract

Abstract: The purpose of this study was to systematically examine the geographic disparity of geocoding error among different geocoding solutions. The research questions include: (1) What are the positional accuracies and matching rates of various geocoding techniques? (2) Are there any significant differences of geocoding quality in rural and urban areas? In this study, 1100 residential addresses scattered across Texas, USA, were address-matched using eight different geocoding platforms, including the ESRI ArcGIS Address Locator, CoreLogic PxPoint, Google Maps API, Yahoo! PlaceFinder, Microsoft Bing, Geocoder.us, Texas A&M University Geocoder, and OpenStreetMap (OSM). The geocoded locations for each method were validated against the GPS data and manual digitization. Using GPS data as reference data, the desktop geocoding using parcel data achieved the highest positional accuracy with a mean error of 24.8 m, whereas the Google Maps API was the best among the six Internet solutions with a mean error of 31.7 m. All geocoding solutions, except Geocoder.us and OSM, achieved a matching rate >95%. It is important to note, however, that the OSM geocoding revealed decent positional accuracy in terms of median and 5% trimmed mean errors, indicating the potential of crowdsourcing approach to produce an accurate geospatial data set. The geocoding errors between urban and rural areas were significantly different in most geocoding solutions but there was no consistent and monotonic trend. Excluding the errors of outliers, defined as the lowest 5 percentile, indeed changed the direction of urban versus rural accuracy in OSM, PxPoint, and ArcGIS geocoding.