Evaluation of the positional difference between two common geocoding methods

Abstract

Abstract. Geocoding, the process of matching addresses to geographic coordinates, is a necessary first step when using geographical information systems (GIS) technology. However, different geocoding methodologies can result in different geographic coordinates. The objective of this study was to compare the positional (i.e. longitude/latitude) difference between two common geocoding methods, i.e. ArcGIS (Environmental System Research Institute, Redlands, CA, USA) and Batchgeo (freely available online at http://www.batchgeo.com). Address data came from the YMCA-Harvard After School Food and Fitness Project, an obesity prevention intervention involving children aged 5-11 years and their families participating in YMCAadministered, after-school programmes located in four geographically diverse metropolitan areas in the USA. Our analyses include baseline addresses (n = 748) collected from the parents of the children in the after school sites. Addresses were first geocoded to the street level and assigned longitude and latitude coordinates with ArcGIS, version 9.3, then the same addresses were geocoded with Batchgeo. For this analysis, the ArcGIS minimum match score was 80. The resulting geocodes were projected into state plane coordinates, and the difference in longitude and latitude coordinates were calculated in meters between the two methods for all data points in each of the four metropolitan areas. We also quantified the descriptions of the geocoding accuracy provided by Batchgeo with the match scores from ArcGIS. We found a 94% match rate (n = 705), 2% (n = 18) were tied and 3% (n = 25) were unmatched using ArcGIS. Forty-eight addresses (6.4%) were not matched in ArcGIS with a match score ≥80 (therefore only 700 addresses were included in our positional difference analysis). Six hundred thirteen (87.6%) of these addresses had a match score of 100. Batchgeo yielded a 100% match rate for the addresses that ArcGIS geocoded. The median for longitude and latitude coordinates for all the data was just over 25 m. Overall, the range for longitude was 0.04-12,911.8 m, and the range for latitude was 0.02-37,766.6 m. Comparisons show minimal differences in the median and minimum values, while there were slightly larger differences in the maximum values. The majority (>75%) of the geographic differences were within 50 m of each other; mostly <25 m from each other (about 49%). Only about 4% overall were ≥400 m apart. We also found geographic differences in the proportion of addresses that fell within certain meter ranges. The match-score range associated with the Batchgeo accuracy level "approximate" (least accurate) was 84-100 (mean = 92), while the "rooftop" Batchgeo accuracy level (most accurate) delivered a mean of 98.9 but the range was the same. Although future research should compare the positional difference of Batchgeo to criterion measures of longitude/latitude (e.g. with global positioning system measurement), this study suggests that Batchgeo is a good, free-of-charge option to geocode addresses.

Comparación de las aplicaciones de Google y Yahoo para la geocodificación de direcciones postales con fines epidemiológicos

“…Estos porcentajes se aproximan a los obtenidos en un estudio 14 en el que un software GIS geocodificó el 97% de las direcciones. En otro trabajo este porcentaje con el programaArcGis fue de 94% 13 .…”

Using Routinely Collected Administrative Data in Public Health Research: Geocoding Alcohol Outlet Data

“…There have been a number of studies that have assessed geocoding techniques, positional accuracy (Zandbergen 2008;Ward et al 2005;Duncan et al 2011) and the impacts of geocoded or matched outlet proportion and positional accuracy upon resulting analyses (Krieger et al 2001;Ratcliffe 2004;Zandbergen 2007). In particular, Zandbergen (2007) has shown that wherever possible, high resolution address data (building-level) should be employed as the reference dataset particularly when finescale analysis is being performed.…”

Response to Geocoding-protected health information using online services may compromise patient privacy - Comments on “Evaluation of the positional difference between two common geocoding methods” by Duncan et al.

“…We thank Mak for his positive comments about our article previously published in Geospatial Health (Duncan et al, 2011), and for bringing attention to the important issue of potentially compromising patient privacy (including protected health information) when geocoding such data using online services. We acknowledge that we did not discuss the issue of confidentiality when geocoding data using online services, and we appreciate the opportunity to briefly reflect on this topic.…”