Geocoding accuracy
Index > research > geocoding accuracy

There are a number of questions that need to be addressed with geocoding. One relates to the accuracy of a geocoded point.

The geocoding process involves placing a dot on a map to represent an address or other location. In the crime mapping field, this usually means a crime site or the location of a call for service. Geocoding is a process undertaken, usually automatically, by a GIS.

The first question is, how accurate is this dot, automatically placed on the map, in relation to the actual house it is often trying to represent? I studied this using cadastre land parcels in the Eastern Suburbs of Sydney. Over 20,000 addresses were geocoded and compared to both their address plots as represented by cadastral records, and also to see if the plotted point was in the same census tract as the cadastral centroid.

Using the default 10 metre offset from the road that is hardcoded into MapInfo in versions 5.5 and earlier, the following results were obtained:

 
10 metre offset
Points in correct cadastre polygon
10%
Points in wrong cadastre polygon
26%
Points not in any cadastre polygon
64%
Points in different census tract from cadastral centroid
7.5%
   
5% trimmed mean distance from geocoded point to cadastre centroid
31 metres (100 feet)

Newer versions of MapInfo and ArcView allow for more control of the geocoding operation. Through extensive testing of a variety of different options, the following settings were found to be most effective for the Eastern Suburbs of Sydney. Use of a 15 metre inset and a 25 metre offset reduced much of the error, as can be seen in the following table. Caveat: These results worked well for the Eastern Suburbs of Sydney, a compact urban area. It is likely that these settings will be better than a 10 metre default setting in most urban environments, but rural crime mappers should consider using different settings.

 
10 metre offset
25 metre offset, 15 metre inset
Points in correct cadastre polygon
10%
47%
Points in wrong cadastre polygon
26%
35%
Points not in any cadastre polygon
64%
18%
Points in different census tract from cadastral centroid
7.5%
5.0%
   
5% trimmed mean distance from geocoded point to cadastre centroid
31 metres (100 feet)
20.5 metres (67 feet)

For a reference and paper that discusses the accuracy of geocoding as above, see:

Ratcliffe, J. H. 2001 'On the accuracy of TIGER type geocoded address data in relation to cadastral and census areal units'
International Journal of Geographical Information Science 15(5): 473-485. Download the paper here.