shrug-l: Geocoding behavior

Johnson, Jay johnsonj@talgov.com
Thu, 18 Dec 2003 09:41:07 -0500


We've run across two odd ESRI geocoding behaviors.  Assume use of ESRI
default geocoding parameters.

1)  If we geocode an unscrubbed list of addresses from one of our legacy
databases against parcels (polygons) and some of the GIS polygon addresses
contain a "0" in the housenum field (such as for undeveloped property), then
the ESRI geocoding engine will give a score of 77 for an intersection style
address.
EXAMPLE:  TENNESSEE ST & MONROE ST will successfully geocode to the first
instance of "0 Tennessee St".  This is not desirable.

2)  If we geocode the same list of addresses against parcels (polygons) and
instead of "0"s the source polygon addresses contain a <NULL> in the
housenum field, then the geocoding engine will give a score of 76 for an
address that has a housenum that doesn't appear in the database.
EXAMPLE:  3700 CAPITAL CIRCLE SE will successfully geocode to the first
instance of "<NULL> Capital Circle SE".  This is not desirable.

It is apparent that for undeveloped properties either a "0" or a <NULL> in
the housenum field will yield erroneous results that on the face of them
have pretty good scores (76 or 77).  By way of contrast, an address with a
small typo, such as "4716 Planters Ridge Dr", only produces a match score of
52 for the actual address of "4715 Planters Ridge Dr".  It seems odd that
the <NULL> gets such a high score by comparison.  We wouldn't want to remove
the non-addressed parcels from the underlying data, as it may be useful in
certain circumstances to query for parcels without an assigned address.  

I'm interested to hear how other people are dealing with these sorts of
addressing issues and how you are treating addresses in your geocoding
source data that contain no valid housenum.  Thanks.

Jay Johnson
GIS Project Manager
City of Tallahassee
(850) 891-8099


Jay Johnson
GIS Project Manager
City of Tallahassee
(850) 891-8099