Address Type Rank - What is the Precision Order?
I've geocoded around 25K addresses and I'd like to sort them into 2 buckets, high vs low precision. To me a high score is meaningless if it can only find it to the nearest postal centroid for instance.
Currently I have ordered the address level results into the following numerical buckets. My goal is to quickly highlight & bucket addresses ESRI isn't as confident in (ex: IF Address-level > 6 then low precision match). As these represent customer addresses I felt (esp for L1) if ESRI can't get it to a street level I doubt we could even send anything to this address. So essentially this is helping us see bad addresses quickly.
Is this the right order they should be in? I wasn't sure if locality should be above postal, and I feel like POI could be higher. What is the right stank rank order these should be in, I couldn't find a definitive listing.
I've also viewed the following topics on this as well:
- https://community.esri.com/message/701456-re-geocoding-accuracy-how-to-determine?commentID=701456#co...
- Geocoding: Delivering High Location Accuracy | ArcGIS Blog
- Service output—ArcGIS REST API: World Geocoding Service | ArcGIS for Developers
- Category filtering—ArcGIS REST API: World Geocoding Service | ArcGIS for Developers (liked this one but I want a more rigid ordering system, ex: a stack rank)
Also it was confusing to figure out where to put this question, is this the right place for it?
For the most part I think you are right. It really depends on the level of accuracy you want, if you want to the actual location with in a matter of 100 feet or so only use 1-3. I'm not familiar with Subaddress, unless it is the same as using the full address and the ZIP+4, in which case that will depend on your geocoder's source data. If derived from the USPS data then it should be great, if it is not, they could be really bad. I am assuming all of your geocoders though are from ESRI directly, in which case the quality of that one would depend on when ESRI last acquired Postal data.
For the geocoders 4-10, they are not really going to get you any closer than with in a few miles of the location unless by sheer random chance your location lies near the center of the centroid. If you want to geocode with better quality try setting your minimum match score to 90 and the minimum candidate score to 70. Then only take those matches with the higher score. Again though those locators less than 4 in your rank are going to be just vicinity locations. You would think that StreetInt should be good, but not all streets are coded correctly for their actual numbering scheems, so some are ok, but vastly more are off by a few blocks.
