How accurate is 80% match score compared to 100%?

LindseyWinn1

I've been researching the match score for geocoding, and can't find how ESRI defines accuracy, especially when the default is 80% as opposed to 100%. Does anyone know how ESRI decides this and how reliable 80% is? I use it for research and want to be as accurate as possible, but also have as many matched addresses as possible. Thanks!

JoeBorgione

I've been researching the match score for geocoding, and can't find how ESRI defines accuracy, especially when the default is 80% as opposed to 100%. Does anyone know how ESRI decides this and how reliable 80% is? I use it for research and want to be as accurate as possible, but also have as many matched addresses as possible. Thanks!

I don't know the actual stats, perhaps someone can from ESRI chime in; I've found it often depends on the type of locator you are using.  Certain elements of the address are weighted differently, like prefix & suffix direction, type etc.

Here in Utah we use a grid system where the actual address is a point on a grid.  For example,
4104 S 3305 E
is 41 blocks south and 33 blocks east of the 0,0, point downtown.

Consider the evil twin of this address; 4104 S 3305 W.  In terms of geocoding with a US streets locator the two addresses are almost the same, yet if the evil twin existed, they would in actuality be miles apart.  I can't remember what the percentage of hit is between these two, but it's pretty close;  in a 9-1-1 response situation, it would be up to the lawyers to sort it all out in the end.  For a coupon mailer it's just another one that goes into a black hole.

So, when you ask how 'reliable' is an 80% hit, you have to ask yourself; What's at stake with an 80% hit?  Everyone wants to be as accurate as possible whether you are dispatching 9-1-1 calls, mailing coupons, or in your case doing research.  However, IMHO accuracy in geocoding is mostly a function of the data you are matching against as well as the data you are matching.

Another discussion is accuracy versus precision.  When geocding against streets, you can only hope to be accurate; that is the address falls on the correct block.  Lets say you street has a range of 100-200 on the even side and 101-199 on the odd side.  Logically the address 150 would be half-way along the block on the even side.  That's accuracy.

The reality might be that the house with 150 painted on the curb is at the end of the block on the odd side of the road.  If you need precision results your street data isn't going to provide it.  Maybe parcel data/point matching would be the way to go.

Hope this helps!