Match if candidates tie, but only if they tie at 100.

JamesColbert · ‎06-09-2014

When geocoding, I don't want my tied candidates to match, unless the candidates are tied at a score of 100. Then I always just pick one or the other of the addresses. Is there a way to do this during automatic geocoding?

I know the answer is likely no, but I couldn't help but ask.

Thanks.

KimOllivier · ‎06-10-2014

Run with ties unmatched first. Then set the score to match at 100 and open again with the Address Inspector. Select tied matches. Then you can go through the tied matches one by one and choose. One by one?? You can't have many! Just calculate how many per hour you can do. I assume you are batch matching. If not, perhaps try building a composite locator that combines combinations of tied addresses, non-tied addresses, scores and so on.

It all seems a bit tedious to me for large numbers of addresses. Maybe you can edit the match field to help the locator decide between the pairs? I use various scrubbing scripts using Python tools to improve the source addresses. Use regular expressions, lookup tables and other information to complete the addresses before you run or rerun the locator. Maybe even select all status = 'U', turn off the shape field and export the match table to run a separate scrub and match. It might be useful to standardise addresses first. This splits them into components that are easier to scrub.

For batch processing if you can't see why the matches are tied, turn on the option to return the scores in a field. Have a look at them to see what component could be improved. For example you might be able to fill in missing zip codes if you know the city, or split the data into geographic regions and turn on 'match in visible extent'.

View solution in original post

KimOllivier · ‎06-10-2014

Run with ties unmatched first. Then set the score to match at 100 and open again with the Address Inspector. Select tied matches. Then you can go through the tied matches one by one and choose. One by one?? You can't have many! Just calculate how many per hour you can do. I assume you are batch matching. If not, perhaps try building a composite locator that combines combinations of tied addresses, non-tied addresses, scores and so on.

It all seems a bit tedious to me for large numbers of addresses. Maybe you can edit the match field to help the locator decide between the pairs? I use various scrubbing scripts using Python tools to improve the source addresses. Use regular expressions, lookup tables and other information to complete the addresses before you run or rerun the locator. Maybe even select all status = 'U', turn off the shape field and export the match table to run a separate scrub and match. It might be useful to standardise addresses first. This splits them into components that are easier to scrub.

For batch processing if you can't see why the matches are tied, turn on the option to return the scores in a field. Have a look at them to see what component could be improved. For example you might be able to fill in missing zip codes if you know the city, or split the data into geographic regions and turn on 'match in visible extent'.