Thread by @LeaveCrosscheck

Thread about Kansas "Strength of Match" Criteria for Crosscheck /1

Kansas pays for and administers #Crosscheck, a voter registration list maintenance tool, on behalf of all 28 member states. Member states/2

send voter registration records to KS every year; KS sends them back file of "possible duplicate registrants" for processing. /3

We've tweeted often that #Crosscheck ignores missing and mismatches SSN4 thus placing millions of "false positives" on these files. /4

In our research, we learned that Kansas calculates a "strength of match" score for each "possible duplicate registrant" from #crosscheck /5

sent to KS County Election Officers. (This calculation is not provided to other member states.). Here is the strength of match grid:

And here is the first page of the Instructions for County Election Officers

From the highlighted portion of the instructions, we see that KS excludes some Crosscheck records before sending them to county officials /9

Specifically, records with mismatched MIDDLE NAMES are not provided to KS county officials but records w/mismatched SSN4 are. /10

Here is a table summarizing the strength of match criteria used by Kansas for internal processing.

Analysis of the scores Kansas assigns shows that SSN4, middle name, and suffix (such as Jr/Sr) are assigned relatively equal weights, /12

with a middle name mismatch *more* disqualifying than an SSN4 mismatch. This is consistent w/KS decision to exclude mismatched middle /13

name from the file sent to their own county officials. The maximum score assigned to a record w/mismatched middle name is 20; /14

the maximum score assigned to a record with a mismatched SSN4 is 50. (Scores range from 10-95). /15

Records with missing SSN4 have scores that range from 70-85, and records with mismatched SSN4 range from "no score" to 50. /16

At this point, it may be tempting to start looking at examples to see what passes the "common sense" test. Don't. /17

There is an entire branch of science dedicated to data matching such as this. Looking at KS' strength of match scores used inside KS, /18

I suspect that KS has not consulted w/such experts. More weight on middle name than SSN4 raises this suspicion (among other things) /19

This is a KS-specific calculation so what's the significance for #crosscheck as a whole? /20

First, it reveals that KS has detailed data on the match quality of "possible duplicate registrant" pairs. /21

Second, it shows that KS feels that their county officials need more information about the match quality to make good decisions. /22

Third, it shows that even *KS* knows that some of the data on the #crosscheck file isn't worthy of review by human eyes. /23

Fourth, it implies that KS has not consulted w/experts in data matching which may explain why six states have left #crosscheck /24

This last point is the most critical. Kansas is entrusted with 100 million voter registration records every year, /25

& the way they handle their own #crosscheck file raises questions about whether they're using good science in handling the database. /26

#ksleg & #crosscheck member states should ask about whether #crosscheck matching algorithms they are in keeping w/ current data analytics/x