Indiana's voter-purging software removes voters without notice, is wrong 99% of the time

The Interstate Voter Registration Crosscheck Program identifies possible duplicate voters by looking at registrations by people with the same name and birthdate; a joint study by researchers at Harvard, Yale, and Microsoft found that 99% of the people it identifies as duplicate voters are not duplicate voters — that is, it has a 99% false positive rate.

States that use Crosscheck have historically interposed some kind of due process: if the system accused you of being a duplicate voter, you'd get a chance to rebut its accusation. But Indiana's Republican state government has removed those checks, so voters who are fingered by this wildly inaccurate system immediately lose their vote, with no warning or chance to present evidence to the contrary.

Common Cause and the ACLU are suing the state.

The program is backed by Kansas Secretary of State Kris Kobach, who is also running Trump's voter fraud investigation.

This week's This American Life has an excellent segment (transcript) on the methodological problems with identifying duplicate voters by looking at name/birthdate matches, building on the painstaking work of Sharad Goel and colleagues.

Goel's paper makes several excellent points, including:

* Many states use January 1 as the default birthday when a voter's birthday is not known, so lots of people will "share" that birthday

* People with names like "June" tend to be born in June (and "Autumn"s are born in the fall, and Carols, Christines and Jesus's are the likeliest Christmas birthday names), so there is a higher probability that two Junes will share a birthday than random chance suggests

* The birthday paradox is a well-understood, but counterintuitive phenomenon by which the probability of two random people out of a pool of 23 will share a birthday is 50%, and the probability rises to 99.9% with just 70 people: so if 70 Juan Gonzaleses are born in 1971, there's a 99.9% chance that two of them will share a birthday

* Finally, there's a low, but measurable, rate of error in voter rolls. Those errors account for virtually every case of alleged duplicate voting that isn't covered by the other three factors.

Lawsuit: Indiana Purging Voters Using Software That's 99% Inaccurate
[Kelly Weill/The Daily Beast]

(Image: Demos)

(via /.)