Georgian law on "exact match" could deprive 909,540 voters of voting rights, my research reveals

Georgian Secretary of State Jack Kemp, Republican candidate for governorship against Democrat Stacey Abrams, has suspended more than 53,000 voters so far, due to the disparity of names in their voting records and other sources such as driving licenses and social networks. Security cards. If the measure comes into force, voters whose information does not exactly match the different sources will have to bring a valid photo ID to the polls on polling day to be able to vote. This could reduce voter turnout, either because some voters do not have an ID card, or because voters do not know if they are eligible. Proponents of the rule say that it's only about prevent illegal voting.

[Why gerrymandering is going to get even worse]

But does it lack a hyphen, an initial instead of a full middle name, or is the mere fact of not distinguishing a letter from the name of an elector is there any evidence sufficient that the voter is not what he claims to be? How would we know?

Researchers often need to match records – and do things well

In this case, researchers often ask this question. When performing empirical scientific research, they often have to link various sets of data with the help of an imperfect identifier, such as agency names or individual addresses. Although it can be tedious, it is crucial that the matches are correct. Match the bad records, and any analysis can be totally unreliable. This leads many data analysts to keep only exact matches.

However, although incorrect matches may cause problems, it is also possible to delete records that need to be matched, but have small differences. The elimination of these records can also corrupt an analysis.

That's why I spent the last three years helping to develop an algorithm which uses a probabilistic record linkage called "fastLink" which not only speeds up and automates record linkage, but also tells the analyst the probability that an incorrect match of two records is actually correct.

In one recent study Co-written with colleagues Ben Fifield and Kosuke Imai, we apply the algorithm to the issue of voter identification. The results raise serious concerns about Georgia's law on exact correspondence – and its likelihood of preventing tens of thousands of valid voters from voting.

This is how we did our research

We worked on linking two 2014 and 2015 national voter files collected by L2 Inc, a non-partisan national society that provides voter data and related technology for campaigns. All active voters in 2014 appeared in the 2015 dataset, which means that we knew that a real match still existed. But many records had typographical inconsistencies that prevented exact matches.

Our analysis revealed that the "exact match" approach would only connect 66% of the voters who were actually the same, correctly identifying about 91 million voters. In other words, the "exact match" would exclude nearly 40 million records that actually referred to the same voter – without voting rights for many Americans.

What does this mean for Georgian voters?

The Georgian archives had a higher proportion of accurate matches than the one we found at the national level – but 30% of voters still do not match exactly in that state.

On the other hand, using our algorithm, which corresponds almost perfectly to L2's internal correspondence records (r = .99), we are able to match nearly 127 million registered voters – or 93% of all voters data from 2014. Of those whose records did not match exactly, we found that 25% had a probability of at least 99% to be correct, while 28% had a probability of at least 99% being correct. less 95%.

Using our algorithm, in other words, 91% of Georgian voters would be allowed to vote, ie 3,941,342 citizens entitled to vote – while the "exact match" would only erase 70%, excluding 909,540 eligible citizens. the right to vote.

I've also tried to connect the electors of the ANES (American National Election Study) 2016 to the voters' registers contained in the level 2 data using two methods: the exact match and a improved version of fastLink I recently developed.

The results appear in the table below. As you can see, the "exact match" method misses a substantial part of valid matches. Although our algorithm validated 60% of voter registrations, the "exact match" validated less than 30% on average.

And, in accordance with the concerns of opponents of the Georgian measure, non-white voters are particularly likely to be deprived of the right to vote. Correspondence rates using exact matching are respectively nine and six percentage points lower for black and Hispanic voters compared to white voters.

The Georgian Law on Accuracy is the latest in a series of voter identification measures that critics say is a thinly veiled crackdown on elections. Whether intentional or not, the Georgian rule of "exact correspondence" will disproportionately prevent minority voters from voting.