The Icosahedron #10 ft. Matt Knee on the Vote-by-Mail Surge & Predicting the Unpredictable

The traditional method for modeling voter turnout is looking at historical voting patterns, supplemented with enthusiasm measures. When election rules or voting dynamics change, however, information on who voted in the past becomes a lot less useful. To ensure quality turnout models for our clients, we have analyzed this year’s primary elections in which voting rules changed; this has allowed us to apply those patterns and lessons to future turnout models.

While adjusting turnout of occasional voters for a more enthusiastic electorate is rather straightforward, the main challenge comes from voters who are voting in primaries for the first time. While some of these voters are just young people getting in the habit of voting, many are not, and are instead general election voters newly voting in primaries or are new to voting altogether.

Between white-hot political enthusiasm, the effects of COVID-19, and some jurisdictions mailing absentee ballot requests (or actual ballots) to every voter – or otherwise increasing convenience voting – there is a lot to sort out.

The sending of absentee ballot requests or ballots to every voter is especially disruptive. We analyzed the vote history available for the Georgia primary election data as part of our efforts to update our turnout modeling process.

Here are some of our findings and solutions.


  1. In 2016, 20,103 mail-in votes were cast in Georgia, where or 3.5% of the total vote. In 2020, 485,894 mail-in votes were cast, or 49% of the total vote.
  2. 69% of primary voters were first-time primary voters. In 2016, it was 17%.
  3. While 55% of new primary voters had voted in general elections, 45% were entirely new.
  4. 32% of new voters in the 2020 primary were under 45. Only 12% of overall voters were. However, there were also plenty of older new voters: 32% were 65 or older.
  5. While Georgia does not have party registration, our party models suggest these new voters were disproportionately Democratic.


  1. We built models of new voters in Georgia to adjust turnout in other areas that specifically mailed absentee requests to their voters. By examining the patterns of what type of voter is likely to begin voting when offered an absentee ballot, we can apply this to other areas that have made this offer.
  2. We built new voter models based on historical and 2020 new voters, so as to adjust new voters independent of rules changes.
  3. By combining historical models, rule-change models, new voter models, enthusiasm models, and political judgement, we can better predict voter turnout in these unprecedented times. So far, we have deployed these methods in Oklahoma and Alabama primary races, and accurately predicted turnout within a percentage point each time.
  4. We will continue to build models, as data becomes available, to find and apply the patterns and effects of voting rule changes.

If you’re¬†interested in what turnout looks like in your district, contact us.

WPA Releases New Poll on 2024 General Election

Comments are closed.