See how The Arizona Republic's voter turnout projection compared to reality

Earlier this year, The Arizona Republic set out to project how many state residents would vote in the November election.

The projection formed the basis for a tool to help readers know how many votes were left to count in the presidential election ― and it proved on-target.

The Republic's projection was that 3.37 million Arizonans (or 3,377,525 to be exact) would come out to vote this year, about 77% of the eligible electorate. The Arizona Secretary of State's Office reported this week that 3,389,319 people voted in the presidential contest.

The difference between those numbers is about 0.35%.

Not everyone who cast a ballot actually voted for one of the presidential candidates; among the total ballots cast, The Republic's projection was about 1.5% lower than the actual number.

"That’s impressive!” Paul Bentz, senior vice president for research at the Arizona political consulting firm HighGround, said. Before the election, Bentz told The Republic he expected that between 76% and 78% of eligible voters would turn out.

More than 3,420,000 Arizona voters ― or about 78.5% of the state’s eligible electorate ― cast a ballot in this year’s general election, according to data from the Secretary of State’s office.

Most who participated did cast a vote in the presidential race: fewer than 40,000 Arizonans didn’t pick a presidential candidate, figures from the Secretary of State’s office show.

With more people mailing their ballots in late — combined with Arizona’s batch of Election Day drop-offs — coming up with an accurate projection was "a little harder" this year than in the past, Bentz said.

“Having done this for 20 years … we often say that there isn’t an Election Day in Arizona, there’s an election month,“ Bentz said. “A lot of people made this almost more like an Election Day than what we're used to.”

How The Republic estimated voter turnout

The Republic’s data team modeled and analyzed historical election data to come up with a turnout estimate, which was rounded down to the nearest 10,000th vote.

Reporters used a combination of several approaches — from the simple to the complex — to generate predictions, then averaged across them to create a rounded election turnout projection.

Members of the data team worked with Christopher Weber, director of The Arizona Voter Project at the University of Arizona, to model turnout for the presidential election. Weber holds a doctorate in political science and has a background in statistics.

With Weber's guidance, a reporter used data collected by L2, a research company that combines voter registration information from the Arizona Secretary of State's Office with demographic information on registered voters across the nation, to perform several regression analyses.

The statistical methods allowed for a better understanding of the relationship between voting in prior presidential general elections and factors like an individual's age, voting history and party affiliation.

“If you’re going to predict whether someone votes or doesn’t, I can’t think of a better measure than did they vote in the last election,” Weber told The Republic. “You built a regression model that takes input from prior elections to predict the present. And because there is a remarkable degree of stability in whether people vote, particularly now, your indicator is probably as best as you’ve got.”

Reporters used cross-validation, a machine learning industry-standard method, to ensure their models would perform well on new data and generate more accurate predictions.

Reach Sahana Jayaraman at [email protected]. Follow her on X @SahanaJayaraman

This article originally appeared on Arizona Republic: Arizona voter turnout projection from The Republic was just 0.35% off