The vote count showed, without a doubt, that VoteCast is the new standard in election research. At 5 p.m. ET, VoteCast’s estimates of vote choice correctly projected the winner in 92% of the 35 races for U.S. Senate and 36 races for governor. The average error in those races was only 1.2 percentage points.
But VoteCast is about more than winners and losers. In the months since Election Day, our partners at NORC at the University of Chicago have looked in depth at the data created from VoteCast’s nearly 139,000 interviews with voters and non-voters. What they found proves in several ways that our approach to election research delivers accurate, trustworthy data about the American electorate.
Here’s one example. After every election, the U.S. Census Bureau conducts a comprehensive review of the electorate as part of its Current Population Survey. This data is widely accepted at a benchmark reference of who voted.
Overall, AP VoteCast’s portrait of the national electorate largely matched that found by the Census Bureau, falling within 1–2 percentage points for all age, gender, racial and ethnic groups, and education levels.
That means that on election night in 2018, journalists using AP VoteCast had highly accurate data that could help them – in real time – explain the why of Election Day.
You can read more about AP VoteCast’s performance in the 2018 midterm elections in this report from NORC, including a review of our data collection efforts and how we confirmed –- using state voter files -– our survey successfully reached and classified people who did, and did not, cast a ballot.