AP VoteCast performance

Explaining the why of Election Day — at the moment polls close

In the weeks after Election Day, the U.S. Census Bureau conducts a nationwide study on voters and nonvoters, examining demographics and voter trends. This addition to the Census Bureau’s Current Population Survey is widely cited benchmark data and is viewed as an authoritative assessment of the electorate.

It takes weeks for the Census Bureau to conduct this research on households that are part of the Current Population Survey. When it arrived in April 2019 following the 2018 midterm elections, it showed that AP VoteCast delivered — in real time on Election Day — an accurate picture of the composition of the electorate by age, gender, race/ethnicity and education.

Delivering accuracy in Election Day outcomes

Journalists who depend on AP VoteCast to help set their newsroom agenda do so with data they can count on. In the 2018 midterm elections, they made coverage decisions — hours before polls closed and officials started to count ballots — based on data that was remarkably accurate in reflecting the choices of the American electorate.

• AP VoteCast’s vote choice projections at 5 p.m. were correct in 65 of 71 races for governor and U.S. Senate, with two races viewed as too close to call. Among its four incorrect projections were two races in Florida, both among the closest in the country and both of which required a recount to resolve.
• As polls closed, AP VoteCast said Democrats would beat the GOP by 8.4 percentage points in the total number of votes cast for the U.S. House of Representatives. The actual result was 8.6 percentage points.
• Overall, AP VoteCast’s average vote choice error at 5 p.m. on Election Day was 1.2 percentage points; the median error was 1.5 percentage points. In recent elections, the error at 5 p.m. in favor of Democratic candidates as reported by the in-person exit poll was often measured in double digits.

After action research confirms accuracy of AP VoteCast methodology

In the weeks after Election Day, research and data scientists from NORC at the University of Chicago conduct an extensive review of AP VoteCast’s survey methodology to ensure it delivers on its promise to provide accurate and informative data in real time.

Following the 2018 midterm elections, one aspect of that AP VoteCast review was a voter validation study. The research sought to confirm that respondents to the random sample portion of the survey who reported they cast a ballot in the election actually did so, through an examination of state voter files, which are public records.

The results of that study found that AP VoteCast’s likely voter model correctly classified more than 9 in 10 respondents as voters or nonvoters. Moreover, 94 percent of those classified as voters did vote, according to state voter files.

Read NORC’s full review of AP VoteCast’s performance in the 2018 midterms.