The Washington Post and George Mason University’s Schar School of Policy and Government have published a final poll measuring Virginia voters’ support for candidates in the last weeks of the gubernatorial campaign. The poll doesn’t forecast an outcome of the contest between Democratic former governor Terry McAuliffe and Republican ex-private-equity executive Glenn Youngkin. Instead, it examines what voters think about key issues in the campaign and how those attitudes are shaping voting decisions — including a glimpse of how different turnout scenarios may affect the outcome of the race.
This year’s surveys incorporated methodological changes aimed at addressing challenges in achieving a representative sample of Virginians — challenges that were made clear in the 2020 presidential election, when state-level polls across the industry had the highest error in at least 20 years, according to a report by the American Association for Public Opinion Research. The Post’s polls that year had a mixed record, accurately estimating support for Joe Biden and President Donald Trump in some states, including Virginia, but overestimating Biden’s support elsewhere.
One specific challenge is called differential nonresponse: While a poll like The Post’s contacts a random sample of the population, different categories of people may respond at different rates. For instance, college graduates and people who volunteer in their community have long been more likely to answer surveys than those in other groups. Polls have historically addressed this issue by calibrating samples to match census benchmarks for the population by gender, education, age, race, ethnicity, sex and region. Yet errors across many 2020 pre-election polls suggest that more needs to be done.
So this year, Post-Schar School polls adopted a survey method that measures and adjusts for the rate at which partisan groups participate in the poll. The samples for the polls were primarily drawn through a database of registered Virginia voters maintained by the nonpartisan data firm Aristotle as well through a random sample of all Virginia cellphones and landlines, matching them to that database where possible.
Virginia does not register voters by party, but the Aristotle database includes a prediction of every voter’s party identity based on his or her individual turnout in partisan primaries, political donations made, and other indicators. Aristotle’s predictions are fairly accurate, with roughly three-quarters of voters classified as Republican saying they identify or lean toward the Republican Party in the September Post-Schar School poll, with a similar level of accuracy for Democrats.
When the polls were completed, the partisan makeup of the sample was compared with the database of all voters and then weighted so that the proportion of Democratic, Republican and unknown party voters matched the Virginia electorate as a whole. That was in addition to weighting adjustments for other demographics. The Post and Schar School used a similar method in the 2017 Virginia governor’s election. More technical details about the survey’s methods are available here.
These changes may improve the accuracy of these polls in the long run, but it will take time to assess their effectiveness. Polls are not precise instruments — even a perfect random sample of 1,000 voters has a margin of error of about three percentage points, too large to determine a leader in a close election. And likely voters have never been easy to identify, especially with turnout swinging sharply over the past decade.
More on the 2021 Virginia election
Latest news: Virginia politics
House of Delegates: Four Democratic incumbents, one Republican ousted in primaries