Likely-Voter Screen

A likely-voter screen is the set of questions a pollster uses to filter the survey sample down to people who will actually turn out. Asking “who do you support?” of every adult produces one number; asking it of the subset who will vote produces a different — usually more reliable — number.

Common screening questions include past-vote history, self-reported intent (“how likely are you to vote, on a scale of 1 to 10?”), interest in the campaign, and registration status. Pollsters combine these into a model that probabilistically labels each respondent as a likely voter, a possible voter, or unlikely to vote. The sample is then weighted accordingly.

Likely-voter screens systematically favor the party whose voters score higher on the screen — historically older, whiter, higher-income voters, all of whom turn out at higher rates. This means LV results often diverge from registered-voter results, sometimes by several points. The polling-average reading depends on which subset the underlying polls report.

The 2016 and 2020 polling errors were partly screen-based: pollsters’ screens missed late-deciding non-college voters whose past-vote history did not predict their current intent. Most pollsters have since modified their screens; whether the modifications are sufficient is one of the most contested questions in polling.