Likely voter model

Because not everyone a pollster surveys will actually vote, pollsters build a likely voter model to identify or weight respondents by their probability of voting, so the reported results reflect likely voters rather than all adults or all registered voters.

Models typically combine self-reported intent to vote, past turnout history from voter files, a form of administrative data, enthusiasm questions, and demographic predictors, either to screen out unlikely voters with a hard cutoff or to weight each respondent by an estimated turnout probability in a softer model. This differs from ordinary survey weighting, which corrects for who was sampled relative to the full population's demographics, not for who among them will actually show up to vote; the two adjustments are usually applied together, on top of a response rate that is already imperfect.

The choice of likely voter model is one of the largest sources of disagreement between pollsters, especially close to an election, and a major reason different polls of the same race can diverge even when their underlying samples are similar. Pitfalls include models built on past turnout patterns misfiring when an election has unusual dynamics, such as a newly mobilized demographic or unexpectedly depressed turnout, and self-reported vote intent tending to overstate how many people will actually participate.