Exit Polls

Surveying Voters as They Leave the Polls

Exit polls intercept voters as they leave polling places and ask them a short questionnaire about how they voted and why. In the US, the National Election Pool (a consortium of major media organizations) conducts exit polls at hundreds of randomly selected precincts on Election Day, supplemented by phone surveys of early and mail voters.

Exit polls serve two purposes. First, they provide the demographic data that television networks use to 'call' races before all votes are counted. If the exit poll shows a candidate winning 58% of the suburban vote in a state where suburbs determine the outcome, networks can project the winner with confidence. Second, they reveal the 'why' behind election results: which issues motivated voters, how different demographic groups voted, and what drove the outcome.

Limitations and Controversies

Exit polls have several known biases. Young voters are more willing to participate than older voters. Voters at busy urban precincts are harder to sample than those at quiet suburban ones. Early and mail voters, who are surveyed by phone rather than in person, may respond differently. In 2004, exit polls initially suggested John Kerry was winning, leading to premature celebration among Democrats before actual vote counts showed a Bush victory.

In developing democracies, exit polls play a different role: they serve as a check on official results. If the exit poll shows one candidate winning by 10 points but the official count shows them losing, this discrepancy can be evidence of electoral fraud. Exit polls were instrumental in exposing vote-rigging in Ukraine's 2004 Orange Revolution.

Worked Example: Building the Topline From Subgroups

Exit polls do not just report an overall number; they build it from demographic pieces, and understanding that arithmetic shows both their power and their fragility. Consider a simplified statewide electorate split into three groups (figures below are illustrative, chosen to demonstrate the method):

Candidate X's estimated share is the turnout-weighted sum of the subgroup shares: 18.2 + 16.0 + 17.0 = 51.2%. Notice two things. First, the turnout weights matter as much as the vote shares. If non-college whites actually make up 45% of the electorate rather than 40%, X's total drops by roughly a point, because that group leans against X. This is exactly the assumption exit polls can get wrong, and it is the same education-weighting problem that haunted pre-election polls in 2016.

Second, subgroup margins of error are large. A group that is 25% of a 2,000-respondent exit poll contributes only about 500 interviews, giving a margin of error near 4-5 points on that group alone. So a headline like '68% of voters of color backed X' could plausibly be anywhere from the low 60s to low 70s. The topline is more stable than any single crosstab, which is why you should treat demographic breakdowns as informative estimates, not precise measurements.