Turnout modeling

Turnout modeling predicts the probability that a specific registered voter will cast a ballot in a given election. It is built primarily from the voter file's recorded history of which past elections a person voted in, along with demographic data, election type, and sometimes modeled or survey-derived variables, to produce a score for each voter rather than an aggregate turnout forecast for the whole electorate.

This differs from a pollster's likely voter model, which serves a different purpose: pollsters use likely-voter screens to decide which survey respondents to include or weight more heavily when estimating an election outcome from a sample, while campaign turnout models score the entire voter file to decide who to contact and how. Both rely on similar signals — past voting history is the strongest predictor in each case — but one drives a sample selection decision and the other drives a targeting decision.

Campaigns use turnout scores to prioritize Get-out-the-vote (GOTV) analytics spending on voters likely to respond to mobilization, and to combine with persuasion modeling and microtargeting scores when planning outreach. A common pitfall is treating turnout score as fixed: it reflects a baseline propensity that field programs are specifically designed to move, so a well-run mobilization effort should shift the very behavior the model is predicting, and models require regular retraining as voter rolls, contact history, and electoral context change.