How does YouGov project elections?

During many elections, YouGov uses a statistical method called multilevel regression with post-stratification (MRP) to estimate how people are likely to vote.

This approach combines survey responses from YouGov panel members with population data to produce estimates of voting behaviour across the country.

How the model works

Step 1: interviews

Election models begin are based on interviews with YouGov members about whether they plan to vote and which candidate they support.

These studies often involve very large numbers of interviews to help us understand how voting intentions vary across different demographic groups and locations. For example, our recent 2024 U.S. election model used responses from tens of thousands of registered voters, with earlier stages including nearly 100,000 interviews.

Step 2: Model training

We use these responses to train a statistical model that identifies patterns in how different groups of people are likely to vote. The model then applies those patterns to the wider electorate.

Step 3: Post-stratification

In simple terms, the model:

1. estimates how likely people are to vote
2. estimates which candidate they are likely to support if they do vote
3. adjusts these estimates so they reflect how common different groups of voters are in the real population.

(For example, if the model estimates how younger voters and older voters are likely to vote, we then scale those estimates based on how many younger and older voters there are in the population.)

This final step — called post-stratification — helps produce overall estimates of support for each candidate across all registered voters, not just those who were interviewed.

Why our model is successful

Even large surveys cannot include enough respondents from every location or demographic group. Statistical modelling helps fill these gaps by combining survey responses with reliable population data.

This makes it possible to:

• estimate results at the state and district level
• better represent smaller or harder-to-reach groups
• track how voting intentions change over time

Election projections based on modelling reflect the current state of public opinion, not a guaranteed prediction of the final result. Because opinions can change before election day, estimates are updated as new survey data becomes available.

Our strong track record

YouGov has successfully used MRP modelling in several major U.S. elections — including 2016, 2018, 2020, and 2024 — as well as in elections in other countries.

Thank you for being a YouGov member.