This is the first release of our model estimating 2024 presidential election votes in every state, based upon nearly 100,000 recent interviews of registered voters. We show Kamala Harris leading Donald Trump by 50% to 47% just before their first debate. However, the race will be determined by who wins the most electoral votes, not popular votes, and, as it currently stands, the race is a toss-up.
We have Harris leading in 22 states and Washington D.C. with 256 electoral votes and Trump leading in 25 states with 235 electoral vote.
270 votes are needed to be elected and, according to our model, the outcome will be determined by races that we currently rate as toss-ups: Arizona, Georgia, Pennsylvania, and Nebraska’s 2nd District. All of these contests are within the margin of error and could go either way.
The model is based on 95,643 interviews from YouGov’s U.S. panel, conducted between August 23 and September 5, 2024 as part of the SAY24 project, a collaboration between Stanford, Arizona State, and Yale Universities. We have been interviewing most of these panelists on a regular basis since December 2023 with periodic re-interviews — either monthly or quarterly, depending upon the panelist. This is a much larger sample than is normally used for opinion polling and allows us to make estimates for each state. Panelists have been asked about their likelihood of voting and who they intend to vote for, along with a host of other questions.
A unique feature of this study is that we can track the same people over time and see how they shift — or don’t shift — as the campaign progresses. Unlike most polls, which draw a new sample each time a survey is conducted, we can distinguish voters switching between candidates and not voting and, on the other hand, variations due to changes in sample composition. So far, in 2024, we have seen striking stability in voters’ candidate preferences. Even after the first presidential debate on June 27 — between President Joe Biden and Trump — or the attempted assassination of Trump on July 13, there was little movement in vote intention.
Another important feature of this dataset is that we have matched the participants to TargetSmart’s national voter file. This means that, except for people who tell us they intend to register between now and the election, we have a sample of verified registered voters. We have also linked voters to tabulated vote in the precinct where they are registered. This helps ensure that the sample is representative of different geographies, some of which have been underrepresented in prior years.
Even with a sample of this size, we are still short of data in some key areas and hard-to-reach demographics. In Maine and Nebraska, for example, electoral votes are allocated by congressional district, and we have to estimate votes for each. In small states, we have correspondingly small samples. And in large states, some groups — such as younger voters and rural voters — can be in short supply.
Our approach to estimating the electoral college vote is based upon a multilevel regression with post-stratification (MRP) model. We have used this approach successfully in past elections in the U.S. and elsewhere. It uses a statistical model to predict votes for everyone on the national voter file, whether or not they belong to YouGov’s panel. The 95,643 interviews for our panelists are used to train a model that classifies people as likely Harris, Trump, or third-party voters — or non-voters — and then this model is applied to the entire voter file. We then aggregate these predictions — in what is referred to as post-stratification — to estimate votes for all registered voters.
The model has three stages: (i) estimate the likelihood of voting; (ii) conditional upon voting, what is the probability of voting for a major-party or third-party candidate; and finally (iii) predict support for Harris and Trump among major party voters. We also used MRP models to estimate votes in the 2020, 2018, and 2016 elections.
We will be updating our electoral-vote model in the beginning of October, based upon approximately 30,000 interviews that we are conducting this month. We will then reinterview the full panel in late October and release a final forecast prior to the election.
In addition, we have created models for every Senate and House race in the U.S., with the first release of estimates expected next week and periodic updates as the campaign progresses.
We caution that these models are based upon what people tell us they plan to do. A small share of registered voters (6%) say they are undecided, but the majority tell us their minds are made up. However, people can change their minds and, if they do, we should see these changes reflected in our model updates. These results reflect our best estimate of the current state of the race.
Image: Getty
