On November 5, American voters are called to the polls to elect the 47th president of the United States. They will have to decide between the Republican Donald Trump and the Democrat Kamala Harris, in a duel that some observers announce as one of the closest in the history of the country. To follow this standoff, franceinfo publishes average voting intentions calculated from polls compiled by the political analysis site 538. Every day, we update our graphs in order to follow the latest trends at the national level as well as than at the level of American states.
National polls are useful for tracking major campaign dynamics. This is the case, for example, with the withdrawal of Joe Biden on July 21. The current president withdrew in favor of his vice-president Kamala Harris. This gave rise to a clear increase in voting intentions in favor of the Democratic candidacy. In a few days, the curves turned around and Kamala Harris overtook Donald Trump.
But the American presidential election is not just about the number of votes. In the United States, the president is not directly elected by the people. He is chosen by a college of 538 electors distributed between each state. A candidate who comes first in a state wins all the voters associated with it. This is the principle of “winner takes all” (“winner takes all”) which prevails in most states (except two, Maine and Nebraska). The map below shows the average voting intentions in states where polls have been published over the last 14 days.
A category of States stands out. These are territories with a large number of electors in which the candidates are neck and neck. These are the pivotal states, or swing states in English. According to franceinfo calculations, there are eight of them: Arizona, North Carolina, Georgia, Florida, Michigan, Nevada, Pennsylvania and Wisconsin. It is in these territories that the outcome of the election should be decided.
“The polls at the level of these states are all within the margin of error and they are generally less precise than the national polls, with often lower means and an electorate more difficult to target”explains François Vergniolle de Chantal, professor of American studies at Paris Cité University.
This specialist believes that opinion surveys in these territories nonetheless remain very important. “They reveal that there is no collapse in the Trump vote, which may otherwise be underestimated”he specifies. The grid of graphs below shows the evolution of voting intentions in the pivotal states.
The averages displayed on this page are calculated over rolling 14-day windows, from a selection of polls from the database compiled by the American political analysis site 538 (owned by ABC News). Each survey used in our calculations was filtered and weighted according to a list of reliability and quality criteria established by franceinfo, based on the methodological notices of 538, New York Times of The Economist and interviews with specialists.
The Democratic Party and the Republican Party are not the only ones to present candidates for the American presidential election. There are also third-party applications which differ by state. Polls which take into account these “small candidates” are not included in the calculation of our averages.
More than 200 pollsters are referenced for the 2024 presidential election alone. To sort through this list, franceinfo only uses certain polling institutes, based on the reliability scores assigned by the site 538, on a scale of 0 to 3. At the national level, all the pollsters we use have a rating greater than or equal to 1.9/3. In the pivotal States, this threshold is lowered to 1.5/3. There is no filter for other states.
Some polls are sponsored by candidates’ supporters or their campaign teams. These investigations are described by the 538 site as “partisan”. They can overestimate by six points the result of the candidate who is supported, according to the New York Times . These partisan polls are dismissed by Franceinfo.
Polls can give several results depending on the type of sample questioned: voters likely to vote, voters registered on the electoral lists or individuals of voting age. In this case, franceinfo only keeps the most qualitative result, that is to say that of respondents who are thinking of going to vote or, failing that, those who are registered on the electoral lists. We calculate an average when a survey includes multiple questions for the same profile of respondents.
Some surveys are carried out continuously. In France, these investigations are called “rolling”. They make it possible to produce results updated daily, by renewing only part of the sample of people questioned each day. For surveys of this type, franceinfo only uses surveys carried out over non-overlapping periods, starting from the most recent publication.
A different weight is assigned to each survey depending on its sample size. The larger this is, the higher the weight of the survey will be for calculating the average. Sizes are capped at 5,000 to avoid giving excessive importance to the largest surveys. The quality of the sample is also taken into account, with greater weight given to voters likely to vote compared to others, and to voters registered on the electoral lists compared to individuals of voting age.
To prevent a single polling institute from having too much weight compared to the others, a weighting coefficient is applied according to the number of polls published by each pollster over a sliding window of 14 days.
In the map representing voting intentions in American states over the last 14 days, certain results are described as very close: neck and neck. To enter this category, the confidence intervals of the polls must overlap and the difference between the averages of the candidates’ voting intentions must be less than 5 percentage points.