Question 1

What polls are included?

Accepted Answer

We only use the highest quality state and national polls in our polling averages. That means we only include polls that are transparent about how they are conducted, are not paid for by candidates or partisan groups and use methods that have demonstrated accuracy. The Post considers each survey’s methods in detail, including whether the sample was drawn using random sampling methods, how respondents are contacted and how samples are stratified or weighted to accurately represent the electorate. Historical accuracy in elections or other settings is also considered in identifying which polls to include. Our polling averages reflect a Harris-Trump matchup, following President Biden’s decision to withdraw from the presidential race on July 21, although we continue to put some weight on polls from before Biden dropped out. These polls do not heavily influence our averages, but do help stabilize the averages in states for which we only have a few polls since Biden dropped out. We also now include polls in our averages that measure support for third-party candidates following independent presidential candidate Robert F. Kennedy Jr.'s decision to drop out of the race on Aug. 23 as fewer voters now support a third-party candidate in polls. Responses of undecided, “other” and those backing a third-party candidate are assigned to the two candidates, based on their vote share in the poll. Responses of not planning to vote are not included. Our polling averages account for the difference in margin of polls that explicitly include third party candidates and those that do not. Read more about the polls included in our averages.

Question 2

What other data is included?

Accepted Answer

Our averages also factor in how each state voted in the last two presidential elections. This isn’t our preferred method for gauging where the presidential race stands, but if we don’t have a lot of polls, this is a helpful proxy. The more polls we have, the less our averages rely on this metric.

Question 3

How are the averages calculated?

Accepted Answer

There are three factors that drive our polling averages: state polls, national polls and a state’s presidential voting history in the last two elections. Our preference is to prioritize state polls in our calculations as they’re the best indicator of what’s happening in a given state, but we also use national polls in each of our state’s polling averages. National polls give us a sense of the race overall and where it might be headed next, and in states where we don’t have a lot of high-quality state polls, we can use national polls and a state’s voting history to see how the presidential race is shaping up there.

Question 4

How do you account for the possibility of a polling error?

Accepted Answer

At the individual poll level we try to account for the possibility of a polling error in three ways. First, we include a larger sampling error than a poll’s standard margin-of-error calculation suggests. Research shows that the standard margin-of-error calculation (based on a poll’s sample size) underestimates the uncertainty present in a poll’s result. Next, we also try to account for the possibility of measurement error, or the idea that error that can be introduced when any process (in this case, people’s public opinion) is being measured. Finally, another source of error we try to account for is the uncertainty that comes with the choices (like weighting or likely voter screens) that pollsters make in the polls we use. But we also try to account for systematic errors in our averages, like the possibility of a polling miss on the scale of what we saw in the presidential elections of 2020 and 2016. We found in the last few presidential cycles that the average modeled polling error in competitive states was 3.5 percentage points, so to account for this, our averages factor in the 90th percentile possible error (i.e., how bad would the error be in the worst 10 percent of cases). However, this doesn’t affect the output, or what our averages indicate the most likely outcome is. After all, it’s impossible to know before an election whether a polling error incorrectly measures support for Democrats or Republicans. Rather, our aim is to show where our estimate falls in a range of uncertainty. Our polling average page shows how our 2024 state averages would change under a 2020, a 2016 and a 2012 polling error.

Question 5

What about the economy? Does that affect your averages?

Accepted Answer

Economic data is not included in our polling averages because it doesn’t improve one’s understanding of the current state of a race (unlike in a forecast when it can be helpful). However, we do use economic data to give readers a sense of what might happen between now and Election Day. We prepared two scenarios, one in which the economy improves until Election Day and one in which the economy deteriorates. Our model examines how the economic situation affected previous elections, looking back all the way to 1952. What we found is not that surprising: If the economy is doing well, the incumbent party (in this case, Harris) does better; if the economy is doing poorly, then challengers (Trump in this cycle) do better. In choosing economic indicators for our model, we determined that gross domestic product per capita growth and the University of Michigan’s Consumer Sentiment Index were most useful. We also decided that our proxy for a good economy would be economic change between now and Election Day similar to what we saw in the 1996 presidential election, while a bad economy would be represented by changes similar to what we saw in the 2008 presidential election. It’s important to remember that our economic outlook is not a prediction for what will happen; instead, it’s supposed to give you a sense for how things might change. A lot of things are possible between now and Election Day.

Question 6

Are you going to release the code of your model?

Accepted Answer

We really want to and are working on that. When we release the code, we’re also hoping to publish a more technical explanation. Acknowledgments Our polling averages are based on the previous work of Drew Linzer and Pierre-Antoine Kremp, and we’re very thankful for their contributions in this field. We also want to thank David Arbour, John Cherian and Drew Dimmery for their help in the modeling process. Prior to Robert F. K thank David Byler, Will Jennings, Christopher Wlezien and Robert Erikson for sharing the polling data they have collected. Without them we would not have been able to build this model.K

How The Washington Post’s presidential polling averages work

Everything you need to know about The Post’s polling averages, from the polls we include to how we handle polling error and uncertainty.

Election 2024