Your 'Citizen Forecast' calls the presidential race for...

ANALYSIS

These days, poll number of vote intention and other data (aka "the fundamentals") are the go-to tools for forecasting election outcomes. Analysts spend hours, writing hundreds of lines of code, to craft these forecasts—only to have them fiercely debated by election nerds. Campaigns and the media use these models to shape election narratives, raise funds, and communicate with supporters. But there's another, simpler method called a "citizen forecast," or as you might know it, "the wisdom of the crowds."

Instead of asking people who they will vote for, which requires that people be honest about that question AND the survey’s likely voter model (their best guess of what the electorate will look like) to be accurate, you instead ask survey respondents to predict the outcome of the election. Researchers then would take the average of the respondent’s predictions to find the “citizen’s forecast” of the election.

Researchers suggest that asking voters to predict the outcome of an election is more accurate than asking them who they personally plan to vote for due to the diversity of information respondents draw from. When asked to predict, people draw from a broader range of sources—like insights from their social circles, news, personal experiences, and historical trends. On the other hand, asking about their own vote only requires them to reflect on their personal choice. The idea is that when you gather predictions from a large, diverse group, you tap into all of that information like some sort of supercomputer. Another reason this method may work better is that predicting an outcome rather than stating personal preferences might reduce the fear of judgment, otherwise known as a desirability effect.

Fun fact: researchers have found this approach is actually more accurate than polling of vote intention. Further, the Trafalgar Group used a similar strategy to predict Trump’s win in 2016 (though they changed their methods in 2020, which got the election wrong), and we applied it ourselves for the 2022 Illinois gubernatorial race. Spoiler alert: our citizen forecast outperformed the polls of vote intention and forecasting models, more accurately predicting Governor Pritzker's margin of victory over Darren Bailey than voting intention questions or election forecasts of the race. Illinois voters 1, election nerds 0.

Now, with a U.S. presidential election just around the corner, you can probably guess where we’re headed with this! One big question that came up after 2022 was whether citizen forecasts could outperform traditional polling or electoral forecasts in a closer race. Let’s face it, the Illinois race wasn’t exactly neck-and-neck. Most credible polls didn’t even try to pretend it would be. But what about a much tighter contest? Enter the 2024 presidential race, which is, at the very least perceived, to be a close one.

To put the wisdom of the crowds to the test, we gathered responses from 2,000 Americans as part of a broader project, making sure our sample reflected the diversity of America in terms of gender, race/ethnicity, education, region, community type, partisanship, and income. Our survey ran from September 25 to October 3—based on prior research, the sweet spot for a citizen forecast is about 30 to 40 days before the election.

We also wanted to compare our citizen forecast with traditional forecasting models, so we pulled predictions from popular sources on September 30, right in the middle of our data collection. Since the Electoral College is what really decides the presidency, that’s where we focused on for our means of predicting the election.

In a fun twist inspired by research on how well people can guess economic factors, we ran an experiment. Half of our respondents were reminded the Electoral College results from 2020, while the other half weren’t. Both groups got a map with the Electoral College values for each state that did not contain previous or predicted results. In keeping with previous research on the wisdom of the crowds, we used median score to filter out extreme values from people who went with unrealistic totals (like 538 or 0 for either candidate).

Here’s how the predictions stacked up:

Across the board, whether they had information or not, our respondents predicted narrow Electoral College victories for Kamala Harris. Interestingly, those without the 2020 election info were slightly more confident in their predictions. This suggests a slight impact for information, but whether it makes voters more or less accurate in their forecast, we won’t know until after the election is called.

In addition to asking respondents their prediction for the national result, we asked them which candidate they thought was most likely to win in the respondent’s home state. We tallied up the predictions in each state and awarded the state’s electoral vote to the candidate with the most predicted success. That process produced the map below:

UIS Center for State Policy and Leadership

/

uis.edu/cspl

This forecast is the most bullish for Harris with a 305 to 233 tally in her favor. To be sure, there are some unlikely predictions on the map—you won’t find many people predicting Harris to win Arkansas or Trump to win all of the votes in Maine—but there is evidence to suggest respondents will do even better at predicting events when they are a little closer to home. (Maybe they know something the rest of us don’t!)

One of the things to note about the above is that by design our citizen’s forecast was out of the field before the final month of the election. Considering how close this election has been, it’s entirely possible that slight shifts in the electorate over the past month would be missed by this method or that the same voters may have a different prediction one week out as opposed to one month out. Maybe that will be our next extension of work in this area.

Before Election Day tomorrow, drop your predicted Electoral College outcome in the comments and see if you do better than either the pros or the citizen’s forecasts!

Thanks for reading.