| Dear Readers: Debra Leiter and Michael Lewis-Beck, two political scientists with lengthy experience in election forecasting, suggest that the public’s opinion on who is going to win (as opposed to asking respondents for whom they plan to vote) has had some predictive value in both the United States and abroad. In today’s guest piece, they summarize four separate polls of citizen expectations of who was going to win the presidential election, with a notable change happening when Kamala Harris replaced Joe Biden as the presumptive and then actual Democratic nominee against Donald Trump. They have previously written about this topic for the Crystal Ball in May 2023 and early June 2024.
— The Editors |
KEY POINTS FROM THIS ARTICLE
— Asking people who they believe will win the election, as opposed to who they are going to vote for, can be an effective tool for trying to figure out who is favored in an election.
— A series of four different polls asking Americans who they believe will win the election showed that more expected Donald Trump to win during the summer, but more recently these expectations have moved in favor of Kamala Harris.
How citizens forecast the election
The 2024 U.S. presidential election has been characterized by uniquely high levels of uncertainty. And of course, there are good reasons why. While just in June of this year the race seemed to be essentially a repeat of 2020, the dynamics of the race have changed, sometimes dramatically, in a few short months, perhaps most dramatically the change in presumptive Democratic nominee from incumbent Joe Biden to Vice President Kamala Harris. But with as much uncertainty as we see discussed in the election, citizens may have a stronger sense of how the election is turning. With respect to forecasting US presidential elections via public opinion polls, the most popular approach consults voter intention surveys, where respondents are asked something like, “In the November election, which candidate do you intend to vote for?”
But an alternative approach treats voters not just as a weather vane for campaigns, but as a source of information about the overall state of the election. This approach uses voter expectations, where respondents are asked something like “In the November election, which candidate do you think will win?” A growing body of evidence indicates that “citizen forecasting” (CF), as the latter method has been called, makes for more accurate predictions of the winner. Indeed, studies of CF in the United States and the United Kingdom, as well as work on other democracies (such as Canada, France, or Germany) have demonstrated that voter expectations outperform voter intentions in terms of predictive accuracy, yielding a higher level of statistical and substantive significance. Inspired by this impressive track record, we have been systematically recording citizen forecasts with regard to the 2024 U.S. presidential election, via nationally representative samples of American voters, as drawn by the Verasight survey team.
Initial soundings from this research were published in the Crystal Ball and again, about a year later. These preliminary results showed President Biden behind former President Donald Trump, in terms of voter expectations—even 600 days before the election, citizens reflected substantial uncertainty about Joe Biden’s ability to win reelection.
In four waves starting in June 2024, we see dramatic shifts in citizen beliefs about the elections. While June showed a competitive race with a minor Trump advantage, one month later marked a nadir in citizen expectations for the Democratic candidate. But more recent waves show a true reversal of fortunes for the Democrats, with citizens flipping their previous estimations on their head, as seen in Figure 1 below.
Figure 1: Citizen forecasts for 2024 presidential election
Note: The question asked in each iteration of the poll was “Which candidate do you think will win the most votes in the November 2024 US Presidential Election?” Donald Trump was the Republican option in all four polls; Joe Biden was the Democratic option in the first two polls and Kamala Harris was the option in the final two polls.
Source: Verasight polls conducted June 20-26, July 20-22, Aug. 20-26, Sept. 20-Oct. 2. N = 750 for all four polls.
With the campaign heating up, June revealed the Trump expectations lead had narrowed. Trump outpaced Biden by only 4 percentage points, 46% to 42%, a gap that was statistically insignificant. (Verasight survey, Wave 1, June 20-26, 2024.) The data are weighted to incorporate standard demographic characteristics and partisanship.)
The percentages, as point estimates, suggest Biden was behind. However, taking into account the margin of error (+/-4 percentage points), the race was, arguably, a tie. (Moreover, when the question was asked about a win of the “Electoral College” directly, the results showed no substantive change in terms of interpretation). Importantly, at this point in the race, while 76% of Democrats believed their party candidate could and would be successful, 86% of Republicans were confident that their candidate could succeed.
These findings, from Wave 1, provide a telling backdrop for the results from our three subsequent Verasight CF waves (put into the field in July, August, and September). Wave 2 (July 20-22), took place one week after the Trump assassination attempt (on July 13) and during the Biden resignation (on July 21). With President Biden suddenly dropping out, it was not immediately clear who the Democratic nominee would be. While Harris presented an obvious option, other names were floated in the media. Indeed, the day after Biden dropped out was a low point in our data for citizen belief that the Democrats could win. Indeed, even Democratic respondents had lost the faith, with just 58% of Democrats thinking a Democratic candidate could win, versus a 90% estimation among Republicans of their candidate’s eventual success. This uncertainty about the nominee, coupled with Biden’s decision to resign late in the campaign, appears to have impacted citizens’ current assessment. The July iteration of the poll showed that 54% of respondents thought Trump would win to only 32% for Biden.
We observe a surge in expectations of a Trump victory, with the Republican choice leading the Democratic choice by 22 percentage points. The encircling dramatic political events appear to have had an effect. What particular events? We get some clues by breaking the results down according to respondent day-of-interview. Recall that Biden announced that he was ending his campaign on July 21; thus, from that moment, news began to spread. If we look at partisan expectations of a Democratic victory by day, we see it drops by 9 percentage points, from July 20 to July 22, while the Republican expectations rise steadily to a plus 11 percentage points by July 22.
With respect specifically to the effects of his resignation, these pre-post results are highly suggestive. That decision was surrounded by other crises, including the assassination attempt, uncertainties about candidate succession, and the possible loss of incumbency advantage that Biden might have provided. Taken together, these events amounted to a massive shock. The question then becomes how long the negative shock would be felt. An answer emerges from the data of Wave 3 (Aug. 20-26). Kamala Harris officially secured the Democratic Party nomination as their presidential candidate on Aug. 5, two weeks or so before this new voter sample was interviewed about their expectations regarding candidates Harris and Trump.
We observe a dramatic reversal of fortune. Voter expectations of a Harris victory are at 56%, compared to only 40% for Trump. Expectations of a Democratic candidate win now exceeded expectations of a Republican candidate win by 16 percentage points. One issue, of course, concerns whether this favorable gap toward Harris would sustain itself beyond the initial enthusiasm of August. The data from Wave 4, gathered around the end of September (Sept. 20-Oct. 2), indicate that expectations of a Harris win sustained themselves at least through that point in the campaign, with a large gap (of 13 percentage points) between Harris and Trump remaining: 55% thought Harris would win to only 42% for Trump.
An important element of citizen forecasting is understanding who is predicting specific outcomes. By the final wave of our panel, we see that a Harris victory is expected by most major demographic groups. Harris has the edge for both men and women (54% and 56%), across all age demographics—although Trump performs best among the 31-55 age range (45%)—and for respondents with either some college (55%) or college graduates (61%), although a Trump victory is still predicted by those with high school-level educations (50%). Partisans, as we might expect, do estimate that their candidate has an advantage, but for the first time in the election, Democrats have an equally strong belief that their candidate will win compared to Republicans (86% vs. 84%). As the election draws nearer, more and more citizens from across demographic groups were increasingly painting a more certain portrait of the 2024 election outcome.
Figure 2: Citizen Forecasts by Demographic Groups
Note: The question asked in each iteration of the poll was “Which candidate do you think will win the most votes in the November 2024 US Presidential Election?” Donald Trump was the Republican option in all four polls; Joe Biden was the Democratic option in the first two polls and Kamala Harris was the option in the final two polls.
Source: Verasight polls conducted June 20-26, July 20-22, Aug. 20-26, Sept. 20-Oct. 2. N = 750 for all four polls.
Conclusions
This current citizen forecast points to a Harris victory in November. How confident can we be in this forecast? Fortunately, we have solid baseline data on the accuracy of voter expectations from past US presidential elections. In the American National Elections Survey (ANES), they have regularly posed the question, “who do you think will win the coming election?” These responses have been shown to reliably forecast the winner. For example, one of us (Michael Lewis-Beck) and Charles Tien demonstrated that the majority of voters (on average 71%) surveyed accurately forecast the winner, and in the aggregate called 9 of 11 elections (1956-1996). (Moreover, the median ANES interview date was Oct. 7, quite close to our dates for Wave 4). In our other recent work, from 2000-2020, voters still seem able to do a good job forecasting U.S. presidential elections, with voters correctly forecasting the winner 66% of the time. Of course, close races are hard to call, such as 2000 or 2016, and this uncertainty reflects itself in the expectations percentages. However, across the entire series, 1956 to 2020, whenever the expectation percentage has exceeded 50%, as is the case with the Harris-Trump race, the forecast of the presidential winner has always been correct. In general, there exists the concern over whether these expectations convert into a win for Harris in the Electoral College. We did ask our respondents this question—“Which candidate will win the Electoral College vote in November 2024?”—and they continued to expect a Harris victory. In sum, in an uncertain election, citizens are telling an increasingly clear story about who will win—and November will be their ultimate test.
| Debra Leiter is an associate professor in the Department of Political Science and Philosophy at the University of Missouri-Kansas City and the secretary- treasurer of the Political Forecasting Group. Michael S. Lewis-Beck is F. Wendell Miller Distinguished Professor of Political Science at the University of Iowa. He has authored or co-authored over 330 books and articles and is a past president of the Political Forecasting Group of the American Political Science Association. |