Prospective voting and the issues and leaders model: Forecasting the 2024 U.S. presidential election

Abstract

This paper presents the Issues and Leaders model for forecasting the 2024 U.S. presidential election’s popular vote, grounded in the theory of prospective voting. The model focuses on voter perceptions of candidates’ competence in handling key issues and leadership qualities, departing from traditional retrospective models based on incumbent performance and economic fundamentals. Designed with campaign strategy in mind, the model provides actionable insights for messaging and public positioning. Throughout the final 100 days of the campaign, it consistently predicted a close race. Its Election Eve forecast—a near tie with Donald Trump at 50.2% and Kamala Harris at 49.8% of the two-party vote—was more cautious than most polling-based forecasts, many of which showed a Harris lead. Ultimately, Trump won the popular vote by 1.5 points, and the model’s average error across the campaign was just 0.65 points. By emphasizing dynamic voter evaluations, the model offers a forward-looking framework for understanding electoral behavior and improving election forecasting.

Introduction

Building models to forecast elections has a long history in political science. Traditionally, many models have been grounded in the theory of retrospective voting. This theory posits that voters reward incumbents for strong performance and penalize them for poor outcomes, with performance typically measured using economic indicators and supplemented by public opinion data, such as the incumbent’s job approval rating (Tien and Lewis-Beck, 2025). While the retrospective voting model offers a compelling forecasting framework (Norpoth, 1996), its assumptions become less reliable in open-seat elections where no incumbent is running—such as the 2024 U.S. presidential election, following President Biden’s withdrawal from the race. Some models account for open-seat elections this by incorporating variables like whether the incumbent is running (Mongrain et al., 2025) or how long the incumbent has held office (Abramowitz, 2016).

However, voting behavior is not purely retrospective. Many voters also base their decisions on prospective evaluations, assessing candidates by their expected future performance rather than past outcomes. Lewis-Beck and Tien (1996) demonstrated that including prospective variables—such as expectations about peace and prosperity—can enhance the predictive accuracy of election forecasts, highlighting their theoretical and empirical relevance. Building on this, Nadeau and Lewis-Beck (2001) refined our understanding of when prospective evaluations matter most. They found that such forward-looking assessments are especially influential in open-seat elections, where a sitting president is not seeking re-election. In the absence of a direct performance record for the incumbent party’s new candidate, voters shift their attention toward anticipated (economic) performance, making prospective evaluations more decisive than retrospective ones.

Despite strong theoretical and empirical support for prospective variables, most forecasting models still rely primarily on retrospective specifications (Lewis-Beck and Stegmaier, 2013). Nevertheless, many models have adapted somewhat by incorporating prospective elements indirectly—for example, by substituting retrospective indicators such as incumbent approval with trial-heat polls that reflect voter preferences for current candidates. Foundational work by Campbell and Wink (1990) and Erikson and Wlezien (2021), which offered one-off forecasts approximately 2 months before Election Day, laid the groundwork for today’s more dynamic forecasting approaches. Contemporary models such as Nate Silver’s Silver Bulletin, The Economist’s forecast, and the now-discontinued FiveThirtyEight—build on this foundation by providing daily updates that integrate economic indicators with recent polling data. These models have become widely followed, offering a steady stream of predictive insights to satisfy the growing public demand for real-time election forecasts.

Despite their popularity, the decision-making implications of these models are limited. While they can help campaign strategists allocate resources, particularly in identifying competitive states where funds might be best spent, they fall short in offering actionable insights on campaign messaging. These models cannot advise on which issues should be prioritized, nor can they provide guidance on shaping a candidate’s public image or tailoring communication strategies during the election cycle.

The Issues and Leaders model, first introduced by Graefe (2013) to predict the 2012 U.S. presidential election and subsequently applied in 2016 and 2020 (Graefe, 2016, 2021), was developed to address these limitations by offering strategic insights for campaign decision-makers. The model incorporates short-term campaign dynamics by focusing on voters’ perceptions of candidates’ competence in handling key issues and leadership qualities. It departs from traditional election forecasting by building on prospective voting theory, emphasizing expectations about future performance rather than past outcomes or economic fundamentals. This forward-looking approach can provide campaigns with actionable insights, potentially enabling strategists to craft more effective messaging that highlights candidates’ strengths in issue-handling and leadership. As shown in Table 1, the Issues and Leaders model demonstrated strong predictive accuracy across the 2012, 2016, and 2020 elections, with forecast errors consistently within one percentage point of the actual two-party vote share for the incumbent party. These results held despite varying lead times, which were determined by the publication schedule rather than selective use of favorable data. For comparison, polling averages at the time of each forecast show much larger errors, particularly in 2020. Across the three forecasts from 2012 to 2020, the model’s mean absolute error was just 0.84 percentage points—roughly one-fourth the error of 3.36 points for the RCP polling averages—highlighting its robustness and reliability, even when issued weeks or months before the election.

Table 1.

Forecast accuracy of the issues and leaders model compared to polling averages, 2012–2020.

					Polls		Issues and leaders model
Election year	Incumbent party	Forecast date	Days to election	Election outcome	RCP average	Error	Forecast	Error	Forecast source
2012	Democratic	Nov 05	1	52.0	50.4	−1.64	51.3	−0.66	Graefe (2013)
2016	Democratic	Oct 10	29	51.1	53.2	2.12	52.1	0.99	Graefe (2016)
2020	Republican	Aug 19	76	52.3	46.0	−6.32	51.4	−0.87	Graefe (2021)

Note: Forecasts and polling averages refer to the incumbent party’s two-party popular vote share. Forecast error is calculated as the difference between the forecast and the actual result (forecast – actual). Model forecasts and RealClearPolitics (RCP) polling averages are taken from the same day, with forecast dates corresponding to the publication dates in the original sources.

Model description

The Issues and Leaders model builds on the well-established framework that party identification, issue positions, and candidate evaluations are key determinants of vote choice (Asher, 1992; Campbell et al., 1960; Clarke et al., 2011). Unlike models that predict survey responses, it is explicitly designed to forecast the incumbent party’s share of the two-party popular vote. While survey data are used to capture perceptions of issue-handling competence and leadership, the dependent variable is the actual election outcome.

Vote equation

The model’s OLS specification is given by V_t = P + bI_t + cL_t, where V_t refers to the incumbent’s share of the two-party popular vote, while I_t and L_t represent the incumbent’s issue handling competence and leadership perception, measured t days before Election Day. The intercept P serves as a proxy for underlying partisan predispositions in the electorate.

Measuring issue-handling competence

Following Campbell et al. (1960: 170), the model assumes that for an issue to influence voting behavior, three conditions must be met: voters must be aware of the issue, perceive it as important, and believe that one candidate is better equipped to handle it. To measure issue salience—combining voter awareness and importance—the model relies on data from Gallup’s “Most Important Problem” question, which asks voters to identify the country’s most pressing issues. Each issue is categorized as either economic, foreign policy, or other. The percentage of respondents citing issues in each category is cumulated, and the relative salience of each category is determined by dividing the cumulative mentions in each category by the total mentions across all categories. Additionally, the model draws on polling data that ask respondents which candidate they trust more to handle specific issues—such as, “Who do you trust more to handle the economy, Kamala Harris or Donald Trump?” All polls were sourced from FiveThirtyEight.com. The dataset includes 586 issue-handling questions drawn from 87 unique surveys conducted between October 2023 and November 2024. These questions span 55 distinct issues, which are grouped into three categories: economic (14 issues), foreign policy (10), and other (31). In total, the dataset comprises 135 entries on economic issues, 95 on foreign policy, and 356 on other issues. On each day, the model calculates the two-party support for the incumbent on each issue, then averages the incumbent’s support across all issues within each category. The incumbent’s issue-handling score is then determined by averaging the scores across the three categories, weighted by the relative salience of each category. This score is updated whenever new polling data become available, using exponential smoothing, which assigns 30% weight to the most recent data and 70% to the previous average.

Measuring leadership perception

To measure leadership perception, the model relies on polling data that ask respondents which candidate they believe would be the better leader—for example: “Regardless of how you intend to vote, who do you think is a stronger leader: Kamala Harris or Donald Trump?” All polls were sourced from FiveThirtyEight.com. The leadership component is based on 22 entries from 22 unique surveys conducted between February and October 2024, each assessing voter perceptions of the candidates’ leadership qualities. In contrast to the issue-handling component, which covers a variety of policy areas, the leadership polls focus exclusively on the single dimension of who is perceived as the stronger or better leader. For each poll, the model calculates the incumbent’s two-party support to generate a leadership perception score. This score is continuously updated over the course of the campaign using exponential smoothing, assigning 30% weight to the most recent polls while retaining influence from earlier data—ensuring responsiveness to evolving voter sentiment without overreacting to short-term fluctuations.

Model estimation

The Issues and Leaders model is estimated as a time-series regression using aggregate data from 13 U.S. presidential elections between 1972 and 2020, with each election treated as a single unit of observation. This approach allows the model to capture long-term trends in voter behavior and campaign dynamics. The vote equation can be recalculated daily leading up to Election Day, enabling the model to reflect how the influence of leadership and issue-handling perceptions may evolve over time. Table 2 presents the estimated coefficients for two points in the forecast horizon: Election Eve and 100 days prior to Election Day. The results illustrate that both leadership and issue-handling perceptions are important predictors of the incumbent vote share, with their effects strengthening as the election approaches. For instance, the coefficient for issue-handling increases from 0.307 at day −100 to 0.465 on Election Eve, while the model’s explanatory power (R²) rises sharply from 0.335 to 0.965 over the same period—demonstrating the growing predictive value of voter evaluations as the campaign progresses.

Table 2.

Estimated coefficients of the issues and leaders model: comparison between election eve and 100 days before the election.

Effect	Election eve	100 days before election
Party identification	10.844	24.615
	(2.4505)	(12.434)
Leadership	0.315	0.235
	(0.0444)	(0.139)
Issue-handling	0.465	0.307
	(0.0496)	(0.214)
R2	0.965	0.335

Notes: Estimated based on data from N = 13 elections from 1972 to 2020, standard errors in parentheses.

Figure 1 displays the trend lines for the model’s regression coefficients and intercept over the final 100 days of the campaign. These trends illustrate how the predictive influence of each component evolves as Election Day approaches. Early in the campaign, around 100 days before the election, the intercept is relatively high—just under 25 percentage points—suggesting that a large share of the incumbent vote can be attributed to baseline partisan alignment or party identification. As the campaign progresses, the intercept steadily declines to around 10.8 points on Election Eve, indicating that fewer voters rely solely on party loyalty when making their final decision.

Figure 1.

Trend lines of model coefficient estimates over the last 100 days of the campaign, based on elections from 1972 to 2020.

At the same time, the coefficients for issue-handling competence and leadership perception increase steadily, reflecting a growing emphasis on candidate evaluations as the campaign progresses. Voters appear to place increasing weight on which candidate is better equipped to address pressing issues, with issue competence gaining particular importance. Leadership perception also becomes more influential, though it consistently plays a somewhat smaller role than issue-handling in shaping voter preferences. These shifts imply that as Election Day nears, voters move from relying on partisan cues to making more informed judgments based on candidate-specific strengths. This evolution is mirrored in the model’s improving fit over time, with higher explanatory power as the campaign draws to a close.

Forecasting the 2024 U.S. Presidential election

Figure 2 tracks Kamala Harris’ lead in issue-handling competence and leadership perception from July 28th, 100 days before the election, to Election Eve. While Harris began with and maintained a slight edge over Donald Trump in issue competence, she started at a significant disadvantage in leadership perception, trailing by 20 points. However, over the course of the campaign, Harris steadily narrowed this gap, bringing it down to less than five points.

Figure 2.

Harris’ lead in issue-handling, leadership perception, and the predicted vote margin in the Issues and Leaders model and polls.

The 2024 forecast is produced by applying the vote equation—using coefficients and the intercept estimated from historical data (Figure 1)—to the observed values of issue-handling competence and leadership perception for the 2024 candidates. The resulting forecasted vote share is presented in Figure 2 as Kamala Harris’s lead in percentage points. As an out-of-sample, ex ante prediction, the forecast relies solely on model parameters derived from data covering 13 presidential elections between 1972 and 2020. By excluding any data from the current election cycle in the model estimation, this approach ensures the forecast remains independent of contemporary outcomes, thereby preserving its validity as a true out-of-sample prediction.

The forecast remained remarkably stable throughout the campaign, ranging between a 1.6-point lead for Trump and a 1.2-point lead for Harris, with Trump maintaining an average advantage of 0.2 points. On Election Eve, the model projected a near tie, predicting a narrow lead for Trump of 0.4 percentage points in the two-party popular vote (50.2% vs 49.8%). Compared to most other forecasts, including polling averages, this projection was notably more cautious about Harris’s chances. For example, the PollyVote.com (Graefe, 2025) tracked estimates from eight different poll aggregators. The typical poll aggregator had Harris up on average by one percentage point on Election Eve.

In the end, Trump won the popular vote by a margin of 1.5 percentage points (50.75% vs 49.25%). The final forecast of the Issues and Leaders model therefore underestimated Trump’s vote share by about half a percentage point. Over the entire 100-day forecast horizon, the model’s average error was just 0.65 percentage points. As in the previous four elections, the model demonstrated a high level of accuracy in predicting the incumbent party’s share of the two-party popular vote.

Discussion

The 2024 election outcome provides another validation for the Issues and Leaders model. A particularly notable insight is the role of leadership perception in forecasting Trump’s narrow victory. While Harris held a modest edge on issue competence, Trump maintained a consistent lead in leadership perception—a factor that gained predictive weight closer to the election. The model anticipated Trump’s advantage before most polling averages reflected it, highlighting its potential to signal electoral dynamics that might otherwise be overlooked.

Beyond forecasting accuracy, the model offers practical value by providing a real-time lens into campaign dynamics. By emphasizing prospective voting, it helps identify evolving voter priorities and candidate strengths. This makes it a potentially useful tool not just for forecasters, but also for campaign strategists, journalists, and political observers.

Like other contemporary forecasting models such as Nate Silver’s Silver Bulletin, The Economist’s forecast, and the now-discontinued FiveThirtyEight—the Issues and Leaders model can be updated continuously as new polling data become available. What sets it apart is its exclusive focus on voter perceptions of issue-handling competence and leadership, rather than structural indicators or general vote intention polls. This focus enables the model to deliver more detailed and actionable insights into campaign dynamics, offering a deeper understanding of why candidates gain or lose support—not just who is ahead.

Strategically, the model points to three levers for improving electoral performance. First, candidates should strengthen their reputations on key issues—by asserting ownership of favorable issues or closing perceived competence gaps on less favorable ones. Second, candidates can shape the public agenda by directing attention to issues where they have an advantage and downplaying those that favor their opponent. Third, leadership perception plays a critical role. Since such perceptions are influenced by relatively stable traits—such as personality, professional background, and appearance—parties should consider these factors carefully during the nomination process.

The model also holds value beyond campaign strategy. For journalists, analysts, and researchers, it offers a theory-based alternative to conventional “horse-race” reporting. By connecting public perceptions to predicted vote outcomes, the Issues and Leaders model helps explain the underlying forces shaping electoral dynamics—offering insight not just into who is winning, but why.

At present, however, the model is limited to forecasting the national popular vote. This constraint reflects the lack of state-level polling data on voter perceptions of issue-handling competence and leadership qualities. Expanding the model to incorporate state-level inputs would significantly enhance its utility—enabling Electoral College forecasts and offering more localized insights for campaign planning. Such an extension could capture regional variation in issue salience and candidate evaluations. For example, immigration may resonate more strongly in border states like Arizona and Texas; abortion may play a larger role in states with recent legislative changes such as Ohio or Florida; and economic concerns may dominate in industrial swing states like Michigan or Pennsylvania. A state-level version of the model could help campaigns tailor their messaging to local priorities and improve both strategic targeting and forecast accuracy.