The field of election forecasting has grown rapidly, especially since 2008, and its growth has accelerated in recent elections. The availability of more data, easily accessible data, and major statistical and methodological advances have all contributed to a surge in the number of news outlets, data journalists, and political scientists forecasting US elections. The increased interest in election forecasting is reflected in this year’s symposium collection, which features 20 forecasts of presidential (both popular and Electoral College votes), House, and Senate elections. We selected these forecasts using a fast and fair process that started almost one year ago. We sent out an open call to the discipline in November 2019, in which we specifically encouraged new forecasters and underrepresented groups to submit proposals. We received 22 submissions to be included in the symposium. At that stage, we decided to be as inclusive as possible and desk-rejected only a limited number of the initial submissions, mostly for being outside of the scope of the symposium. To blind-review each manuscript, we relied on a prearranged team of dedicated outside reviewers, and we also asked each author who submitted a proposal to act as a blind reviewer. Based on these quick and timely reviews, we accepted 15 articles including three critical reviews of the forecasting field. Without the quick turnaround by the authors and reviewers, this endeavor would have been impossible.

These forecasts are rendered in an election year when predicting the outcome is more challenging than ever, in particular for models that rely on the fundamentals and are informed by historical patterns. As of September 2020, the COVID-19 virus has resulted in more than six million infections and more than 200,000 deaths in the United States alone (John Hopkins Coronavirus Resource Center 2020). This unprecedented pandemic is not only affecting everyday life in the United States but also policy priorities. In addition, social-distancing policies are forcing candidates, parties, and voter-mobilization organizations to rethink their campaign and mobilization practices. Government measures to slow down the spread of the virus have led to “the deepest global recession in decades” (World Bank 2020). As the virus spread, the American economy cratered, resulting in negative economic growth and a spike in unemployment. Thus, the economic indicators that feature prominently in many political forecasting models are extreme outliers this year. This symposium addresses head on these challenges presented by the pandemic. For example, Abramowitz (Reference Abramowitz2020) does not include the economy in his forecast model and Lewis-Beck and Tien (Reference Lewis-Beck and Tien2020) argue for a lower limit in their economic indicators.

There are other challenges to forecasting elections in 2020. First, historically, the American electorate swings between the Democratic and the Republican parties with some regularity, but extreme partisan polarization now reduces the number of potential swing voters. As a result, recent elections have been marked by smaller margins of victory than in the past (Abramowitz Reference Abramowitz2012). Second, the outcome of the 2016 presidential election has drawn attention to a daunting challenge for US election forecasters: the recent and recurring disconnect between the popular vote and the Electoral College vote. Many of the forecasts published in 2016—much like public-opinion polls (as argued by Kennedy et al. Reference Kennedy, Blumenthal, Clement, Clinton and Durand2018)—provided fairly accurate predictions of the two-party vote share. On average, the forecasts were only 1.6 percentage points off Clinton’s vote share (Cuzán Reference Cuzán2020). Still, Trump’s Electoral College win caught many by surprise because it occurred despite his losing the popular vote.

In summary, a pandemic-induced recession, increasing party polarization, and a growing disconnect between the popular vote and the Electoral College vote all call on the field of forecasting to innovate. As the collection of articles in this symposium shows, forecasters are responding to these challenges. The forecasts in this collection rely on new indicators, data, and methods. Reflecting on the question, “What should be predicted?,” several models importantly choose to forecast the Electoral College vote for the first time. The end result is a collection of new and old forecasting models and forecasters who innovate theoretically and methodologically.

A maturing field also needs to reflect on the value and usefulness of election forecasting. Three contributions in this symposium offer such a critical reflection by evaluating how political scientists historically performed as election forecasters (Cuzán Reference Cuzán2020), by thinking thoroughly about the role of predictions in political science (Dowding Reference Dowding2020), and by drawing attention to the risks as well as the opportunities of a forecasting exercise for the discipline (Victor Reference Victor2020).

THE CONTEXT OF 2020

The challenges to forecasting in 2020 also present opportunities. The election will test the quality and accuracy of the forecasting models like no other. Will the strong theories that have been applied successfully to forecasting American elections still apply in a polarized electorate in the midst of an economic recession caused by a pandemic?

Will the strong theories that have been applied successfully to forecasting American elections still apply in a polarized electorate in the midst of an economic recession caused by a pandemic?

Forecasting has shown that parties that have been in power for only one term almost never lose reelection. Incumbent parties that have been in power for eight years or longer, in contrast, are more likely to lose as voters believe it is “time for change” (Abramowitz Reference Abramowitz1988). This year’s election pits an incumbent against the vice president of the previous administration. The last time there was this type of matchup was in 1984, when incumbent Ronald Reagan defeated Walter Mondale, who was Jimmy Carter’s vice president. With a first-term president on the ballot, the “time-for-change” theory predicts a Trump reelection. Abramowitz’s theory will be put to a strong test in Reference Abramowitz2020, but it is telling that Abramowitz himself abandons the time-for-change variable in his 2020 forecast model.

We also have learned that economic conditions in an election year matter to the success of incumbent parties. Well-established models included here (Abramowitz Reference Abramowitz1988; Erikson and Wlezien Reference Erikson and Wlezien1996; Lewis-Beck and Tien Reference Lewis-Beck and Tien1996; Lockerbie Reference Lockerbie, Garand and Campbell2000) use or have used various economic indicators (i.e., Gross Domestic Product, Gross National Product, consumer sentiment about the economic future, income growth, and leading economic indicators) to make accurate forecasts. This year, most of the economic indicators that forecasters rely on are huge outliers. The advance estimate of GDP was −32.9% (annualized) for the second quarter of the election year, and disposable personal income fluctuated widely because of federal stimulus checks and state “shelter-in-place” orders. Will the forecasts that continue to use economic indicators from this topsy-turvy economy affected by the pandemic miss the mark?

Early forecasting models also showed that presidential elections are largely referenda on the incumbent (e.g., Brody and Sigelman Reference Brody and Sigelman1983; Lewis-Beck and Rice Reference Lewis-Beck and Rice1984). Voters reward incumbents who have handled economic and noneconomic issues with competence. This concept typically is measured by presidential approval ratings (see, e.g., Abramowitz Reference Abramowitz1988; Lewis-Beck and Tien Reference Lewis-Beck and Tien1996) or in trial-heat polls (Campbell and Wink Reference Campbell and Wink1990; Erikson and Wlezien Reference Erikson and Wlezien1996). However, will presidential approval ratings still strongly predict outcomes in 2020? Extreme party polarization has changed politics and voting behavior in the postwar period. This year’s extreme polarization presents a challenge to accurate forecasting as more voters evaluate presidents through partisan-colored lenses. Through all of the chaos of Trump’s first term, what remained fairly stable was the president’s approval rating. Trump’s Gallup approval rating never dropped below 35% and never rose higher than 49%. Considering that during Trump’s first term, voters witnessed the Mueller report, impeachment, jailed White House staffers and campaign staff, a soaring stock market, large tax cuts for corporations and the wealthy, the COVID-19 pandemic, mass social and racial protests sparked by George Floyd’s murder, and a shocking break from political norms on an almost-daily basis, it is surprising that Trump’s Gallup approval has fluctuated by only 14 percentage points. By comparison, during his first term, George W. Bush’s approval rating had a range of 44 percentage points with a high of 90% to a low of 46%; Barack Obama’s first-term approval ratings had a range of 26 percentage points (from 66% to 40%). Does this stability in approval ratings caused by extreme polarization make it more or less useful as a predictor of the vote?

Finally, perhaps the toughest test that American presidential forecasting faces head on this year relates to the dependent variable. Previously, most models predicted only the incumbent party’s share of the two-party popular vote (but see Jérôme and Jérôme-Speziari Reference Jerôme and Jerôme-Speziari2016). In 2016, the forecasting postmortem published in PS concluded that 2016 was a great year for political science forecasting because 8 of the 10 published forecasts predicted the two-party popular vote within two percentage points (Campbell Reference Campbell2017, table 1). However, the 2016 election marked the second time in the last five elections that the popular-vote winner was not sworn in as president. In response, political science forecasters appropriately shifted their focus to the winner of the Electoral College vote and away from the popular vote. The dependent variable in seven of our published forecasts is now the Electoral College vote.

Table 1 Summary of the 2020 Presidential Electoral College Vote Forecasts

Note: *Murr and Lewis-Beck provide two forecasts for the Electoral College: 334 and 357. We report the average of those two forecasts in the table.

THE 2020 PRESIDENTIAL AND CONGRESSIONAL FORECASTS

With these challenges described for the forecasting field in 2020, we now turn to what are the actual predictions. In this collection of articles, each forecasts the presidential election, two also forecast the House election, and one forecasts the Senate seat change as well. Seven of the articles provide a forecast of the Electoral College vote, and 10 offer a prediction of the two-party popular-vote share for the incumbent (i.e., Trump).

To summarize the predictions of this year’s models, we present separate tables for the Electoral College vote (table 1), the two-party popular vote (table 2), and the congressional seat change (table 3). Tables 1 and 2 highlight the variation in the level of forecast (i.e., national and state). Many of the Electoral College models produce a forecast for each state, from which an Electoral College vote forecast is generated. The popular vote forecasts, however, still mostly analyze national-level data.

Table 2 Summary of the 2020 Presidential Two-Party Popular-Vote Forecasts

Table 3 Summary of the 2020 House and Senate Forecasts

Turning to the point estimates of the models in this symposium, forecasts for both the Electoral College vote and the popular vote disagree about who will be the next president of the United States. Table 1 shows that of the seven models that offer an Electoral College forecast, five predict a Biden win. The unweighted average of these seven forecasts is 237 Electoral College votes for Trump. Collectively, the contributors to this symposium thus forecast a Biden win in the Electoral College, but it is not unanimous.

Collectively, the contributors to this symposium thus forecast a Biden win in the Electoral College, but it is not unanimous.

The predictions for the two-party popular vote summarized in table 2 similarly suggest that Biden is the favorite to win. The unweighted average of these forecasts is 47.8%, implying that the forecasters collectively predict Trump to lose the popular vote. Again, however, this is not a unanimous forecast—with two of 10 models forecasting a (narrow) majority of the popular vote for Trump.

This collection also includes two contributions that leverage their models to render a forecast for the House and one for the Senate. Lewis-Beck and Tien use their political economy model to predict seat change in the House and the Senate, whereas Lockerbie applies his prospective model to forecast seat change only in the House. Both models predict that the Republican Party will lose seats, although the predicted loss varies widely (table 3).