The 2020 presidential election presents forecasters with unique challenges. First and foremost, the election is taking place in the midst of one of the most severe crises that the United States has faced in the past hundred years: the coronavirus pandemic. Not only has the pandemic already killed more than 185,000 Americans, with thousands more likely to die before Election Day; it also has produced the most severe economic downturn since the Great Depression, with the unemployment rate reaching double digits and real GDP declining at an annual rate of 32.9% in the second quarter of 2020.
President Donald Trump already was facing a stiff battle for reelection before the coronavirus pandemic hit the country—consistently trailing the Democratic frontrunner, Joe Biden, in almost every national poll and most swing-state polls in January and February. In recent months, however, Trump’s situation has become increasingly dire, with his approval rating falling from the mid to the low forties and his deficit against Biden in national polls rising from an average of 5 to 6 points to an average of 8 to 10 points.
The “time-for-change” model relies on three predictors to forecast the outcome of presidential elections: the incumbent president’s approval rating in late June or early July, the change in real GDP in the second quarter of the election year, and a dummy variable based on whether a first-term incumbent is running for reelection. This time-for-change factor reflects the fact that first-term incumbents like President Trump generally enjoy a significant advantage—even after controlling for their approval ratings and economic conditions (Abramowitz Reference Abramowitz2016).
The extraordinary conditions under which the 2020 presidential election is being contested mean that the time-for-change model requires substantial modification. There are good reasons to expect that in 2020, two of the model’s predictors—the change in real GDP in the second quarter and the time-for-change dummy variable—will not perform as they normally do.
Although the US economy is currently experiencing a severe downturn, with real GDP falling at an unprecedented rate in the second quarter, voters do not appear to hold the incumbent president responsible. This undoubtedly is because the recession was deliberately induced to try to control the spread of the deadly coronavirus. Thus, despite the massive rise in unemployment and decline in real GDP, Trump’s approval ratings on handling the economy—although they have fallen—have generally remained positive.
It also is the case that in the past 10 to 20 years, assessments of economic conditions have become increasingly divided along party lines, with supporters of the president’s party consistently rating the state of the economy much more positively than supporters of the opposition party. This growing partisan divide in how voters view economic conditions means that economic trends may have less impact on election results than in the past.
Increasing partisan polarization also may have the effect of reducing the electoral advantage of being a first-term incumbent. That advantage was based mainly on the ability of the incumbents to appeal to voters across party lines who might be reluctant to replace a president after only one term in office. However, rising partisan polarization means that voters today are much less willing to support a candidate from the opposing party for any reason. This is especially true in the case of Donald Trump—the most divisive president in modern history according to a Gallup Poll (Jones Reference Jones2020).
A SIMPLIFIED INCUMBENT ACCOUNTABILITY MODEL
Given these concerns, for the 2020 presidential election, I used one predictor to forecast the results: the incumbent president’s net approval rating in late June. In addition, I made two other modifications to my typical presidential forecasting model: (1) using the electoral vote rather than the popular vote as the dependent variable, and (2) limiting the analysis to contests with a running incumbent. Along with the current forecast, I present conditional forecasts based on the president’s net approval rating in late October.
For this forecast, I used only elections like 2020 with a running incumbent. The reason for doing this is that the connection between the incumbent’s job-approval rating and the election outcome is much closer in these contests than in those without a running incumbent. When an incumbent is running for a second term, the election is largely a referendum on that incumbent’s performance during the previous four years. When there is an open-seat election, the public’s appraisal of the incumbent’s performance matters but not nearly as much.
I argue that the extraordinary circumstances under which the 2020 presidential election is taking place, along with deepening partisan polarization, mean that objective economic conditions are unlikely to have much effect on the outcome of the election. This is because it is the coronavirus pandemic that seems to be driving voters’ assessments of President Trump’s performance. Thus, whereas assessments of his handling of the economy continue to be more positive than negative, assessments of his handling of the pandemic—as well as assessments of his overall job performance—have turned decisively negative since April.
…objective economic conditions are unlikely to have much effect on the outcome of the election. This is because it is the coronavirus pandemic that seems to be driving voters’ assessments of President Trump’s performance.
FORECASTING TRUMP’S ELECTORAL VOTE
Table 1 displays the results of a regression analysis of incumbent electoral votes on the late-June approval rating as well as a forecast of the number of electoral votes that Trump would receive in the 2020 presidential election based on an approval rating of −15% at the end of June (Abramowitz 2020). The table also includes the estimated probability of a Trump victory based on this forecast and the standard error of estimate of the regression equation. The model was estimated based on the 11 elections with running incumbents between 1948 and 2012.
Table 1 Regression Analysis of Incumbent’s Electoral Vote with Late-June Net Approval Rating, 1948–2012
Sources: Electoral votes from uselectionatlas.org; presidential approval from the Gallup Poll; and 2016 Trump late-June approval from FiveThirtyEight average.
With a late-June net approval rating of −15%, this model predicts that Trump will lose the electoral vote to Biden by a decisive margin of 319 to 219. However, because the model has an adjusted R2 of only 0.64 and a standard error of estimate of about 92 electoral votes, the model gives Trump about a 30% chance of winning the election. In other words, at this point, according to this simple incumbent-accountability model, Biden is a solid but not overwhelming favorite to win the 2020 presidential election.
As an additional check on the accuracy of the model, I conducted an out-of-sample forecast test for the 11 elections with running incumbents between 1948 and 2012. The model correctly predicted the winner of the electoral vote in nine of the 11 elections. The two erroneous forecasts were for the 1948 and 1976 elections. President Truman was predicted to lose narrowly in 1948. His victory was probably due to a rising approval rating between June and October—however, Gallup did not measure Truman’s approval rating in October. Gerald Ford was predicted to win reelection in 1976. His defeat probably reflected the unusual circumstances of that election: Ford had recently succeeded the extremely unpopular Richard Nixon after Nixon resigned the presidency in disgrace. Voters may have been influenced as much or more by their opinion of Nixon as their opinion of Ford.
We also can estimate a model based on the incumbent president’s net approval rating in late October. Table 2 displays the estimates for this model based on the 10 elections with running incumbents between 1956 and 2012. October approval data were not available for 1948. Not surprisingly, the model is considerably more accurate than the one based on late-June approval ratings. It has an adjusted R2 of 0.82 and a standard error of estimate of only about 69 electoral votes. The estimates in table 2 can be used to produce conditional forecasts of incumbents’ electoral votes based on their net approval rating in late October. Those predictions, along with the estimated probabilities of victories for the incumbent, are shown in table 3.
Table 2 Regression Analysis of Incumbent-Candidate Electoral Vote Share with Late-October Net Approval Rating, 1948–2012
Sources: Electoral vote data from uselectionatlas.org; presidential approval from the Gallup Poll.
Table 3 Conditional Forecasts of Trump Electoral Vote and Probability of Victory Based on Late-October Net Approval Rating
Note: Predictions based on the regression equation in table 3.
The conditional forecasts displayed in table 3 indicate that if Trump’s approval rating remains unchanged in late October from where it was in late June, there is a strong likelihood that he would be defeated in an Electoral College landslide. Under this scenario, Biden would be expected to receive 361 electoral votes to only 177 for Trump, and the president would have only a 9% chance of winning. Based on these results, to have a reasonable chance of winning the election, Trump would have to raise his net approval rating to much closer to the neutral point. A net approval rating of zero would result in a prediction of 262 electoral votes for Trump to 276 for Biden and would give the president a 46% chance of winning.
…if Trump’s approval rating remains unchanged in late October from where it was in late June, there is a strong likelihood that he would be defeated in an Electoral College landslide.
CONCLUSIONS
In 2016, Trump won a majority of votes in the Electoral College but lost the national popular vote by more than two percentage points. The huge discrepancy between these popular- and electoral-vote margins was due to his narrow victories in several swing states, including Florida, Michigan, Pennsylvania, and Wisconsin. However, based on the results presented in this article, Trump’s chances of repeating this feat in 2020 appear to be slim.
When an incumbent president is running for a second term, the election is always largely a referendum on the president’s record during his first term. Normally, an important component of that record is the performance of the US economy, especially during the first half of the election year. In 2020, however, due to the devastating impact of the coronavirus pandemic on American society and the economy, it appears likely that the election will turn much more on the public’s assessment of the president’s handling of the pandemic—which, by the summer of 2020, had turned decidedly negative.
It also appears unlikely that Trump will enjoy the electoral advantage that typically accrues to first-term incumbents. Partisan polarization has drastically reduced the ability of incumbent officeholders at all levels to appeal to voters across party lines. Moreover, unlike previous incumbents, Trump has made little effort to expand his base of support during his time in office.
Based on these considerations, I present a simple incumbent-referendum model for forecasting the outcome of the 2020 electoral vote. The president’s late-June net approval rating of −15% yields a forecast of a decisive 319–219 Electoral College vote victory by Biden. However, the model still gives Trump about a 30% chance of winning the election due to uncertainty about what will transpire between June and November, and the fact that many of the key swing states that will decide the election in the Electoral College tilt slightly more Republican than the nation as a whole. However, if Trump’s approval rating remains at −15% in late October, the model predicts an even more overwhelming defeat, with 361 electoral votes for Biden and only 177 for the president. At that point, Trump would have only a 9% chance of winning the election.
DATA AVAILABILITY STATEMENT
Replication files are available on Dataverse at https://doi.org/10.7910/DVN/NQWLFH.
