In 2016, Donald Trump shook the political scene by becoming the Republican nominee for president and by subsequently winning the White House. His election came as a surprise to a majority of pollsters and political observers who expected Hillary Clinton to become the 45th president of the United States. But on November 8, 2016, Hillary Clinton instead became the fifth candidate in US electoral history to win the popular vote (PV) but to lose the presidency as she failed to get more electoral votes (EV) than her main opponent.
Most forecasting models for American presidential elections provide estimates of the national two-party vote. Because popular vote winners generally win a majority of Electoral College votes, these models can assume that their forecast offers a clear indication of who will end up in the Oval Office. Evidently, this assumption is not always warranted. In 2016, Hillary Clinton outpaced her Republican opponent by close to three million votes and nevertheless lost the election to Donald Trump, who achieved his 2016 “tour de force” by winning key swing states in the Rust Belt region by razor-thin margins—more precisely, Michigan, Pennsylvania, and Wisconsin.
The Electoral College makes the geographical dimension of presidential elections particularly salient for the forecasting community. A number of scholars have acknowledged this fact by proposing modeling strategies that use the state as the primary level of analysis (see, e.g., Berry and Bickers Reference Berry and Bickers2012; Campbell Reference Campbell1992; Campbell, Ali, and Jalalzai Reference Campbell, Ali and Jalalzai2006; Klarner Reference Klarner2012).Footnote 1 Steven Rosenstone (Reference Rosenstone1983) was perhaps the first political scientist to embrace this approach. His forecasting model was more comprehensive than previous similar efforts, because it included various indicators related not only to economic matters but also to partisanship, incumbency, region, and a number of political issues.
In this short article, following the path set by Rosenstone and others, we develop a forecasting model that draws on data from the 50 states, as well as the District of Columbia, since the 1980 presidential election. This State-by-State Political Economy Model (2SPE Model) produces vote share forecasts for the two major party candidates in every state that can then be used to make a projection of the Electoral College result.
THE MODEL
The 2SPE Model acknowledges that the outcome of US presidential elections is decided by the Electoral College in which the balance of power is determined by the results in locked states, battleground states, and swing states. Such an approach makes it possible to grasp the dynamics of change in the states caused by, among other factors, variations in the economy and residential mobility (see Tiebout’s [Reference Tiebout1956] analysis of “voting with one’s feet”).
The 2SPE Model acknowledges that the outcome of US presidential elections is decided by the Electoral College in which the balance of power is determined by the results in locked states, battleground states, and swing states.
Our model correctly predicted the presidential elections from 2004 to 2012 (see Jérôme and Jérôme-Speziari Reference Jérôme and Jérôme-Speziari2012b). In 2016, however, the model failed to foresee Donald Trump’s Electoral College victory, especially in the key states of the Rust Belt: Michigan, Ohio, and Wisconsin (see Jérôme and Jérôme-Speziari Reference Jérôme and Jérôme-Speziari2016). This led us to make some adjustments to the model. We kept the initial blocks of explanatory variables: the variation in unemployment rate over the term of office (using a longer lag, however, in the revised model), the president’s job approval rating depending on whether the incumbent is seeking a second term or not, the indices measuring Democratic or Republican partisan domination in the older and newer strongholds of each party, the potential “nuisance” power of independent and third-party candidates, the score of the challenger party’s candidate in the primaries, and dummy variables for scores that are traditionally out of the norm, as in the District of Columbia. Then we updated the model by using presidential popularity measured by state. This represents a considerable improvement, because it enables a more accurate assessment of the local impact of a decline or an increase in the president’s approval rating. We also added a new variable looking at the partisan composition of state legislatures. We hypothesize that the incumbent party candidate stands in “friendlier” territory in states where his or her party controls the legislature. Additionally, the revised model uses the two-party vote share (instead of the raw vote share) of the incumbent party candidate as its dependent variable.
Jackknife out-of-sample forecasts for previous elections (1980–2016) indicate that the model is quite accurate when predicting who will win in each state and who will be elected president. The model correctly identifies the state winner in 82% of cases (420/510) and the Electoral College winner in 9 out of 10 elections, the sole exception being the 2000 election. For the 2016 election, it predicts Donald Trump to win with 272 electoral votes (but to nonetheless lose the popular vote).Footnote 2 This is a considerable improvement over the previous specification of the model, which gave Hillary Clinton the upper hand with 319 electoral votes.
The model correctly identifies the state winner in 82% of cases (420/510) and the Electoral College winner in 9 out of 10 elections, the sole exception being the 2000 election.
The revised model can be found in table 1.Footnote 3 As can be seen, a one-point increase in the unemployment rate of a given state in the second quarter of the election year compared to the last quarter of the previous election year leads to a loss of 0.38 percentage points in that state’s incumbent two-party vote share. In states where the legislature is controlled by the party of the president, the incumbent party candidate gets a 2.30 percentage point boost. An approval rating of 50% in a state leads to an average gain of 15.69 points in that same state when the president stands for reelection and of 11.83 points in an open race. Partisanship also matters: “old” and “new” strongholds of the incumbent candidate’s party offer bonuses of 7.36 and 4.57 points, respectively. Finally, various local electoral peculiarities and circumstances (like the presence of an independent or a third-party candidate) also leave their imprint on the incumbent’s vote.
State-by-State Political Economy Model: Pooled Time Series, 50 States and District of Columbia, 1980–2016
a Including new southern Republican strongholds.
Note: Dependent variable: Incumbent party candidate’s two-party vote share (2P-INCV). Significance levels: * p < 0.05; ** p < 0.01; *** p < 0.001 (two-tailed).
THE FORECAST
Using the model in table 1, we plugged in current numbers to estimate the incumbent party candidate’s share of the two-party vote in each state; from this, we can predict who will win the EV in each state. Adding up the EV generates our 2020 forecast, which suggests that Donald Trump will be defeated by Joe Biden next November. The model gives 308 EV for the Democratic candidate against 230 EV for the incumbent president (Jérôme Reference Jérôme2020). Using electoral weights based on voter registration (see section N of the appendix in the online supplementary materials), we converted vote share forecasts at the state level into a national vote forecast. Biden is projected to win 51.69% of the national two-party vote.
Figure 1 shows the predicted vote share for the Democratic and Republican candidates, as well as the predicted global outcomes (i.e., who should carry the state) in each state for the upcoming 2020 presidential election.Footnote 4 These predicted outcomes can be compared with the actual results of the 2016 election.
Predicted 2020 Election Results by State
CONCLUSION
Until the beginning of 2020, Donald Trump could argue in favor of his reelection by pointing to the state of the economy. The nation’s low unemployment rate (around 3.5%) compensated for his rather low approval rating. However, the deterioration of the economy caused by the COVID-19 outbreak has considerably harmed Trump’s economic record. The nationwide unemployment rate skyrocketed to 14.7% in April 2020. The COVID-19 pandemic has without a doubt damaged the economy, but chaotic management of the health crisis likely increased its economic impact. In several states won by Trump in 2016, the variation in unemployment is much higher than the average variation over his time in office. In a context where the partisan balance remains relatively stable in each state compared to 2016, the economy had earlier allowed Trump to compensate for his low popularity, giving him a fair shot at reelection. According to the 2SPE Model, this is no longer the case.
Joe Biden therefore seems to have victory within his grasp. However, it will not be an easy one. If one calculates the probability of crossing the 50% mark according to the margin of error of the model (i.e., ±5.02 percentage points),Footnote 5 the probability of a Biden victory falls below a 75% percent “comfort zone” threshold in six states—Colorado, Florida, Michigan, New Mexico, Pennsylvania, and Virginia—representing 92 EV. Joe Biden will have to be mindful of the misfortune of Hillary Clinton, who campaigned little in states presumed to be “toss-ups.” The 2SPE Model shows that, despite promising polling numbers, the Democratic candidate will have to campaign to the end to ensure his victory.
Joe Biden will have to be mindful of the misfortune of Hillary Clinton, who campaigned little in states presumed to be “toss-ups.” The 2SPE Model shows that, despite promising polling numbers, the Democratic candidate will have to campaign to the end to ensure his victory. Joe Biden will have to be mindful of the misfortune of Hillary Clinton, who campaigned little in states presumed to be “toss-ups.” The 2SPE Model shows that, despite promising polling numbers, the Democratic candidate will have to campaign to the end to ensure his victory.
DATA AVAILABILITY STATEMENT
Replication materials are available on Dataverse at https://doi.org/10.7910/DVN/QUPBSR.
SUPPLEMENTARY MATERIALS
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S1049096520001377.
