Forecasting elections is fraught with peril. This year is more difficult or perilous than most recent elections. There were presumptive nominees for each party, but then the incumbent Democrat was cast aside. We were in the situation when, for the first time since 1956, both parties looked as though they would nominate the same candidates as last time. Moreover, it was the first time since 1892 that we would have had an incumbent president running against a former president. There was an attempt on the life of the former president the weekend before his renomination. Clearly, this election season is atypical. The oddities of this year should encourage us to be humble in our forecasts.
Despite the rather unusual nature of this election season, we can still forecast the election. Unlike pundits, we will be making forecasts that are subject to an external reviewer assessing the process by which we made our forecasts. It will not be a seat-of-the-pants judgment. Outside observers can examine our models. If we revise our models, it is transparent.
If we revise our forecasts, there will be a record of what we have done. In a sense, we are a bit like weather forecasters. We try to offer a long-range forecast that makes it possible to plan for the future. The earlier the forecast, the greater the opportunity for one to make use of that information. We also want to take new information into account as it becomes available. Both kinds of forecasts have their utility. We want to be able to plan for the future. Nonetheless, we also want to be able to take advantage of new information as it becomes available. Here, the effort is on making the long-range forecast.
For the forecast of the 2024 election, presidential and the House elections will be using the largely the same model as used in 2020. Here, the focus is on the long-range forecast of the election. The forecast makes use of data that were available as of July 5. The long-range forecast has the advantage of allowing more time to adjust behavior before the event. Someone who forecasts the weather well in advance allows people more time to adjust their behavior than a forecaster who gives you a two-minute warning of a storm approaching. People who pay attention to long-range political forecasts can change their behavior considering the information presented by political forecasters.
INFLUENCES ON PRESIDENTIAL ELECTION OUTCOMES
There are, no doubt, many potential variables we could include in the model. In the interest of parsimony and recognizing the small number of cases, we need to be selective. The literature on elections is replete with economic models. Because there are a small number of cases, we cannot include multiple variables representing economic conditions. One of the ongoing debates is whether voters look to the past or the future when casting a ballot. In one sense the decision is easy. The retrospective and prospective economic items are strongly correlated (.84).
Recognizing there are still debates on this topic, I follow the lead of Fiorina (Reference Fiorina1981), Lewis-Beck (Reference Lewis-Beck1988), Lockerbie (Reference Lockerbie2008, Reference Lockerbie2020), and Nadeau, Lewis-Beck, and Bélanger (Reference Nadeau, Lewis-Beck and Bélanger2013), whose models focus on the prospective. I and those who find themselves in disagreement with me can take some comfort in the strong relationship between the retrospective and prospective economic items at the aggregate level. As with earlier endeavors, I make use of the item from table 8 of the Survey of Consumer Attitudes and Behavior. Instead of averaging the scores from the second quarter, as done in the past, I make use of the percentage replying in the negative from June of the election year.Footnote 1 This is because the volatility of the item and the desire to have the latest information consistent with an early forecast. The specific question is “Now looking ahead—do you think you (and your family living there) will be better off or worse off financially a year from now, or about the same?”Footnote 2
I should note here that this measure includes no sense of attribution to either party. One can think the economy will be worse off in the future and still vote for the incumbent party because one believes that the incumbent party will be better than the opposition. The retrospective item has one advantage. Even though it does not reference the incumbent party, there is no ambiguity about which party controls the White House during this time.
Aside from economics, there are other factors that are relevant for forecasting presidential elections. Specifically, Abramowitz (Reference Abramowitz, Campbell and Garand2000) has shown that the incumbent party’s share of the two-party vote is negatively related to how long it has controlled the White House. Specifically, having held the White House for two terms puts one in a worse position than does holding it for just one term. Nonetheless, we should note that in the last two elections, the incumbent party has lost after holding for just one term. Before that in 2012, the incumbent won with a lower percentage of the vote than was won in the original election. To account for the penalty for having held the White House for a long time, I make use of the log of the time the incumbent party has controlled the presidency.Footnote 3
In the interest of humility and transparency, this forecast model was off by quite a bit in 2020. Donald Trump was forecast to be the victor by a comfortable margin. The actual result was a comfortable victory for Joe Biden. There are several potential explanations for this. The unprecedented nature of the pandemic may have heightened anxiety about the overall state of the economy. Even with the modest pessimism about the economy in 2020, the pandemic may have given many people pause about voting to return Trump to the White House.
Similarly, the protests following the death of George Floyd and the rise of the Black Lives Matter movement might have made people more uneasy about the future. The noneconomic anxiety might have also made people less inclined to give Trump another four years. The 2020 election underscored the limitations of these models in the face of unprecedented national and international events. I still have confidence that the prospective model will perform well.
Consequently, I do not think there needs to be a fundamental change in the model. Even with the error of 2020, the model performs reasonably well over time. Nonetheless, forecasters and those who use forecasts should be aware of the role of events outside the model in influencing the accuracy of the forecast. Similarly, we ought not to be too quick to jump to current events to revise our models or explain mistaken forecasts. If we do so, we run the risk of engaging in ad hoc speculation that cannot be tested.
PRESIDENTIAL ELECTION FORECAST
Table 1 shows the results from the equations forecasting elections. The results of this effort are similar to those of earlier years. The more pessimistic people are, the more likely the incumbent party is to lose. As before, the longer a party has controlled the White House, the lower its share of the vote. The forecast for 2024 is that the incumbent party will receive 49.09% of the two-party vote. This and the popular press discussions of the election suggest a very close election.
Forecasting Equations 1954–2024
Note: Significance levels in parentheses.
We can look at the out-of-sample equations to assess the utility of this model. To assess the model’s performance, we take one case out of the data set, reestimate the equation, and use the results to forecast the excluded election. We then repeat this process for each case. The results are shown in table 2. I have put the election years forecast correctly in bold. The average absolute error is 3.3 percentage points. Even with all the vagaries of presidential elections, the model does an excellent job of forecasting presidential elections.
Out-of-Sample Presidential Forecasts and Errors
Influences on House Election Outcomes
House elections are a bit different from presidential elections. The House elections are more likely to be influenced by the incumbency. We know that if an incumbent seeks reelection, that incumbent is likely to be successful. We are confident in predicting that the overwhelming majority of House incumbents who seek reelection will win reelection (Alford and Hibbing Reference Alford and Hibbing1981; Collie Reference Collie1981; Ferejohn Reference Ferejohn1977). Open seats present the best opportunity for a party to pick off seats from the other party. To account for this, I include an interaction term between the number of open seats and whether it will be a good or bad year for the incumbent party. Midterms are, by definition, a bad year for the president’s party. On-year elections are a little more complicated. If the vast majority of the electorate believes one party is likely to be victorious, we should expect it to pick up more seats, if there are a large number of open seats. The number of open seats is multiplied by -1 if it looks like a bad year for the incumbent president’s party, by +1 if it looks like a good year for the incumbent president’s party, and by 0 if it is neither a good nor a bad year for the incumbent president’s party.Footnote 4
Looking at table 1, we can see that this equation does a reasonably good job of forecasting House elections. The more pessimistic people are, the more likely the incumbent party loses seats. The more open seats in a good (bad) year there are, the more seats the incumbent party is likely to gain (lose). The forecast for 2024 is that the incumbent party will lose 12 seats. We can, as before, look at the out-of-sample equations to assess the utility of this model. The results are shown in table 3. The average absolute error is 16.8 seats.
Out-of-Sample House Forecasts and Errors
CONCLUSION
When we look at the forecast for the 2024 election, it is essentially a jump ball. The Democratic party is likely to lose the presidency, and the House is likely to remain in the hands of the Republican party. If the forecasts of a Democrat losing the White House and not regaining control of the House of Representatives are correct, it will mean a return to unified government, at least for the presidency and the House of Representatives.Footnote 5
ACKNOWLEDGMENTS
I would like to thank the reviewers for questions and suggestions that made this brief manuscript better. I would also like to thank Samuel Knell for his questions and comments on earlier iterations of this project.
DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available at the PS: Political Science and Politics Harvard Dataverse at DOI: https://doi.org/10.7910/DVN/FMRDPF.
CONFLICTS OF INTEREST
The author declares that there are no ethical issues or conflicts of interest in this research.
Open access
