Recall in a recent post, Election Polls and Prediction, I observed that presidential approval ratings and the state of the economy are ‘fundamentals’ that portend election outcomes. Contrary to predictions based on available polling data, forecasters that utilize the fundamentals publish their expectations well in advance of the election.
Let’s consider 2020 predictions from 2 notable fundamental models.
Typically, when an incumbent president runs for re-election, major factors that determine the likelihood of victory include performance of the economy and presidential popularity. The vote is therefore expressed as a function of the two measures: Presidential popularity = Gallup presidential approval rating mid-year in July. Economic growth = change in Gross National Product over the first two quarters of the election year.
Beginning in 1948, the political economy model has correctly predicted 15/18 presidential contests – 83%. The model failed in 1960, 1968 and 1976 – two very tight contests and the first campaign after Nixon’s resignation. However, the model correctly predicted the last 10 elections – with a very small forecasting error.
This year, the model projects a comfortable Joe Biden win.
Tweaking the model for the Senate races, it also projects a 12-seat gain by Democrats – flipping majority control from Republicans to Democrats.
The time for change model also accounts for the state of the economy and presidential popularity but adds a third variable – incumbency. The addition recognizes that first-term incumbents enjoy significant advantages, and very few fail to win re-election. The last president to lose re-election was George H.W. Bush in 1992. Before Bush, it was Carter in 1980 and then all the way back to Hoover in 1932.
Noteworthy, the model’s author, Alan Abramowitz, acknowledges the unique circumstances this year. Specifically, he argues that two predictors – the change in GDP and incumbency would not perform as usual and should be removed from the model.
First, while GDP plunged dramatically in the second quarter, voters do not appear to hold Trump responsible. To control the spread of the virus, much of the economy was deliberately shutdown. Despite the unprecedented rise in unemployment and drop in GDP, Trump’s approval ratings on handling the economy remain positive.
This fact does not square with theory. During grim economic circumstances, voters are expected to punish the incumbent. Yet today, a majority of Americans give Trump positive ratings on the economy.
Second, historically high partisan polarization diminishes the electoral advantage of first-term incumbency. Part of the advantage of incumbency is the ability to appeal to voters across party lines that are often reluctant to replace a president after only one term. Trump’s penchant to appeal to his own party base makes this unlikely. Indeed Trump established a new record for partisan polarization – 82 point gap in approval rating between Republicans and Democrats.
Abramowitz thus casts aside two main variables and settled on a single predictor for the 2020 forecast. He calls it a ‘simplified incumbent accountability model’. Trump’s net approval rating in late June represents that single predictor.
Gallup June’s approval rating was 40%, with 55% disapproving for a net of – 15. The model generated a 29.5% chance of a Trump victory.
Abramowitz transforms his model from 3 predictor variables to 1. And that change appears inconsistent with the fundamentals philosophy. The theories behind the fundamental measures should apply across every election context. That’s why the measures are called fundamental. If scholars can pick and choose measures based on specific electoral conditions, then let’s not pretend the model generalizes across elections.
Nevertheless, Abramowitz breaks from tradition and focuses on the particulars. He argues that in 2020, late October presidential approval ratings are the best means of predicting the winner. He then offers several predictions based on potential net approval scores.
For example, if Trump’s October net approval rating remained at the June reading of -15, his chances of victory slide to a meager 9%. However, an October net approval score of -5 improves Trump’s chances to 31%.
|October net approval||Chance of Trump victory|
On September 28, Gallup reported Trump’s net approval at – 6 (46% approval). The margin of error associated with the estimate is plus/minus 4 points, which means approval could be as high as 50% or as low as 42%.
For the moment, let’s assume the approval estimate is on the high side at 50%. When approval and disapproval are even – net zero, Trump’s chance of victory edges close to 50%. To the extent that Abramowitz is right, a positive net approval rating in October pushes the odds of a Trump victory over 50%.
Thus, down the stretch, keep an eye on Trump’s public approval ratings.
The two fundamental models view the election as a referendum on how well the president handled economic and non-economic issues. The better the performance, the greater chances that the president wins reelection. Trump did not perform well. Both models foresee a comfortable Biden victory. And the political economy model forecasted a Senate majority for Democrats.
Yet the models departed on the unusual circumstances of 2020.
The political economy model remained faithful to the fundamental approach. The authors recognized the devastating changes brought about by COVID-19, including a dramatic economic collapse, and the widespread protests inspired by George Floyd’s killing. They argue the consequences of these events are already baked-in the measures and expressed through the model’s predictions. Change in GNP captures the economic collapse and the political turmoil manifests in Trump’s July job approval rating.
Meanwhile, circumstances prompted Abramowitz’s time for change model to abandon economic performance and political incumbency. He projected a winner based on a single measure – presidential popularity and then offered an additional prediction based on hypothetical popularity ratings in late October.
Finally, consider the basic lesson of fundamentals: They can predict election outcomes before the campaign starts. This fact demands we reconsider the relevance of the day-to-day horserace analyses and the significance of the campaign itself.
The news media spend much energy and untold hours tracking candidate speeches, events, public appearances, and private behaviors – connecting nearly every action to electoral fortunes. Detailed candidate characteristics and every word spoken seem to matter.
Yet, it’s all trivial compared to the president’s handling of the economy and the pandemic. This is what the fundamental models tell us. They are as accurate – often more so – than models that consider the campaign as the causal agent of change – pure polling models.
Sure, strategy and events move polls and media reports contribute to our understanding of elections. But none of this is foundational to predicting the winner. Rather, the fundamental models capture the essentials. And the essentials determine the outcome before we begin democracy’s greatest spectacle.
Pundits, media personalities, and campaign consultants smirk at such assertions. After all, they occupy a pivotal role and they believe in their capacity to persuade and generate significant change.
Not a single advocate of the fundamentals approach denies that people can produce change and that campaigns are important. Advocates make clear though that larger economic and political structures cause election outcomes. Pundits, media personalities, consultants and candidates work within those constraints.
 For Senate races, the model uses real disposable income over the first two quarters of the election year and adds a midterm variable to recognize when contests are held during presidential election years. Finally, it adds a measure for the number of seats the president’s party has up for reelection.