Downside #1: Partisan Polarization Perverts the Fundamentals

If there is a conventional wisdom that has developed around forecasting the general election winner in presidential elections, in my view, it is the “Time for Change” model developed by Alan Abramowitz and its variants. That is, three fundamental features of a presidential election have correctly forecast every presidential election since 1988: party incumbency, status of the national economy, and presidential approval. These are the fundamental features of an electoral context that I teach undergraduate students. This symposium in PS includes many other valuable models, with a high degree of consistency among them.

In general, models rely on two categories of inputs: static features of the election cycle (e.g., “the fundamentals”), such as how long the incumbent party has been in the presidency, and variable features of the election cycle, such as incumbent approval ratings and economic statistics. When the predictors in the variable category have low variance, however, they operate more like fundamental features. In this way, most forecasters agree that day-to-day campaign events and news cycle highlights are less predictive of the outcome than are fundamentals. But this is where 2020 gets interesting and presents particular challenges to forecasting.

Unfortunately, two particularly salient features of the 2020 election season make forecasts both more appealing and less reliable: polarization and the pandemic. First, as politics has become nationalized (Hopkins Reference Hopkins2018) and partisanship is thought of as teamsmanship (Hetherington and Weiler Reference Hetherington and Weiler2018; Mason Reference Mason2018), predicting the outcome of a high-stakes election in a polarized political environment becomes of increasing interest to more people. Although a polarized electorate may have greater interest in forecasts, the polarization reduces the reliability of the measures in their models. For example, if we think of fundamentals like incumbent approval ratings or status of the economy, it turns out that Republicans and Democrats have very different evaluations of these factors (Pew Research Center Reference Center2020). Measuring the overall approval rating may provide an eagle’s-eye view of an incumbent’s popularity, but if partisans view the incumbent and their likelihood of supporting the incumbent party systematically differently, then it introduces a systematic bias for which the model does not account. Polarization therefore drives up interest in election forecasts and drives down their reliability.

Moreover, as a political scientist, I am uneasy about publicly recognizing the predictive nature of the state of the economy for understanding elections, because it tacitly suggests that voters justifiably assign credit or blame to presidents for economic conditions. Political science research shows that this sort of retrospective evaluation of candidates is a flawed means of holding elected officials accountable, largely because generic human cognitive biases (such as recency bias) cause us to overvalue recent or salient events over more consequential events (Achen and Bartels Reference Achen and Bartels2016). This suggests that the state of the economy is sometimes an irrational reason to credit or blame the president, but nevertheless, humans are flawed, and the factor remains predictive. A political scientist understands the difference between a predictive factor and a normative claim, but we should not expect everyone to.

In addition, the COVID-19 pandemic also drives up interest, and down reliability, of election forecasts. The pandemic has curtailed or eliminated professional and collegiate sports, and Americans may be extra hungry for a team to root for, a race to watch, or highlight reels to relive. The now immense and highly visible election forecasting industry fills that demand. Yet, even though demand may be up for election coverage, the pandemic itself messes with fundamental indicators included in forecasts. Statistics that capture the economic downturn, for example, may reflect the pandemic more than the president and therefore provide a tainted indicator for an election forecast. In this way, coronavirus infects the forecasts with measurement error, while stoking the public’s interest in forecasts’ findings. Not a great combination.

Finally, President Trump is an anomaly that creates challenges for forecasters using out-of-sample predictions. As a candidate in 2016 he was heavily discounted because of his lack of any experience in elected office and his scorched-earth campaign style that broke many norms of strategy and behavior in a national campaign. As the incumbent in 2020, Trump is less of an anomaly now than in 2016, because he is, well, the incumbent, but he is still an unconventional candidate in many respects. He defies democratic norms in nearly every public appearance by disrespecting political opponents, glorifying authoritarian practices, revering autocratic leaders, and undermining democratic institutions. His campaign style in 2020 so far is fairly similar to the antagonistic campaign he ran in 2016 that emphasized themes of misogyny and racism. For the first time in history, Americans have the opportunity to reelect an impeached president. Talk about your out-of-sample events.

In a world where the fundamental indicators are diseased, are they valid predictors of the election? If not, how can forecasters adjust their models without being arbitrary and introducing personal bias? The whole idea of fundamentals is that structural commonalities across time explain political outcomes that are governed by the Constitution and by fixed political contexts. If we discount the fundamentals because of anomalies in the cycle, then we are effectively admitting that fundamentals are a weak model. But that seems like an overstatement of reality.

At the end of the day, forecasters who rely on variable factors of the election may overcorrect their models in arbitrary ways that reduce the value of their forecast; yet those who rely on fixed factors may insufficiently account for the odd features of this particular cycle. Damned if you do; damned if you don’t.

Downside #2: Forecasts May Affect Turnout

A study of the 2016 election showed that greater emphasis on forecasts in that cycle increased voters’ certainty about the election outcome and therefore depressed turnout (Westwood, Messing, and Lelkes Reference Westwood, Messing and Lelkes2020). Had 2016 seen fewer or paid less public attention to forecasts, the outcome of the election may have been different. In this way, forecasting itself is not unvirtuous; the business and aggrandized showiness of forecasting imposes costs on democratic institutions that may exceed their benefits to the general public, especially if an election is close.

Downside #3: Outsized Focus on the Election Horse Race

Although a polarized, partisan, and always-in-crisis political environment drives up interest in the election, forecasts contribute to a news environment that focuses on the horse race at the expense of more substantive election topics. To make matters worse, the pandemic likely increases this effect, as noted earlier. Political science research shows, however, that journalists rationally provide election horse race coverage, rather than covering candidate scandals or election issues, because the latter are simply more expensive to write about (Searles and Banda Reference Searles and Banda2019). In 2020, we should anticipate a louder, but lower-quality, election discourse. More forecasts contribute to that effect.

Downside #4: Forecasts Give a False Impression of Science and Certainty

When the public perceives that polling or forecasts are incorrect, it contributes to skepticism about science and media. When forecasts or reporting about forecasts overstates the level of confidence in an estimate or fail to communicate uncertainty associated with that estimate, it gives a false impression about what science is designed to do. Science is for explanation, not prediction. Increasingly, public distrust in science and media is contributing to democratic demise in the United States. I am not claiming that the forecasts are causing this, but how we talk about them matters.