COVID-19 may upset the 'Keys to the White House' prediction

Allan Lichtman, the distinguished political scientist from American University, has demonstrated an uncanny knack to forecast who will win U.S. presidential elections.

Using an earthquake forecasting model adapted to election forecasting, he uses 13 simple true/false questions, labelled the “Keys to the White House,” with six or more false keys indicating that the challenger will defeat the incumbent. Originally formulated to forecast the popular vote winner, Lichtman’s forecast modifications after the 2000 Bush-Gore election resulted in a correct forecast of Trump over Clinton in 2016.

For the 2020 election, Lichtman classifies seven of the keys to be false, indicating that will be elected the next president, and not just win the popular vote, which is certain to occur without the assistance of any models (think Biden’s dominance in California and New York).

Forecasting models, like most analytical models, use sets of input data that coalesce and transform into outputs. These outputs may provide a forecast of something that has not happened yet, or something that is imminent, but no one knows when. Earthquake forecasting models are useful, but like all models, the quality of the input determines the quality of the forecasts. As such, they may predict an earthquake that does not happen (a false alarm) or not predict an earthquake that does happen (a false clear).

Any input-driven model relies on historical information stealthily hidden in the data. Artificial intelligence and machine learning typically require massive amounts of data to be most effective. However, if the data is void of certain extremely rare events, or never observed events, the models may be unable to provide accurate forecasts of such phenomena.

How is this relevant to this year’s presidential election? Examining Lichtman’s 13 keys, four are clearly false (midterm House gains by the incumbent party, no social unrest, no scandals and no foreign or military successes), and six are clearly true (being the incumbent president, no primary challenger to the incumbent, no significant third party candidate, a major policy change, no foreign or military failure and the opposing party’s candidate is not charismatic).

The three remaining keys (positive short-term economic prospects, positive long-term economic prospects and incumbent is charismatic) become the deciding factors. To reach six false keys requires two of these last three keys to be false. Lichtman lists them all as false, with COVID-19 turning the two economic keys. However, there has never been a public health event of the size and scope as the current COVID-19 pandemic. As such, the data used to create the 13 keys and the true/false rules could not have interpreted the economic downturns driven by a public health crisis. Looking to early 2020, prior to the nation’s shut down, economic prospects both short term and long term were positive, which would have given Trump at most five false keys, not seven, an indicator for his reelection.

If the model splits the two economic keys, given that the stock market continues to remain strong, the deciding factor becomes whether the incumbent is charismatic. Lichtman declares that he is not. Others, particularly those in his base on the right, would argue otherwise. Given that elections are all about turn out, if Trump’s base is motivated to show up and vote, the Trump charisma key may turn true.

What this exercise demonstrates is that in spite of large leads in the polls, this election is closer than many wish to acknowledge. Lichtman’s forecast is likely correct. However, like forecasting earthquakes, sometimes false alarms occur and earthquakes just do not occur.

Sheldon H. Jacobson, PhD, is a professor of computer science at the University of Illinois at Urbana-Champaign.  He applies his expertise in data-driven risk-based assessment to evaluate and inform public policy. He is the founder of Election Analytics at the University of Illinois, a STEM learning laboratory for election forecasting.