Political science has its limits when it comes to presidential prediction

“Who’s going to be the Democratic presidential candidate?” “Is Trump going to win again?” “What impact will the impeachment, Senate trial, and possible war with Iran have on the 2020 election?” As a political scientist who specializes in American politics and election law, these are the questions I am already bombarded with going into the next election. I am not alone.  

Political scientists often are called upon to predict who will win elections. Soon, leading scholars in political science will release their 2020 prediction models, often well ahead of Iowa caucuses and the start of primaries or general elections. These models say a lot about what political scientists think about American politics — and also what political science is and examines.

The study of politics goes back to the Greek philosophers Plato and Aristotle. Yet modern political science is a construct of the post-World War II era, a response to the challenges of modern science that was able to build a nuclear bomb, design a computer and land a man on the moon. Looking at these accomplishments, some sought to make the study of government and politics a science. A scientific study of politics meant building models that could mathematically predict events, much like physicists or chemists can calculate or predict molecular or chemical reactions. Prediction included the concepts of description and explanation. This is what the “science” in political science meant.

Among the holy grails of scientific political science is predicting U.S. presidential elections.  Over the past half-century, political scientists have designed and perfected presidential forecasting models. In 2016, eight forecasting teams published their predictions in PS: Political Science & Politics. Nearly all of them came extremely close in predicting the final vote share for the Democratic and Republican parties, yet generally failed to capture correctly the electoral vote division that produced President Trump’s victory.  

Look to see similar — and potentially highly accurate — prediction this year. Yet, again, they too may fail to foresee who eventually wins the presidential election and why. This failure and the assumptions built into these predictive models are fascinating.

First, the models are aimed at predicting two-party aggregate popular vote totals. While predicting this is fine, they do not tell us who will win the presidency. These models are unable to account for the reality of presidential elections that are 50 separate state elections plus the District of Columbia, where the race is to get to 270 electoral votes. These predictive models make the same mistake as national polls — looking at the aggregate results of something that is not necessarily determinative of the outcome of national elections.  

Second, they assume a two-party model with a first-past-the-post (winner-take all) single member district model of elections. Were states to turn to ranked choice voting, as Maine and other jurisdictions are doing in non-presidential elections, or to adopt other voting systems, these models might not work as well. This is because third parties would alter popular vote totals and change voter calculations these predictive models presently cannot handle.

Third, consider what most of these models use as variables to predict popular vote totals. Many look at some combination of job approval for the president and the economic factors such as GDP growth several months before the election. In the case of economic growth, these models seem to concede to economists the saliency of their variables as drivers of political behavior.  Conversely, these models continue to assume that “It’s the economy, stupid,” as James Carville said in the 1992 presidential election when he urged Bill Clinton to talk about only that when running. But, increasingly, one finds that voters are not rational economic calculators and are not moved by economic variables. Partisanship, values, identity and ideology seem more potent forces motivating choices.

Fourth, think about what these predictive models imply. As currently constructed, they say nothing about who the candidates are, their messages, or campaign strategies. In effect, politics or campaigns and elections are unimportant when it comes to elections. They also discount contingent events, such as a war with Iran or the taking of American hostages by that country back in 1979, which derailed Jimmy Carter’s re-election. The reality is that these variables do matter, and we saw that in 2016 when Hilary Clinton, who won the popular vote, failed to win the presidency in part because of who she was, her messaging, or strategy that ignored the Midwest.

The field of political science, or at least these models, is good at predicting something that may not matter when it comes to presidential elections. Prediction here also fails to describe what really is happening in a presidential election, and therefore is also unable to explain what happened and why.

My field of political science is smart in much of what it does, and in building scientific models.  Yet presidential forecasting speaks to the limits of the field in describing and explaining “real” politics — and we should be wary to bank too much on these models in this presidential election.

David Schultz is a professor of political science at Hamline University in St. Paul, Minn. Follow him on Twitter @ProfDSchultz.