There’s no more accurate way to predict the day’s weather than looking out the window that morning, and there’s no more accurate way to forecast an election than looking at the polls on election eve. There are several academic and nonacademic meteorologists — The Washington Post’s Election Lab, Five Thirty Eight, The Upshot, Princeton Election Consortium, Daily Kos, Pollster and more — providing daily election forecasts. These models currently rely on aggregated preelection polls, with a sprinkle of the fundamentals and a dash of state-specific factors.
Political science forecasting dates to at least 40 years ago, when Edward Tufte published his famous 1975 paper in the American Political Science Review. A Google scholar search on “political science election forecasts” returns over 50,000 results over 40 years and spanning elections worldwide.
For the past 12 years, the political science journal PS has published a symposium featuring midterm and presidential election forecasts by leading political scientists. This year’s symposium is being released Monday. It forecasts sunny days ahead for the GOP — a net gain between four and 16 seats in the House and five to eight seats in the Senate. Three of the five Senate forecasts would lead to unified Republican control of Congress.
With new polls coming out daily, why pay attention to these forecasts?
To continue the analogy, political scientists are the climate scientists of the meteorological world, not the short-term weather forecasters. The PS forecasts are not based on a simple aggregation of polling data. Alan Abramowitz’s model, for instance, captures shifting political sentiments nationwide without reference to a specific race. Joseph Bafumi, Robert Erikson and Christopher Wlezien quite consciously rely only on information collected well before the election. Their approaches parallel the difference between climate science and short-term weather forecasts.
Political scientists enjoy playing the election prediction game, just like journalists, but we draw a bright line between forecasting and understanding. A light-hearted example by Michael Alvarez and Brian Loynd published 20 years ago illustrates the difference. Alvarez and Loynd showed that an American League victory in the World Series contributes nearly 12 percentage points to the Republican Party’s vote total. (One can only speculate about the impact that interleague play has had on this important result!) But Alvarez and Loynd never thought that the World Series had any direct causal link to election outcomes, just like last year’s snowstorms in Washington, D.C., tell us anything about long-term climate change.
The value of the political science models is to assess how well we can predict the election, even significantly ahead of time, based on scientific theory. These models provide a baseline for how well a party “should” do that we can then use to understand the results. For example, the disjuncture between the 2010 forecasts and the results provided evidence for the impact of salient roll call votes on vote outcomes. James Campbell developed the now common metaphor of “wave” elections in response to underestimates of the 1994 Republican victory.
In short, the point of political science models is not just to tell us who will win, but to help us determine why they won.
Phillip Ardoin and Paul Gronke are co-editors of PS: Political Science and Politics. Ardoin is chair and professor of political science at Appalachian State University. Gronke is the Daniel B. German visiting professor at Appalachian State and director of the Early Voting Information Center.