Why political science falls short in predicting elections

My latest blog originally appeared in The Hill on October 5, 2016.

October 05, 2016, 12:55 pm
Why political science falls short in predicting elections
By David Schultz, contributor


Candidates, campaigns and strategies don’t seem to matter in U.S. presidential elections. At least that is the impression often left by mainstream political scientists and their election forecast models. Yet the 2016 presidential election suggests that who the candidates are, what they say, and how they campaign does matter.

A hallmark of modern political science is the scientific aspect of it. It is an effort to create mathematically objective and accurate models that can predict political behavior and events such as elections. Define a list of variables, plug them into an equation, and presto: months before the actual election we can predict a winner.

The leading political science prediction models consider variables that include economic growth, (incumbent) presidential popularity, and how long the current party has been in power. Oddly, many of the best prediction models downplay or ignore political variables such as the impact of presidential debates, the quality of candidates, the role of money in politics and even policy positions — the factors that candidates, their managers, and strategists often discuss and emphasize as critical to electoral success.

The best models for political scientists seem to be the ones that are deployed far in advance of elections, so as not to be tainted by actual politics or the campaigns themselves. Even less statistically driven models seem to assume elections are over before they start. I too am guilty of that, and have argued that because of the Electoral College and partisan alignments, the 2016 election was largely over two years ago except in a handful of 10-12 swing states such as Ohio, or that only about 15 counties will matter this election cycle. Political science models describe how firm partisan and demographic factors such as race and gender are in determining whether and how individuals will vote.

Listen to political scientists and one will think that debates, who the vice presidential candidate is, and even lawn signs do not matter. It seems nothing — especially anything any of us would consider political — really matters. Politics and elections seem to be on autopilot. Theodore White’s classic “The Making of the President” series, while informative, is superfluous in terms of explaining presidential elections. We might as well simply run the models every four years and let them and political scientists pick the next president.

The reality is that politics does matter in campaigns and elections. The very craft of political science in its effort to be scientific often misses that. It misses the subtle shifts in public opinion that occur as a result of a debate. It ignores, especially this year, how candidate character — or lack thereof, as measured by the high disapprovals for both Hillary Clinton and Donald Trump — might affect voting behavior.

The models miss how anti-establishment feelings, whether a voter feels like they could have a cup of coffee with the candidate, potential terrorist attacks, fear of immigration, emails, and a potential massive WikiLeaks dump could shape elections. And these models overlook the impact that get-out-the-vote campaigns, voter ID laws and racism or sexism have on elections.

Perhaps these affects are variables too subtle to measure, or instead they are structural forces that simply escape standard political science models because, oftentimes, these are unpredictable or escape easy quantifying. These are the factors that political operatives consider important — this is real politics.

Real politics is sloppy and dominated by conventional wisdom that it often wrong. It is rumor-fed and hunch- or gut-driven. It’s not scientific.

In real politics voters matter, and they need to show up for candidates to win. Yes, real politics could learn a lot from political science. But political science could learn even more about real politics if it paid more attention to what is happening in politics. Too many of my political scientist colleagues are perplexed this year by the success of Trump, the struggles of Clinton, and why this election is so close. But to really understand and explain the 2016 election cycle, one needs to move beyond macro-statistical models that fail to appreciate how campaigns and elections are affected by micro, subtle and structure forces that take place within an election cycle and are not simply determined well before campaigns even start.