“Donald Trump’s presidency could be a real possibility,” lamented Hillary Clinton’s campaign manager in a fundraising email send out after Clinton’s poll numbers started sliding. News outlets are incessantly reporting that the latest polls reflect a tightening in the race between Trump and Clinton and that one recent CNN/ORC poll showed Trump had taken the lead. Nate Silver, who became famous for applying sabermetrics to politics, today announced that he considers the presidential race “highly competitive.”
In contrast, we hear some in the media reassuring us that now is not the time to panic. Statisticians add their calm, steady voices to the small chorus of writers—such as Jeet Heer at the New Republic and Greg Sargent at the Washington Post—who are trying to soothe us. The scholars behind the Princeton Election Consortium still have the odds at better than 4:1 in favor of Clinton.
Are these writers and scholars who try to pacify the nervous public right? Should we instead listen to the over-simplified headlines, the self-interested campaign operatives, and the morning show pundits?
Did my naming-calling give away my answer?
We should listen to the measured, reasoned arguments of those listed in the former group. Not because they claim some authority with their sophisticated equations—though that is a fine reason. Not even because they average polls together in order to give a big picture view of the race—this is a good reason, too. The best reason to listen to them is that history proves they are right.
Robert Erikson and Christopher Wlezien’s 2012 tome, The Timeline of Presidential Elections, goes a long way toward showing that polls are useful predictors of elections, but not in the way the headlines would have you believe. Erikson and Wlezien use polling data from 15 presidential races dating back to 1952 to show that, by the fall of the general election, the aggregate poll results are extremely good predictors of the election results. Depending on how the polls are averaged together—different organizations have different formulas for which polls are included and how they are weighted—Clinton had between a three and five point lead at the end of last week. This indicates that Clinton will win the election, even if by a small margin.
Erikson and Wlezien also convincingly show that, once the so-called bounce after the Party Conventions settles back down, poll results explain over 80 percent of the variance in the vote. In other words, the average polling results for late August and early September are highly predictive of the real election outcome. This leads them to the conclusion that relatively little changes after the conventions, and that the candidate with the net gain at that point will often lead for the rest of the campaign. After the Republican Convention, Trump easily closed the polling gap of three to four points, tying or even beating Clinton in the various polling averages. After the Democratic Convention, Clinton regained her lead, pulling approximately six points ahead of Trump. Clinton’s convention bounce was substantially larger than Trump’s. Erikson and Wlezien would, therefore, predict that Clinton will maintain some of that lead, though certainly not all of it, for the remainder of the campaign.
This lead will shrink as the election gets closer. In fact, history shows that the gap between Trump and Clinton’s poll numbers will come close to converging on the 50-50 line. That means that a handful of individual polls may show Trump ahead of Clinton, but by averaging them we will see that while they move towards convergence, Clinton will maintain a small lead that will carry her to victory on Election Day.
This isn’t to say that things can’t change. Pollsters and pundits thought Dewey would defeat Truman; Nixon might have beaten Kennedy had it not been for the televised debates; and if Hurricane Sandy hadn’t pummeled the Eastern Seaboard, Obama could have lost to Romney (he would certainly have won by a smaller margin.) In other words, the numbers can be wrong, presidential debates occasionally make a difference, and exogenous shocks might be enough to turn the tide. Unless something completely unforeseen upsets the electoral system before November 8th, history and statistical modeling tell us that we will be saying “Madame President” on November 9th.
P.S. I’m taking two courses on writing short form political commentary. Between that and the election, expect more of the “politics” side of the blog over the next months.