Ludwig-Maximilians-Universität München

Language Selection

Breadcrumb Navigation


Presidential election

She’ll win, she won’t, she’ll win ...

München, 09/26/2016

The US elect a new President on November 8th. With the TV debates coming up, the polls are highly volatile, suggesting a neck-and-neck race between Clinton and Trump. But according to Andreas Graefe’s model, the outcome is already clear.

Source: Picture Alliance / newscom

Why are polls for the American presidential election so volatile? For several weeks the headlines have varied between “Clinton extends lead” and “Trump narrows gap”.
Andreas Graefe:
That’s no surprise. Polling results have always varied substantially in the run-up to US presidential elections. In 2012 Mitt Romney was in front for a short time, in 2008 John McCain led the polls for a while, and the same holds for John Kerry in 2004. However, changes in polling numbers, particularly months before Election Day, do not necessarily reflect changes in people’s vote intention. Recent research shows that the variation in the polling results does not mean that voters are constantly changing their minds. What does vary is their willingness to participate in polls. For instance, in July, after the party conventions in Cleveland and Philadelphia, Hillary Clinton had a commanding lead in the polls. But this was also in the period immediately after Donald Trump had verbally attacked the Khans, the parents of a Muslim member of the US Army who had been killed in combat in Iraq, which might have set off some of his supporters. The more enthusiastic voters are, the more likely they are to respond to pollsters. And Trump supporters are clearly far more excited about their candidate. In the aftermath of Trump‘s attack on the Khan family, however, those who back him were less willing to participate in polls.

In other words, asking voters for whom they would vote if the election were held today is likely to be misleading?
Graefe: One problem with this approach is that it measures factors that are not directly related to voters’ intentions. Another problem is the weighting method pollsters use to derive the final numbers. Last week, journalists at the New York Times carried out a very revealing experiment: they sent the raw data of a single poll to four renowned pollsters and asked them to apply their weighting method. The results ranged from a 1-point lead for Trump to a 4-point lead for Clinton. A five-point difference, based on the same data! The good news is that the quality media have begun to turn away from reacting to every new poll with banner headlines, and are now more likely to focus on polling averages.

Is it not better then to simply ignore polling results?
Graefe: Polls do provide an indication. In recent elections, the candidate who was in the lead after the conventions also won at the ballot box. But polls are among the worst possible methods if your goal is to accurately predict the final vote shares, particularly months before the election. There are better alternatives. One highly accurate method is to simply ask voters who they think will win. Unfortunately, pollsters rarely ask this question. They should, and journalists should report on it.

You use PollyVote to predict the election result. What is this method based on?
Graefe: PollyVote is a research project that aims to demonstrate advances in forecasting research. In particular, we apply one of the major findings in forecasting research over the past 50 years, which is that combining forecasts increases accuracy. So we combine forecasts from six different methods, each of which utilizes different types of information. For example, we use traditional polls but also take into account the opinions of experts, as well as forecasts from statistical models. In the latest PollyVote forecast, Clinton leads by 5 to 6 points. Compared to previous elections, this is a relatively clear lead. Furthermore, Clinton leads in five of the six methods that contribute to the overall prediction. The one exception are models that predict the election outcome based on the so-called ‘fundamentals’, such as the state of the economy. But these models are based on the assumption that we are dealing with ‘normal’ candidates who are running ‘normal’ campaigns. That’s clearly not the case this year.

The first TV debate between Clinton and Trump takes place on the 26th of September. Are their performances in these debates likely to sway voters?
Graefe: The first debate could play an important role, because the third-party candidates, Jill Stein und Gary Johnson, cannot participate. The reason is that neither has the required minimum of 15% support in the polls. Many younger voters, particularly liberals, are currently telling pollsters that they intend to vote for a third-party candidate. The big question is whether or not they will actually do so on Election day. I suspect that the television debates will make it clear to many of them that they only have the choice between Clinton and Trump. And that may be the point at which the polls will swing in Clinton’s favor.

Are the debates likely to have an impact not only on pre-election polls but on the final result?
Graefe: The American society is strongly polarized. Each of the two major parties can expect to gain at least 40% of the vote, no matter who their candidate is. In the end, only a very small proportion of voters decide based on who the candidates are and how they stand on the major issues. This is where the debates can have an influence, but in recent years this influence has been relatively minor. The last election in which the outcome of the debates apparently played a decisive role was the close contest between Al Gore and George W. Bush in 2000.

So then, who will be the next President of the United States?
Graefe: PollyVote has been used to forecast the result of US presidential elections since 2004, beginning more than eights months before Election Day and on the basis of analyses that are updated daily. In the past three campaigns, there wasn’t a single day on which PollyVote failed to correctly predict the eventual winner. This year, Pollyvote has shown Clinton in the lead since January. I would be very surprised if all the forecasting methods were to get it wrong this time.

But you did mention earlier that this year’s campaign is not a normal one …
Graefe: It’s true that this campaign has been full of surprises. But a victory for Trump would be the biggest surprise of all.

graefe_130_webDr. Andreas Graefe is a Research Fellow in the Institute for Communications and Media Research at LMU and in the Tow Center for Digital Journalism at the Columbia University School of Journalism. He also holds a Sky Professorship at the Macromedia University of Applied Sciences in Munich.