Runner and rider. Wikimedia Commons, CC BY-SARunner and rider. Wikimedia Commons, CC BY-SA

In a US presidential election year, Labor Day (the first Monday of September) marks the traditional start of what the Americans call the “fall campaign”. The party conventions have been and gone, the Olympic Games are over for another four years, the candidates have nominated their vice-presidential running mates, and debate preparations are well under way.

In short, we’re nearly there. So how does the race stand – and what’s the best way to take its temperature?

Traditionally, we look to the opinion polls for some indication, and they are as prevalent this year as they have ever been in dominating the news narrative. The RealClearPolitics poll average has Hillary Clinton leading Donald Trump by around five points, ranging from a ten-point lead in the Quinnipiac poll to a three-point lead for Trump in the LA Times/USC poll.

The LA Times/USC poll has always favoured Trump relative to other polls, even including the traditionally Republican-leaning Rasmussen. In fact, its tilt is so large compared to the average that it’s generally a good rule of thumb to add upwards of six points to its Clinton number to find the average of all the polls.

There is, of course, no statistical problem with a biased poll so long as it is consistently biased – that is, in the same direction and by about the same amount. This makes it easy to de-bias. And yet, the RealClearPolitics poll average takes this one at face value, simply adding it to the pool to produce an overall average.


innerself subscribe graphic


Because of the peculiarity of the American election system, however, the national polling average is not as valuable as individual state polling. It’s the states, not the whole country, that decide the election. To be elected president, a candidate must win more than 269 electoral votes, which are apportioned to the individual states based on the size of their delegations to Congress. With the exceptions of Maine and Nebraska who both allocate some of their electoral votes to whoever wins each congressional district, winning a state usually means winning all its electoral votes.

On the basis of state polls, RealClearPolitics predicts that were the election held now, Clinton would win 362 electoral votes, with 176 for Donald Trump. Even if we remove the states judged close enough to be potential toss-ups, Clinton is currently predicted to win 272 electoral votes – still just enough to put her over the line.

But how useful are these sorts of polls, really, for predicting the final outcome?

And they’re off

In one sense, they’re not “predicting” it at all. What they provide is an estimate of how people would vote if there were an election today. If the average is an accurate representation of that, it means that this is Clinton’s election to lose and Trump’s to win.

When a scenario like this presents itself, it’s the trailing candidate who welcomes variance, or anything that shakes up the race. More often than not, they resort to shaking it up themselves – the equivalent of knocking over a chess table when you’re about to be checkmated. John McCain deployed this tactic in 2008 when he chose Sarah Palin as his running mate; at first, she seemed to genuinely upend the race in the GOP’s favour, but she then went on to become a significant liability.

For Trump, the equivalent is perhaps a major terrorist attack on home soil, or a huge scandal to eclipse the Clinton campaign. Even the debates, not widely expected to help Trump’s chances, provide a potential for some upset. President Obama’s shaky performance in his first debate against Mitt Romney in 2012 was a textbook example.

Polls alone can only do so much to account for late-in-the-day wobbles. Those who play in the betting markets, on the other hand, do their best to factor in the current state of the race, and the potential for upsets, and make a best estimate of the likely final outcome.

In a recent paper I co-authored with Dr. James Reade of the University of Reading, we employed huge data sets to compare the performance of opinion polls and betting (or prediction) markets, over a large number of US elections. We concluded that the markets tend to be more accurate than the polls in identifying election outcomes.

There are signs that pundits and reporters are noticing; RealClearPolitics, for instance, also now reports the latest betting odds. On that measure, the site gives Clinton about an 80% probability of winning on November 8.

But before they start measuring a new carpet for the Oval Office, the Democrats would be wise to note that a 20% chance of losing isn’t much different to the 25% chance of drawing a spade from a shuffled pack of cards without jokers. And for all that the polls and the markets rate Hillary Clinton the solid favourite to be the US’s 45th president, there are likely to be many stumbles yet before the fall campaign concludes.

So keep an eye on the numbers, make sure to look beyond the polls – and remember, there’s still all to play for.

About The Author

Leighton Vaughan Williams, Professor of Economics and Finance and Director, Betting Research Unit & Political Forecasting Unit, Nottingham Trent University

This article was originally published on The Conversation. Read the original article.

Related Books

at InnerSelf Market and Amazon