With the not insignificant exception of Donald Trump’s army of die-hard supporters, there is an all but uniform consensus that Hillary Clinton remains the favourite to win this week’s presidential election, despite her recent troubles.
But there is increasingly spirited disagreement as to whether a Clinton victory is merely a likelihood, or something approaching a certainty.
Betting markets are leaning towards the former view, with Sportsbet currently offering $1.20 on Clinton and $5 on Trump, compared with $1.28 and $3.40 before today’s news that the FBI had again cleared Clinton over her email server.
That leaves Trump with a roughly one-in-four chance of defying the odds, which sits comfortably with NBC News/Wall Street Journal and ABC News/Washington Post polls crediting Clinton with a lead of around 4%.
Not all pollsters are singing quite the same tune though, with one particularly striking exception being the poll series conducted by the University of Southern California for the Los Angeles Times.
This poll, which employs an innovative and somewhat controversial methodology, has consistently had far the best results for Trump, with his lead currently at 48% to 43%.
Another strong poll for Trump has been that conducted for Investor’s Business Daily by TIPP, which produced an all but perfect result on the eve of the 2012 election, and has shown a statistical dead heat ever since the FBI email investigation bombshell was dropped on October 28.
It’s by no means unusual for pollsters to offer such a confusing mix of predictions, which is why the work of forecasters and poll aggregators has attracted so much attention in recent years.
The most famous name in the game is Nate Silver of FiveThirtyEight, whose model correctly predicted every state in 2012, and every state but one in 2008.
This time though, even the forecasters are all over the place — and Silver’s model is the most striking outlier.
The New York Times’ Upshot model is fairly typical in rating Clinton an 84% chance of victory, while Sam Wang of the Princeton Election Consortium lands at the other end of the scale by rating a Clinton win a near certainty, just as he did with Barack Obama in 2012.
While Silver’s model agrees that Clinton is the favourite, it is unique among the forecasts in giving Trump a better chance than implied by the odds from Sportsbet, its current reading for Trump being 35.7%.
Silver’s consistent predictions of an Obama victory did much to soothe the nerves of Democrats during periods of polling volatility in 2008 and 2012, but this time the shoe is on the other foot.
Particularly notable was an article published yesterday on the popular left-of-centre website Huffington Post, which excoriated Silver in terms almost identical to those used against him in 2012 by Republican optimists and press corps pundits invested in a close race (and prompted a furious response from Silver in turn).
While the various forecast models differ in myriad ways, two main points of distinction stand out.
One relates to the balance of national state and polling, with Silver placing greater emphasis on the former.
This has made Silver’s model more responsive to the turn in Trump’s favour after the email story broke, since the greater frequency of national polling means it is more up to date (although the very latest results suggest the Trump surge is now wearing off).
The other factor is the degree of uncertainty the models allow for, which is what makes the Princeton model a standout.
In terms of the states each candidate is considered likely to win, the Princeton and Upshot models predict essentially the same result. But the exact levels of probability are considerably higher in Princeton’s case, since it does not factor in a chance of the polls being systematically wrong to an extent unforeseeable from recent historical experience.
With little time left on the clock, expert opinions exist to support a wide range of possible outcomes, with Clinton’s favouritism being the one point of agreement.
Only 48 hours remain until the guessing games can finally be put to rest.