Are Trump's chances being under- or over-played? Credit: Mandel Ngan / AFP via Getty

September 18, 2020   5 mins

“Herd mentality” may not bring an end to coronavirus, despite President Trump’s slip of the tongue, but the phenomenon is never more real than during election season. In the bubble of opinion pollsters, pundits and the commentariat, a conventional wisdom usually forms about what is going to happen — and it is often wrong. The trauma of the shock Trump victory in 2016, missed by most of the media and pollsters, looms large.

So what is the evidence about who is going to win in seven weeks’ time? Which way is the “herd” really facing? Currently the national polls show Biden ahead by 6-8 points; in the battleground states that will decide the election it’s closer, but he’s still solidly ahead. There’s a lot that can still happen, but in the coming weeks are the polls and media narrative more likely to be underplaying, or overplaying the chances of a Trump re-election? To find out, I spoke to some of the world’s leading experts — including one notable dissenting voice.

Watch the full interview below

Nate Silver runs FiveThirtyEight, perhaps the most famous polling website in the world; he faced a good deal of criticism after his 2016 eve-of-election model gave Hillary a 71.4% chance of winning the election (he’s keen to point out that this was lower than many other models — the New York Times had her at 85%, others at 99%). Right now, FiveThirtyEight gives Trump a 24% chance of winning, and Biden is at 76%.

“It’s a weird election because it’s a little hard to diagnose exactly what the conventional wisdom is,” he tells me. “There’s a meme that everybody is overconfident on Biden, but in fact… people are quite cautious on Biden.”

“If you go to betting markets, they have Biden as only about a 55% favourite. To me that seems hard to justify, low on Biden. He is up by 9 points — he does have disadvantages like the electoral college but I don’t think that quite computes
 When our model was released, everyone kind of nodded and said, ‘70% seems about right to me’ … So for better or worse, our forecast and the conventional wisdom are aligned.”

But not everyone in Washington sees it that way. Republicans tend to think that the conventional wisdom is, once again, under-rating the chances of a Trump victory. The pollster-of-choice for this view is Robert Cahaly of Trafalgar Group. He may be a contrarian voice, but he deserves to be taken seriously: he was one of the few in 2016 to correctly call the crucial Trump wins in Michigan and Pennsylvania. His polls show Trump narrowly ahead, and winning enough of the battleground states to secure a victory in November.

So how is it that his numbers differ so dramatically?

“We think what’s happening is what’s technically called the ‘social desirability bias’ — people give an answer to a question being asked by a live caller which is less about honesty and more about being judged positively by the person asking the question. So when there is a candidate who is polarising that it isn’t politically correct to say you’re for, you tend to give an answer that makes you look best.” He estimates this “shy Trump” effect is between 3-5% and he suspects it could be more, and that “live caller” polling is the method in which the effect is strongest.

It is not totally clear what methods Cahaly is using to assess the extent of the effect. He says it involves a combination of different polling methods (less reliance on live caller polls, more digital and robocalls) and asking additional questions such as “who do you think the majority of your neighbours are for?” to ascertain people’s true motivations. But the complexity of the method raises legitimate questions.

Another problem with this theory is that if such a “shy Trump” effect were a big factor, you would expect to see it most powerfully in districts that are overwhelmingly Democratic, and where being a Trump supporter was most shameful —  in highly Democratic parts of California, for example. Trump should have outperformed his polls there in 2016; but in fact the reverse happened. Clinton did even better than the polls suggested she would in those highly Democratic areas, and Trump outperformed in strongly pro-Trump areas.

In theory, those polling companies that do not use a live caller method should mitigate any “shy Trump” effect because you are not talking to a real person, but that’s not happening either. Doug Rivers is Chief Scientist at YouGov, an online-only pollster that makes use of its large panel of survey respondents:

“Hundreds of thousands of people have signed up to take surveys and we interview them repeatedly over time. What that enables us to see is the same people at different points in time and whether they’ve changed or not. What we’ve seen is about 7% of the Trump voters now say they will vote for Biden rather than Trump, and there’s only a 1% offsetting flow in the other direction. So that’s a big deal…

“We never had Hillary Clinton leading by more than 6 points in 2016. And we’ve had Biden at a high watermark of plus 11.” The YouGov model comes out with an estimate of 340 electoral college votes for Joe Biden (270 is the magic number needed to win).

This raises the opposite possibility that, in fact, the “herd mentality” in 2020 is underplaying the Biden advantage more than overplaying it. Pollsters and the media are so scarred from the 2016 shock that they seem to be doing everything they can to emphasise the ongoing chance of a Trump victory, and not blithely predict his defeat. Perhaps this effect is suppressing reporting of the Biden lead, and actually he’s even further ahead than is being reported?

Says Silver: “There was a recent outcome that a lot of people found traumatising for many reasons. You combine that with the fact that people tend to overweight recent outcomes and more salient outcomes — so just from first principles you would think that people might overweight the chance of Trump winning, just based on their emotional reaction. A lot of what they are doing is emotionally hedging.”

At the technical polling level, this could happen by pollsters anticipating another exceptionally high turnout among white working class voters — like the one we saw in 2016 — when they could well turn out in more conventional numbers this time round.

There’s also the matter of minority voters, as Nate Silver explains:

“There’s also a more subtle form of overcorrecting. You take every step to ensure that you’re counting the white, working class, non-college Trump voters, but you don’t take as many steps to ensure that you’re counting hispanic and black voters. There were some states like California and Arizona and Texas where Clinton beat her polls in 2016 — nobody cared because she lost the key states — but there were groups among which Clinton overperformed… We could wake up and say, oh actually, boy, turnout among blacks and latinos was really high, while polls thought it would be mediocre, and so Biden won by 11 points instead of 8.”

The “smart take” you hear most often about the forthcoming election is that Trump is once again being underrated — people somehow feel it in their bones, and the betting markets confirm that, showing a much closer race than the polls indicate. In a way, “you watch, Trump’ll be re-elected” has almost become the new conventional wisdom, and it’s more uncomfortable to predict a big Biden win.

As Nate Silver says, “If I were looking to play the markets and I didn’t have an ethical problem with doing it, I would think the price on Biden is very favourable.” Doug Rivers agrees: “I think the betting markets, which are showing this about 50/50, are underestimating the chances of a Biden win.”

So is Donald Trump toast? “Not quite,” Rivers replies, “but close – he’s browning rapidly.”

Freddie Sayers is the Editor-in-Chief & CEO of UnHerd. He was previously Editor-in-Chief of YouGov, and founder of PoliticsHome.