On the night of the 2019 UK election, a colleague made an astute observation: people always expect the previous election to replay itself. In 2017, many thought the Tories would outperform the polls, as they had in 2015; instead Corbyn drastically outdid expectations. Many then expected a repeat of the 2017 election in 2019, only for Corbyn to do just as badly as the polls had predicted.
The same sort of phenomenon is now playing out in the US. After the shock of President Donald Trump’s victory in 2016, many are reluctant to believe the polls or the forecasting models that are based on them. The Economist gives Biden a 96 per cent chance of victory, while Nate Silver’s FiveThirtyEight and the New Statesman‘s model (both 88 per cent) are similarly bullish on his prospects.
You can find reasons to be cautious if you are looking for them. But, first, know this: 2020 is not 2016. Biden is further ahead in the polls than Clinton was; if Biden loses it should come as a far greater shock.
In the 2016 election, 17 states were decided by fewer than ten points; ten of these were won by Trump, whose tally included the three key Midwestern states of Pennsylvania, Michigan and Wisconsin – the so-called “Blue Wall”. The final RealClearPolitics average across these 17 close states, averaged together, gave Clinton a lead of just 0.6 points. As some analysts said at the time, the 2016 race was always very close, and got closer as election day approached. It was never clear that Clinton was a shoo-in.
The 2016 polls were wrong, but – taken in the aggregate – they were wrong by less than you might think. They pointed to a very close election. Trump won the three critical Blue Wall states by 0.2, 0.7 and 0.8 points. The rest of the 17 close states were split between Clinton and Trump: seven each. Forecasters erred in 2016 by turning what was a very tight set of polls into an overwhelming probability of a Clinton win.
This time around, the forecasts’ strong probabilities of a Biden win are based on much stronger polls. While Clinton’s pre-election lead across the 17 states was only 0.6 points, Biden’s is a sizeable 5.9 points, according to the New Statesman’s 2020 election model, based on the latest polling data. In 15 of the 17 states, Biden’s lead is greater than Clinton’s in 2016; in ten of them his lead is at least four points greater. This gives the polls a lot of room to be wrong. In 2016, these 17 states undercounted Trump’s support by an average of 1.5 points – but Biden’s lead is more than three times greater than that.
Most encouragingly for Democrats in the US, the polling picture in these 17 close states is mirrored in two other sets of data: national presidential polls, and the district-level polling conducted in House races across the country.
First, take the national polls. There have been 262 Biden vs Trump polls since January 2019. A crude average of these puts Biden on 50 per cent and Trump on 42.5. The current RealClearPolitics average, using only the most recent polls, is little different: Biden on 50.8 and Trump on 43.0, a lead of 7.8 points. This margin is exactly what FiveThirtyEight forecasts by election day (with higher vote shares for both candidates).
This is a 5.7-point shift from 2016, as Clinton beat Trump in the popular vote by 2.1 points. A 5.7-point shift very closely aligns with the shift since 2016 across the 17 closest states. Trump won these 17 states by 0.9 points in 2016, and he now trails Biden by 5.9 points across them: a shift of 6.9 points. The swing in both the national and key state polls largely reflect one another.
This swing is also, perhaps most crucially, echoed in the district polls carried out for House races across the US. Democrats are set to expand their majority in the House by between five and 15 seats. This under-discussed fact may be the best indication that Trump is in deep trouble.
In 2018 Democrats won the House by 8.6 points after losing it by 1.1 points in 2016. Since then, the 2018 “blue wave” appears to have scarcely subsided. FiveThirtyEight expects the Democrats to win the House by 6.0 points; RealClearPolitics puts the Democrats ahead by 6.6.
This is encouraging for Biden as the House polls were broadly accurate in both elections: in 2018 they put the Democrats ahead by 7.3 points, in 2016 they put them ahead by 0.6 points. The polls were “wrong” in 2016 – the Republicans won – but the miss was small each time: 1.7 points too generous for Democrats in 2016, 1.4 points too harsh in 2018. With a projected lead of six to seven points this year, Biden and the Democrats appear to be comfortably ahead.
This House lead may be very telling. In 2012, when Obama won re-election, the Democrats overturned their disastrous House showing in the 2010 mid-terms, when they lost the chamber by 6.8 points. House Democrats struck back in 2012, beating the Republicans by 1.2 points.
Trump’s Republican Party, in contrast, appears to have barely made up any ground nationally in the House (and are even likely to lose seats having alienated key suburban voters in certain battleground districts). If Trump is winning nationally, or holding on to the Blue Wall – contrary to all the data that suggests otherwise – we should see evidence of it in the district-level House polling, as in 2016. But we do not.
As Dave Wasserman, a non-partisan elections analyst and the House editor of the Cook Political Report, put it recently: “In 2016, district-level polling in late October showed flashing red warning signs for Clinton in districts dominated by White non-college voters. It wasn’t being detected so much in state-level polling, because the state polling chronically under-sampled those voters.”
Wasserman isn’t offering a retrospective analysis. He identified the trouble Clinton was in ahead of time in 2016. But he sees a very different story this time around. “In my 13 years of covering House races,” he wrote last week, “this is probably the most consistent cycle I’ve seen: Trump is underperforming his 2016 margins by 8 to 10 points in most competitive districts.”
[See also: the New Statesman’s US election forecast]
Trump and his party appear to be being beaten across the board. This shouldn’t surprise us: Trump has been a deeply unpopular president since taking office. A majority of Americans have disapproved of his performance since his inauguration in 2017. And without a strong set of third-party candidates this year (as in 2016), that disapproving majority appears to be almost entirely behind Biden.
Nevertheless, such confident analyses may be wrong. If they turn out to be, and Trump wins, one polling company will thrive: Trafalgar Group, a young and openly Republican-leaning operation that correctly called the 2016 election in Pennsylvania and Michigan, and is again suggesting that Trump could very possibly win next week. Trafalgar will tell you that Trump is within 1 point in Pennsylvania and Wisconsin, and ahead by 2 in Michigan (where FiveThirtyEight thinks Trump trails by 9).
It is easy to be captured by this narrative: Trafalgar was right in 2016, it must be right again. As in 2016, the picture Trafalgar paints is supported by only a handful of other outlier polls: InsiderAdvantage recently put Trump up by 3 in Pennsylvania; Zia Poll put him up 4 in Michigan; and Susquehanna sees a tied race in Wisconsin. Will we look back on these polls as the key data points we ignored?
It is possible, but it would be remarkable – far more so than in 2016, when polls in both the 17 most contested states and across House districts showed a very close race, far closer than most of the media realised or reported. This time is different. Trump is no more popular than he was in 2016, while Biden is far more popular than Clinton. On these two facts, polls of every type – national, state and district – are almost united.
With less than a week until election day, Trump appears to lack a path to victory. The Trump era may be all but over.
NB: This piece was updated on 29 October with data from the New Statesman 2020 election model, which slightly increased the shift to Biden relative to the pure RealClearPolitics polling averages in the 17 close seats.