The polls were wildly off in the Virginia election.
Democrat Ralph Northam had a lead over Republican Ed Gillespie throughout the race — but a lead that notably tightened to 1 or 2 points by Election Day. Pundits and pollsters were all in agreement: Virginia’s polls were leaning toward Northam, but the race was a toss-up.
But on Tuesday night, Democrats scored a landslide victory, taking the governor’s race by 8 points and winning far more delegate seats than expected.
The blue wave reflected a state that not only had denied Donald Trump a Virginia victory in 2016 but also continues to view the president extremely unfavorably. This year, the poll average in the Virginia governor’s race was off by 5.3 percent, Tom Bonier, a Democratic pollster tweeted Wednesday morning. To put that in context, polling was only off by about 1 percent in the 2016 election.
So the poll average was off in the VA Gov race by 5.3%, as compared to only 1.1% in the '16 Presidential. Yet because the polls predicted the winner in VA, we probably won't talk about what we're still doing wrong.— Tom Bonier (@tbonier) November 8, 2017
That the polls were wrong likely won’t get as much attention this year, because they happened to be wrong in the correct direction — showing Northam with a slight lead.
Nevertheless, exactly one year after the 2016 presidential election, a race that made social scientists and pollsters reconsider their fundamental models for evaluating the political climate, the polls are still epically off. In fact, they were more off in Virginia this year than in the 2016 presidential election.
In fact, the poll average missed the margin in Virginia by more than it did in Michigan in '16. Look, polls don't predict turnout. Pollsters do, and some are better than others, but sometimes it's very, very difficult.— Tom Bonier (@tbonier) November 8, 2017
There might be a good reason for this. As the New York Times’s Nate Cohn reported, most public polls, conducted through newspapers and colleges, actually haven’t changed how they do things:
Few if any of the public pollsters that conducted surveys ahead of Tuesday’s elections for governor in Virginia and New Jersey appear to have adopted significant methodological changes intended to better represent the rural, less educated white voters who pollsters believe were underrepresented in pre-election surveys.
Cohn points out there are a few pollsters in Virginia that do weight factors like education — and they did so before the 2016 election.
Those include polls like Quinnipiac’s, for example, which had predictions closest to the actual results in 2017, coming within 3 points.