Votes are still being counted in some swing states around the country, but it's now clear Joe Biden is the president-elect. However, his margin of victory was much closer than many polls predicted ahead of the election.
Some state polls were off by significant margins, reminiscent of 2016 when pollsters widely underestimated President Donald Trump's support. So what do these two elections tell us about the state of political polling in this country? And how should we think about and interpret polls going forward?
To start thinking about those questions, VPR’s Henry Epp spoke with Rich Clark, who is a professor of political science at Castleton University and also conducts the polls that are commissioned by VPR and Vermont PBS. Their conversation has been edited and condensed for clarity.
Henry Epp: So according to the New York Times average, President-elect Joe Biden went into Election Day with a lead of 8.5 percentage points in national polls. And he now stands at just shy of 3 percentage points ahead, 50% to Trump's 47% as votes in the national tally are being counted. So were the polls really as wrong as they seem?
Rich Clark: Well, no. I mean, based on the national polls, I don't think they're far off at all. You started the introduction here with the recognition that both are still being counted. And I also, you know, would point out that even as it stands now for the national polls, really we're within the margin of error.
OK. Pollsters did make a number of changes to how they conduct and weight various groups of people in the polls after 2016. Can you explain what some of those changes were and whether they got us closer to accuracy this year?
Sometimes I think we're expecting too much from polls. When we talk about accuracy and precision, I always imagine people are thinking of a scalpel, and I'm always imagining I'm using a chainsaw. The polls are always giving one value, but it’s – one value is an estimate that should have these error bars around it.
So with that in mind, some of the changes that were made from 2016, it was clear there were problems in many state polls in 2016. However, the national polls in 2016 were about as good as they've ever been. They were – it was about 1.6% off the actual final vote counts for that.
Some of the changes that state polls made were adjusting for education. Not every poll, including the VPR poll up to that point, had been adjusting for education. And all of a sudden, when Donald Trump entered the political landscape, education became a huge factor in vote choice.
More from Brave Little State: These Vermonters Voted For Trump In 2016. What About 2020?
And can you explain a little bit more about what you mean by that? How are you weighting for education?
So typically, we had weighted values to match the census demographics on age, race, location and a number of other factors, not including education, because education and income were so closely associated.
But in 2016, it was clear that lower-educated, especially lower-educated whites, were voting more heavily for Donald Trump than had in the past. And it had been a group that was often – not omitted, but underrepresented in most polling done.
So, giving those groups more weight in what you report out, in terms of the percentage that you expect a candidate to get, is that right?
Yeah, and weighting them closer to their representation in the electorate and in the population generally.
So I guess the big question here after this presidential race, as you're looking at the state polls, were the polls quite off this year?
Well, you know, again, we can go through and cite some polls that were really far off of it. Every time we report a poll, we always report some margin of error. But, you know, it's always the least concern for most pollsters. The job of a pollster is really to minimize error in the research.
One of the errors that we are constantly contending with is non-response error. Who participates? It's no secret anymore that response rates in most polls are around 10% or lower. And so we have that big concern about who is it that's not responding to the polls, and are they different from the people who are responding in the statistics that we're measuring? And that does seem to be the case, which is why we do this weighting to help adjust for non-response.
And so given those changes that were made and given the issues that we're continuing to see, I mean, what is the use for the general public and for journalists of a poll right now? How should we think about and use the information that we see in a poll going into an election?
That's a great question. One of the great concerns of pollsters, especially those in academia, is that the polls are used as the horserace solely, to see who's up and who's down and focus more on strategy than on substance and issues.
The value I see of doing the polls, the horserace polls, is to measure against a projected reality the vote count, to see if our methods are accurate. But what the polls are giving you that you wouldn't have without those, it tells us who is supporting each candidate, whether there's a gender gap, whether people in one region are more likely to support a candidate than another, and also tells us the whys.
After Election Day, you know, every politician likes to say the only polls that matter are the ones that happen on Election Day. Yeah, but those don't explain much. They might tell you how many people voted for Donald Trump and how many voted for Joe Biden. But they don't explain why. They don't explain what were the top concerns of voters. They don't explain why voters were, in Vermont, for example, going overwhelmingly for Joe Biden while at the same time re-electing a Republican governor – Phil Scott.
So the polls give us a lot more information than we otherwise would have that help us understand the electorate and the depth of the issues that are being addressed in each electoral cycle.
So, Rich, you conducted polls for us this year here at VPR and Vermont PBS. And so let's look at the last one you did, which was in September, compared to the actualoutcome of the election. How did you do in terms of the races that you polled on versus how the actual results came out?
We polled back in September, so it was quite a ways before the election. And even with early voting, you know, we were at least a week before the first votes were cast. But with that said, we know that a lot of the national vote had been decided by that point. Our poll had done extremely well in the two state races that we had, the gubernatorial and the lieutenant governor's race, where we were within just small percentages of the actual vote count.
We slightly overestimated the Trump vote and underestimated Biden's vote in our poll, which is different, I guess, from the national polls, where they're looking at underestimating the Trump vote. And then on the race with Peter Welch, we actually overestimated Welch's support against Miriam Berry in that race.
But so overall, what you found in the poll was fairly close, within a few percentage points of the outcome of the election.
That's right.
OK, and so what do you think that says about your methods of polling and about polling in Vermont in general?
Well, I think polling in Vermont is – one advantage we have, or I have, I should say, in polling – is that we have fewer factors to consider without having huge populations that differ dramatically by race or by regions.
I also think our response rates, we constantly find our response rates higher than national averages. And in I think that's because Vermonters don't have the distrust of polls or of the election process that we see in other states, and we know that distrust is also associated with the lack of participation in polls.
Have questions, comments or tips? Send us a message or get in touch with reporter Henry Epp @TheHenryEpp.
We've closed our comments. Read about ways to get in touch here.
