Policy

YouGov CEO on why US election polls missed Trump’s victory

By Oliver Smith 9 November 2016
Summary

What went wrong?

First Brexit, now Trump.

Pollsters were left battered and bruised once again this morning, after an equally embarrassing misstep over Britain’s Brexit vote earlier this year.

On Monday the biggest polling groups in the US – Ipsos Mori, YouGov, SurveyUSA, etc – all forecast a clear route for Hillary Clinton to enter the White House.

Today those predictions lie in tatters, and there’s just one question:

What went wrong?

We spoke to Stephan Shakespeare, CEO of YouGov, to find out:

Oliver Smith: YouGov gave Hillary Clinton a 4% lead on Monday (with a 1.7% margin of error), what went wrong?

Stephan Shakespeare: As far as I know Hillary Clinton is actually going to win the popular vote, only by a sliver, but nevertheless. We were 1 or 2% out, which is perfectly normal, this is not like general election 2015 where we were 3 or 4% out.

And we’re in the middle of a pack of pollsters who were all out by a similar margin.

I’m not trying to be evasive, but it is a slightly different situation because she is preferred by the majority [of Americans], and our state polls were pretty close.

Read more: Election 2015 – Meeting the man with all the answers

OS: Brexit and now Trump are probably the two biggest polling misses in political history, is this a pattern?

SS: We’ve seen establishment politics upended twice now by candidates or campaigns that simply didn’t understand the other side.

That’s something that these two events have very much in common, that there are many people out there who don’t normally vote, but turned out to vote because for the first time they felt they had a cause that represented them, rather than their usual sense of being left out.

There are errors [in the US election], but our errors are in the same direction as in the Brexit campaign, of finding our sample models underestimate the kind of people that typically in an election don’t actually vote. And we need to adapt to that.

Can polling ever account for these historical non-voters?

In our methodology, because we have hundreds of thousands of people online to sample from, we can find plenty of people like you’re talking about, we have many thousands of them on our panel.

Reaching them isn’t the problem, the problem for us is how to model their likelihood to participate in an election.

All typical turnout figures suggested they wouldn’t turn out in these two elections, but they have come out and voted even more.

For instance in Brexit the north [of Britain] showed a higher turnout than the South, which never happens, and it’s very hard to know if something like that will happen.

After these two huge polling misses, can we still trust pollsters?

The Trump election is actually already an improvement – even though it’s obviously a much more shocking result because of its importance to the world – in the sense that if you look at the UK Brexit polls they were much further out.

The [Brexit] telephone polls were 10 or 20% out, but no pollster was that far out for this election.

We are making progress. We know what it is what we should be focusing on, but we can’t be certain whether it’ll be the same thing we should be focussing on at the next election.

Polling is the best model of prediction we have, but no one pretends that it’s infallible or that there’s not a margin of error.

People like to think if only social media could predict or this or that, but as we’ve seen the betting markets are even worse.

We do have to keep improving, but today we are still the best bet.