Donald Trump’s unexpected victory in Tuesday’s Presidential election has sparked a backlash against the pollsters who predicted his opponent, Hillary Clinton, would prevail. The President-elect had not led in the polls since convention season in July but crossed the winning line after taking Wisconsin. Though that particular result was entirely unexpected, the polls were far more accurate than recent rhetoric would suggest.
At the time of writing, Hillary Clinton has received over 200,000 more votes than her opponent. Though President-elect Trump looks set to amass over 300 electoral college votes, with three states still to declare, he trails in the popular vote by 0.2%. This means that the predictions, made by numerous polling sites, of a national poll lead for Clinton of around 3% were within the expected margin of error. Furthermore, almost all pollsters correctly predicted which would be the key swing states. Whilst it doesn’t take an expert psephologist to recognise that Florida would fit this category, the polls showed that perennial swing states Ohio and Iowa were relatively safe for Trump and that typically red states like North Carolina were up for grabs.
This is not to say that there weren’t states the pollsters got wrong. Unexpected victories in Pennsylvania, where Clinton was thought to be up by around 4%, and Wisconsin, not won by the Republicans since 1984, delivered Trump the presidency. Anecdotal evidence of voter suppression in the latter and the occupational demographics of the former may shed some light on these polling errors. Regardless, it’s clear that the national polls – and those in the vast majority of states – were reasonably accurate.
For genuine polling errors we should look closer to home. At the 2015 general election, national polling averages had Labour and the Conservatives neck-and-neck before David Cameron’s party stormed to an unexpected outright victory. The Tories actually outpolled Labour by 6.5% – almost two million votes. The EU referendum polls also proved to be less accurate than those for the 2016 Presidential election. Most poll-of-polls final averages had Remain ahead by around 2%; roughly 4% off the eventual result; 52-48 in favour of Leave. The key to both these polling errors and perhaps Wisconsin too (if not Pennsylvania) is turnout.
Whilst pollsters can be fairly sure than participants will give honest answers, it’s much harder to know whether someone will take their opinion to the ballot box. With the two candidates in this election dividing opinion so starkly along demographic lines, fluctuations in turnout have an even greater effect. Across America it seems that the Black vote was down compared to 2012 – a demographic evidently favourable to Clinton. As yet unconfirmed reports of tens of thousands of voters without the necessary ID being turned away from the polls in Wisconsin, when added to Trump outperforming his polls by an average of around 2%, may explain the unlikely result in this state.
A number of organisations constructed models to derive the percentage chance of each candidate winning the presidency, based on the polls. Though providing an interesting insight into what the polls were showing, these models arguably warped many people’s view of the data. Being told, for example, that there’s a 70% chance Clinton will win didn’t indicate to most people that a number of key states would still be down to the wire. There’s a 25% chance that you’ll correctly guess the result of a coin flip twice in a row; such a feat is therefore less probable than Trump’s victory according to FiveThirtyEight. Admittedly, most poll-trackers gave Trump less of a chance. But add one more coin flip and you’ve crossed the 15% chance predicted by the Upshot. Even models showing Trump with a 10% chance gave him odds of 2:1 in states like Florida and North Carolina.
That Donald Trump’s election on Tuesday was unexpected does not mean that the polls were wrong. Several key swing states were always a toss-up and, Wisconsin aside, no state’s result can really be considered a shock. The difficulties in determining turnout and presenting data in an accessible way contributed to the illusion that pollsters misjudged the race. However, in such a bizarre election, with two remarkably unpopular candidates and changing voter demographics, it’s frankly surprising how accurate the polls were.