How accurate were the polls in the 2019 election?

This article was made possible by our Patreon supporters! Special thanks in particular go to our amazing $5 supporters: Vincent Calabrese, Scott Folan, Josuke Higashikata, Chloe Hopkins, Harry Jackson, Stephen Kaar, Bert Rothkugel, sn, Alan Spence, Alex Wilson

To support Stats for Lefties on Patreon, visit https://www.patreon.com/leftiestats

--

British opinion polls have had a difficult few years. On average, they predicted the wrong outcome in the 2015 general election, the 2016 EU referendum and the 2017 general election. It was probably unsurprising, therefore, that many commentators - myself included - expected the election result to be surprising in one way or another.

As it turned out, the most surprising thing of all happened: the polls were actually right.

On average, the polls in the final week of the election (9th-12th December) showed the following results:

Tories 43%, Labour 34%, Lib Dems 12%, SNP 4%, Greens 3%, Brexit Party 3%

And on December 12th, the actual election result within Great Britain was:

Tories 45%, Labour 33%, Lib Dems 12%, SNP 4%, Greens 3%, Brexit Party 2%

The average polling error for the three major parties in the final week was plus or minus 1.03pts, the lowest average polling error since 1955 - so, in other words, the polls were more accurate in 2019 than in any election for over 60 years. That's impressive. But what was even more impressive were some of the results for individual pollsters. Let's go through the results for individual pollsters and see how accurate they were.

Opinium was the most accurate

Many people were surprised by the large Conservative leads that Opinium was showing throughout the campaign, but in the end, they were proven right - exactly right. As you can see in the graph below, Opinium's final poll predicted 45% for the Tories, 33% for Labour and 12% for the Lib Dems. They estimated that the Tories would win by 12pts; the Tories won by 12pts.

Opinium's average error for the top 3 parties was therefore just plus or minus 0.2pts, the lowest average polling error of any pollster in 2019.

The table below shows the final poll from each pollster, and the average error for the top 3 parties. Note that I have calculated the average error for Qriously by comparing it to the UK-wide result, as unlike other pollsters, Qriously polled the whole of the UK.

Opinium, Ipsos MORI, Number Cruncher Politics and Kantar Public all had average errors of less than 1%. All pollsters estimated that the Conservatives would win - the only difference between the pollsters was in how big the Tory margin of victory was estimated to be.

Overall, the pollsters performed very very well, and the average of polls overall was more accurate than in any election since 1955.