Yesterday the UK voted to leave the European Union by a margin of 51.9% to 48.1%. Our final pre-election poll, published yesterday morning, showed that of those people who were likely to vote, 52% said they would vote remain, and 48% to leave.
Clearly, we are disappointed that this was not more accurate. Having said that, we have consistently said that the race was very tight based on our last two polls, and that the outcome of the referendum was not decided. Indeed, in our final results we published the 26% probability – one in four - that it could be a Leave vote, and pointed out that even in our final poll 13% of voters said they might change their mind when they actually reached the polling booth.
Looking at our polls and those of other companies over the course of the campaign, it is clear that the narrative of a comfortable win for Remain was not something which was backed up by the data: 11 out of the last 23 polls in the last two weeks have had a Leave lead, indeed our penultimate figures published on 16 June, just a week before polling day, had Leave on 53%.
As always, we will of course do our own post-event analysis which looks to understand this and work with others on what the lessons are. One thing we can say is that it is clear turnout had a big impact, especially given the large differences in attitudes towards the referendum by age, class and region, something we have consistently pointed out and shown in our data. Turnout was higher than in the last five general elections, and analysis suggests it was higher in leave supporting areas than regions which supported Remain, such as London.
We have already made changes to our approach following the 2015 election, and many of these (such as correcting for newspaper readership and education levels) we believe have made our samples more representative. We also changed the way we account for turnout, based on the learnings from 2015 that we over-represented voters in our final results. However, turnout in the referendum in many parts of the country appears to have been different, so in this case it had the effect of making of our referendum polling less accurate (if we had used our previous method of accounting for turnout, we would have shown a two point leave lead in our final poll). But the detail of this effect will take further examination: for example, the same approach produced a near-perfect result in our final, separate Scotland-only poll.
We remain strong advocates of the role polling has to play in illuminating political campaigns.