Revealing the secrets of the exit polls
It requires a lot of interviewers to stand outside a lot of church halls or school buildings or whatever to collect the data, to get the data in to us, and for us to analyse it.
Doing an exit poll in the UK is particularly difficult as we don’t have precinct-level counts. Rather than us knowing how each polling district has voted, we take our ballot boxes, we bring them together to a central point and we mix the papers together before they’re counted.
The Conservative and Labour vote varies very substantially from one constituency to another. But the change in a party’s vote from one election to the next varies a good deal less. If we want to get an estimate of change, the only way is, wherever possible, to go to the same polling stations. That gives us about 140 estimates of the changes in party vote share, which we then statistically model.
Virtually all of the recent exit polls have come with results that people found surprising. In 2005, we said that the Labour majority was going to be 66. Most people expected a majority of around 100.
In 2010, there was very sharp surprise because we said the Liberal Democrats are going to win fewer seats than they did in 2005. In 2015, we spent a lot of time interrogating the data, and concluded: “Is there any chance it can be wrong? The answer’s no.” We were reporting Tories at 316 instead of 331. I regarded it as being in the outside realm of what was possible.
There is a historical record of the polls underestimating the Tories and never underestimating Labour. The question is whether it’s going to be a small error or a bigger one.