If we’d done that last June the outcome would have been much less of a shock
The publication of two sets of leader ratings by Deltapoll and YouGov over the weekend has put a lot of attention on these regular trackers which the records suggest are a better guide to what will happen in elections than voting intention polls.
We know that the voting numbers with one exception in GE2017 joined GE2015 and GE1992 in being some way out and gave a wrong view of what would happen. However if we had relied solely in each election on the leader numbers we probably would have predicted it better.
Two tables from YouGov showing the trend in it well/badly ratings for Political leaders in the months ahead of GE2017. As can be seen Mrs May was in a totally dominant position and maintained that lead right through until the final poll when she was showing a 5% negative. This compared with huge leads earlier in the year.
Yesterday the new polling company established by Joe Twyman, formerly of YouGov, and Martin Boon, ex-ICM, produced the first poll for the Observer. Its findings on LAB’s anti-semitism have made the most headlines but I was pleased to see that is highlighting leader ratings and is not focusing everything on voting intention. In fact there are no voting numbers in the poll although the sample was asked what they did at the last general election. That’s good thinking because it will increase the attention that’s given to these findings.
At GE2015 the final polls had it tight but all the different leader ratings had EdM trailing behind Cameron. It was the same at GE1992 when the polls had it neck and next but Major had very clear leads over Kinnock in the ratings.
A big problem with voting intention polling is that unlike almost all other political polling you are not asking respondents for their view on a subject. Rather you are trying to establish whether at a future date they will act in a particular way if at all. So we have to add on all sorts of complicated calculations in relation to whether they will actually do as they say and turn out on the day and vote. Then we can get into more complex analysis of whether particular groups are responding in a different way to the turn out question so their views can be discounted.
Leader ratings ares simpler and I’d argue the record shoes gives you a higher quality of response.