At YouGov we are committed to the constant review of our approach and methods. After every election we compare our polling to the election results to see how well they performed and what we can learn to build and improve on our performance. The bumper round up of elections last week is no different. Over the weeks leading up to last Thursday we conducted polling on three of the mayoral elections (the London Mayor and two of the key Metro mayoral races) and made MRP projections for some council races. In most of these our figures held up well, in London we fell short.
For the regional mayors, we published polls in the West Midlands and the Tees Valley. Both surveys were conducted between the 12th and the 29th April.
Polling on regional mayors is more difficult than national polling. Partly this is simply because it is a more unusual sample, polls at the national level are more frequent, we have more experience and there are many other companies and academics working in the field with whom we can exchange ideas. Polls of the Tees Valley or the West Midlands mayoral area often only happen at election time, and there is less institutional experience to draw upon. It is also more challenging because it is a smaller area - particularly in the Tees Valley - we simply have fewer panellists to draw people from when looking to invite a representative sample.
As such, we are happy with the outcome of both. The fieldwork was conducted well over a week before polling day, so one cannot be certain that public opinion did not move between our fieldwork and the final result, but as things stand, our poll correctly showed Ben Houchen winning clearly and the West Midlands race too close to call (we had Andy Street two points ahead, in the event Richard Parker won by less than a single percentage point).
Following on from our successful modelling in 2023, we again used MRP to project some of the key council races. Our predictions and what actually happened are set out below. We are broadly happy with this outcome, with our predictions bearing out in most cases (the main exception was Reigate and Banstead, where we expected the Conservatives to make some gains, when actually they fell back).
Finally there is our polling in London. Here our polling held up much less well. While we correctly predicted an easy win for Sadiq Khan, his lead was substantially lower than the numbers in our final poll. Opinion polls are often judged more on the narrative than on the statistics. In a very close race, a poll can be seen as wrong even when it is within the margin of error of the actual result because it has the "wrong" winner. The same can be the case at the other extreme - when an election is an easy win for one candidate, a poll can have the right narrative and right winner, even if statistically it is wrong.
This was the case in London. While we still got the winner right, we significantly understated the level of Conservative support. We will be looking at why we got this wrong and what we could have done better.
Photo: Getty
