Since 2015, in which the polls underestimated the Conservative vote share, polling companies have made efforts to improve their predictions by weighting the number of people who say they will vote for each party by the likelihood that they will actually cast a vote. For example, in the latest YouGov GB-wide poll, the Conservatives have a three point lead (42-39). Before weighting for turnout, however, their lead is only one point (31-30). But just how much of a difference does this make to the result?
Using just this poll from YouGov for the English voting intention, as well as the latest single polls for Wales and Scotland, my model predicts 330 +/- 19 seats for the Conservatives and 246 +/- 18 for Labour. With 10000 simulations, the Conservatives won a majority 61% of the time. The histogram is:
But polling companies also provide some of the raw data, so we can run the simulation for the results before the turnout correction has been applied:
Without this correction, the prediction becomes:
- CON 310 +/- 17
- LAB 352 +/- 15
And crucially the probability of an outright Conservative majority drops to just 21%. So just how good are these predictions of turnout for different groups of voters? We’ll find out on June 9th…
Bear in mind, however, that both of the above scenarios are based on a single poll for England – the latest YouGov/Times poll. Simulations using aggregated polling data predict that the Conservatives will get 355 +/- 13 seats, with a 98% chance of a majority.