The “MRP” method has likely captured the overall picture but there should be caution over individual seatsby Peter Kellner / November 28, 2019 / Leave a comment
This is not a sentence I ever expected to write; but I have some sympathy for Dominic Cummings. Boris Johnson’s notorious adviser is worried that too many people will take a Conservative victory for granted, stop worrying about Jeremy Corbyn becoming prime minister, and hence fail to vote Conservative on 12th December. Might complacency and the desire of many lukewarm Tories on a dark, cold and possibly wet December evening to stay at home, cost Johnson his premiership?
Cummings is right to be concerned. The evidence is stacking up that the Conservatives are on course for a clear, and possibly large, majority; but their lead is not impregnable. Like a soccer club two goals up with ten minutes to go, victory is very likely but not yet certain.
Today’s Times brings Cummings no relief. YouGov has questioned 100,000 people. It projects a complacency-inducing 68-seat majority for the Conservatives, who are ahead in 359 seats. Labour is down to 211—just two more than in their catastrophic defeat under Michael Foot in 1983. The SNP is up to 43, while the Lib Dems, with 13, are up just one since 2017—their four expected gains are offset by three projected losses.
Do YouGov’s figures make sense? My large answer is yes; but I have some reservations about particular seats, which I shall discuss below. (Non-interest declared: I left YouGov in 2016 and have had no involvement in the developments of the techniques it has used to make these projections.)
What YouGov has done is employ a method known as MRP (“multilevel regression post-stratification,” since you ask) to estimate party support in each constituency—not so much by recording what voters in those seats say (there are too few, even in a sample as large as this) but by lining up the demographic profile of each constituency with the voting intentions of people like that in similar seats.
This method allowed YouGov to warn the Tories two years ago that they might lose apparently safe seats such as Canterbury and Kensington—while telling Labour that it was in trouble in traditionally heartland seats such as Walsall North and Mansfield. YouGov’s data was not always right: for example, it understated the surge to Labour in a range of Remain-voting seats in London. But the big picture was…