After the United Kingdom’s polling missed the 2015 general election, predicting a popular vote tie between the Conservatives and the Labourites, we had reason to question the predictiveness of the polling on the European Union referendum.
While the science of polling makes sense, it’s clear the practice of polling has diverged from the science. They can’t get anything right!
Six polls came out at the end, ahead of Thursday’s vote to leave the European Union. They mostly got it wrong:
|Actual result||51-49, Leave +2|
|Populus||45-55, Remain +10|
|YouGov||49-51, Remain +2|
|Ipsos MORI||46-49, Remain +3|
|Opinium||45-44, Leave +1|
|ComRes||46-54, Remain +8|
|TNS||43-41, Leave +2|
|Average||46-49, Remain +3|
The polling was wrong. We see Opinium and TNS looking correct, but they were simply at the fringes. If the polling were accurate, we’d have expected results that swung the other way as well. Not just a Remain +10, but also polls with Leave up by double digits.
Polling averages work when the polls themselves don’t have a systemic bias. We can have no doubt they do. They overstated the left last year, and they overstated the EU this year. I fully expect that they’ll trot out their usual excuse, and say there were “shy leavers” in the same way they always claim the eternal “shy Tories” keep their polls from being accurate.
At least in the UK, we now seem to be at the point where the Dewey Defeats Truman-style polling failures are more common than the actual successes. That’s a sign of an industry in complete disarray. Should we believe US pollsters are more accurate?