To anyone getting all warm and cozy with the polls showing that voters will decide to stay with the European Union in Thursday’s Brexit vote, I’ve got just one word of warning: Michigan.
Remember the Michigan primary between Hillary Clinton and Bernie Sanders this year? The polls had Clinton ahead by double digits. Nate Silver’s 538.com, one of the most thoughtful poll crunchers had her up by 21 percentage points. In what the media saw as a startling upset, Sanders won by 1.5 percentage points.
The polls were so wrong in Michigan because of problems with past vote weighting. And the same problems should make you regard all the current polls in the United Kingdom with extreme skepticism.
Polls don’t survey every voter in the electorate. They take a sample and then weight that sample based on the way that different kinds of voters turned out in the last election. In the Michigan primary, because of some extremely odd history–one year the state wasn’t allowed to hold a meaningful primary as punishment for violating Democratic Party rules, for example–there wasn’t a recent election to use in weighting those samples. And it turned out that the participation by different classes of voters was radically different than it had been in the last, very distant, primary.
So the polls were really, really wrong.
Same problem is making the accuracy of the Brexit polls in the United Kingdom even more questionable. In a very interesting podcast on 538.com, Tom Clark, chair of the editorial board at the Guardian newspaper, notes that the last referendum on Europe that poll takers might be able to use for weighting took place in 1975. Since then the electorate has shifted–there’s the rise of the far right, Euro-skeptic, anti-immigant UKIP party, for example. And perhaps even more important the decline of the number of people with landline phones and the rise of Internet polling have changed response rates to polls and have resulted in considerable self-selection by poll participants. Absent a recent referendum vote to use as a model, pollsters have to fall back on weighting from more recent national elections–and no one really knows if voters behavior is the same in national elections and in referendums.
No one knows what the exact result of any shifts in the electorate have been–or how accurate or inaccurate those shifts might make current polls. (And as tempting as it is to extrapolate betting odds from bookmakers to referendum results, no one really knows how that sample of voter opinion relates to the opinions of the entire electorate.)
In other words, it’s quite possible that the Thursday results will bear very little relationship to the polls as they now stand. I think it’s likely that the trends that seem to be moving toward a win by the “stay” in the European Union position are real trends. But there is certainly a non-zero chance that all these numbers are rubbish.
Which, of course, would give a surprise result–one that went against the market’s current belief in the polls–real power to move the markets. And which might be one reason that the Standard & Poor’s 500 Volatility Index (VIX) soared 14.56% to 21.17 on June 22.