The Big “Bad Data” Brexit
The British vote to leave the European Union took the world and most UK politicians by surprise. The backstory is this:
David Cameron, who recently resigned as UK Prime Minister, had been struggling with a severe political problem. His Conservative Party was in danger of splitting because of Europe. In particular, the UK Independence Party (UKIP) had been attracting significant defections from the Conservative Party, which had a sizable Euroskeptic constituency. In an effort to contain this rebellion, at the previous election, which the Conservatives won unexpectedly (if you put any faith in opinion polls), he promised a referendum on Europe.
He never expected the very clear Brexit result. Most of his fellow politicians, most of the UK talking heads and the opinion polls leading up to the vote suggested that the UK public would vote to Bremain. Little did they know. Cameron’s failure to read the runes correctly will doubtless go down as a historic political error.
In my mind, the question is: Can we blame Cameron’s error on bad analytics? Not exactly, but let’s examine this.
Political polling has grown increasingly inaccurate on both sides of the Atlantic. It has become bad analytics. The old methods for gauging public opinion are no longer trustworthy for a variety of reasons:
- If you ask someone which way they will vote, they may not tell the truth.
- The means of gathering opinion data has fallen foul of cost and technology. Gathering data is expensive. It mainly involves phone calls, because face-to-face interviews are too expensive. But phone calls are also expensive, so sample sizes are usually small, which means the results often have a wide margin of error.
- Sampling and adjusting for result skew has been made more difficult by the adoption of mobile phones. It introduces extra skew.
- Internet polling is usually horribly inaccurate.
- The traditional media has changed for the worse and social media is further distorting the whole political process.
The mainstream media and social media impact is, in my view, more profound than many people realize. We have gradually moved to a video-driven-sound-bite-and-tweet world. This has, I believe, altered the impact of mass media on the voter. The voter is fed on a diet of one-dimensional opinions, usually chock-full of emotion and expressed unreliably in a few words. This mitigates in favor of the best slogan rather than any considered opinion, so the voter is more likely to make either an uninformed or an emotional decision.
In regards to Brexit, it appears that opinion was volatile. In a poll of polls, the Bremain led almost all the time except for a few days before the vote, but that corrected in the last polls taken. Conventional wisdom has it that voters tend to become more conservative of the status quo as polling day approaches. Conventional wisdom was wrong in this instance. The vote gave a Brexit majority (52% to 48%), which was greater than the poll of polls had ever suggested.
General Charles De Gaulle, the long serving post-war President of France, once said about referendums: “The problem is that the voters often answer a question you didn’t ask them.” It seems that in the Brexit vote, more UK voters answered the emotional question, “Do you dislike immigrants?” rather than the more cerebral question, “Do you think leaving the EU will damage the UK economy?”
As for the sound-bites and the tenor of the campaign, that sums up the discussion in a few words.
There seems little reason now to have much confidence in political polling because it is establishing a nasty habit of being inaccurate. There are alternatives, and more precise methods can be found. There is the real possibility of using better predictive modeling. After all, if you can predict customer churn precisely, you should be able to predict voter churn with equal accuracy. But it may be that the churn is too volatile to deliver any level of accuracy.
And as for “Cameron’s error,” if his political calculations were founded on opinion polls, he can cry mea culpa. But he had very good reason to mistrust the polls. He had, in fact, recently won an election that defied the message of the polls.