Alastair Meeks on predicting politics better
“How could I have known?” is the cry of palookas at the card tables everywhere. It’s a cry that’s been heard much in political circles in the last few years. In short order, David Cameron won an overall majority in 2015, Leave won the EU referendum in June 2016, Donald Trump won the presidential election in November 2016 and Theresa May lost her overall majority in June last year. None of these events had been expected by the pundits.
But far more significantly, none of these events had been expected by the outperformers. On the day of the 2015 election, David Cameron had prepared and delivered as a practice run a resignation speech. Nigel Farage effectively conceded defeat on the evening of the referendum. Michael Wolff in “Fire And Fury” claims that almost the entire Trump campaign thought he’d lost (and most actually wanted to lose). Almost all of Jeremy Corbyn’s campaign team thought that Labour would lose seats on the night of the June 2017 election.
This matters. In physics or engineering, predictions can be made reliably. In advertising, John Wanamaker, an American department store magnate, joked that half his money was wasted but he didn’t know which half, but even he on his own estimation managed a 50% strike rate. In a world of politics where even the successful don’t know what they’re doing, how are the rest of us supposed to figure out what’s going on?
The problems with polling
The pollsters have had a wretched period in Britain. They got 2015 wrong, as many got 2016 wrong as got it right and they differed so much on 2017 that it’s hard to tell whether the couple that got close got there by luck or judgement.
Polling is hard. It’s hard to get a statistically reliable sample. The people who are willing to pick up the phone or join an internet panel and then answer a pollster’s questions may be unusual in lots of ways. It’s hard to get that statistically reliable sample to be truthful (or to make appropriate adjustments). And, irritatingly, people can change their minds after they have given their responses, meaning that the snapshot is not a prediction.
These are not new problems but they seem to be becoming more intractable. After the 2015 election, ICM reported that under 7% of phone calls placed resulted in a respondent answering their questions.
I don’t have a better indicator or predictor of public opinion than opinion polls and they obviously have some worth, but they need to be treated with a healthy disrespect. I wouldn’t pay too much regard to the claimed margins of error. Past experience has shown that opinion polls are nowhere near as accurate as that.
Tighter races
Adding to the pollsters’ problems, we have seen a series of close contests. Relatively minor errors assume disproportionate importance. 52:48 is not far from 48:52, but that’s the difference between winning and losing. The pollsters mostly got 1997 quite wrong but Labour’s margin of victory was such that no one noticed.
Increased volatility?
Voters seem in some recent elections to be less tied to particular voting patterns than previously. Opinion polls moved massively in the 2017 general election campaign and there is some evidence of late swing in both the 2015 general election and in the US presidential election (though not particularly in the EU referendum vote). This would be a good thing, since it would suggest that politicians need to earn more votes than previously. It would also mean that past polling was of less value than ever before.
Set against that, polls in the UK now seem to have settled into a rough equilibrium. It seems that Brexit and the 2017 general election may have settled the political views of many previously flighty voters for now and they will need something disruptive to make them reconsider. Or perhaps they’ve just turned off for now and are giving their responses on autopilot until they need to think about the problem again properly.
Given recent past volatility, any sense of inevitability going into an election needs to be abandoned. Similarly, the concept of a sure thing needs to be abandoned. The range of possible outcomes of any election should be viewed more broadly than most people have done before.
Beware of false narratives
Correlation is not causation. We all know that and all of us forget it all the time. Theresa May reshuffles her Cabinet and, say, the Conservatives fall in the polls. Does that mean that the one caused the other? Maybe. Or maybe petrol price rises alienated drivers. Or maybe the NHS flu crisis has worried voters. Or maybe Conservative voters have become unusually reluctant to answer the phone because of a spate of nuisance calls.
We all like to build stories around the facts that we have to hand. A particular problem is that political pundits are particularly good fabulists and their stories grip their audiences. Sometimes those stories adequately explain what’s going on. More often, they don’t but even after dissonant information has come to light we are reluctant to discard our stories. We need to be quicker to do so.
Picking through the rubble
That doesn’t leave much standing. It’s important, however, not to confuse two different objectives. If you’re looking to make predictions, the first thing to realise is that the future is fluid. The pundit who has emerged from the last few years with most credibility intact is Nate Silver, who rated the chances of a Donald Trump victory far ahead of most other pundits and who cautioned how risky the 2017 general election was for the Conservatives. Nothing is certain. By all means construct a narrative but be aware that the story is quite likely to be rewritten.
If on the other hand you’re looking for the best bets to place, you have to draw conclusions on skimpy information. Often it is better to be wrong quickly than right slowly. So long as you’re right more often than the odds would suggest or get the opportunity to trade out at a profit later on, the failures won’t matter, though they might be embarrassing (I speak from repeated experience on this). If the electorate is more footloose and polling remains of doubtful accuracy, the value is usually going to be found in the long shots. Equally, if you spot a narrative that looks set to gain currency, selling into that narrative will often be profitable.
Of course, you might well not be right. But so long as you’re keenly aware that there’s a high chance you’re wrong, you can make the appropriate adjustments later on. And if all goes wrong, at least you will have a ready answer to the question: “how could I have known?”