The election is fast approaching and the polls are fluctuating by a few points every day.
It’s got me thinking about how accurate polls tend to be. Almost every newspaper front page recently has had a poll result on the front, celebrating as if it’s the real result on election night.
I decided to take a look back at times in political history where polls have been way off, to see if we can really trust predictions based on bad data.
Fail #1 – UK GE ’92
This one’s pretty much the benchmark for shock election wins. Throughout the campaign, Neil Kinnock’s Labour party had been polling consistently (if narrowly) ahead.
When the votes were in, John Major’s Conservative party won 65 more seats, getting a healthy majority.
Why did this happen?
Pollsters at the time put it down to the classic ‘shy Tory’ effect. Conservative voters are typically less likely to be truthful about who they’re voting for when talking to pollsters. Possibly because they’re rightly ashamed of being Tories.
Today’s pollsters would do well to learn from 1992. A long time has passed since that election, and it would be easy for pollsters to ignore the effect in this election, especially given that the last parliament has been Tory-dominated – people could be less likely to say that they support the status quo.
To counter this effect, polling companies have turned to automatic phone polls or internet polls to counter the effect – they’ve found that people are more likely to be truthful when they’re not dealing with a human. But internet polls are notoriously easy to game.
It’s hard to predict the shy Tory effect, which is just one reason why you should take modern-day polls with a pinch of salt.
Disaster #2 – Thatcher v SDP/Labour
Election predictions are based on people, and people are very changeable.
In the first years of her time in power, Thatcher saw a steady decline in approval ratings for both her and her government, falling almost to as low as they were during the poll tax days at the end of her premiership.
The newly-formed Social Democratic Party were getting plenty of attention, but ultimately with the Tories polling so low, Labour were sitting pretty. In February 1982, putting your retirement on Labour in the next election would have been a wise investment.
But then, the Falklands happened, Thatcher’s approval ratings went through the roof, and Tories won a huge victory.
Why did this happen?
To be fair to the pollsters and predictors, they couldn’t have seen an Argentinian invasion coming. And sure, there’s probably not going to be a shock win for the UK in any war between now and the election. But massive upsets can happen, and it wasn’t really that long ago when it did happen. Don’t underestimate how quickly the public’s mind can change when trying to critically appraise polls.
Legendary upsets aside, there’s a few essential flaws at the heart of election data and predictions, that can be hard to get past.
Basic problems with some prediction methods
For all the ingenious measures that pollsters resort to, there’s still no foolproof system of predicting an election. There’s plenty of commonly used tools, but they all have their flaws.
Calling people up at random and asking who they’re voting for sounds like a good method – you can catch people unawares, in the comfort of their own homes. But most young people don’t have landlines. I certainly don’t. In relying on phone polls, or basing your stories/bets/predictions on them, you’re basing them on a method that excludes young people or those living in rented accommodation from the very start.
Social media chatter
Predominantly, Twitter is made up of young, liberal people, and a lot of the political chatter on there is driving by London-based journalist types. This isn’t a representative sample of the UK at all. There’s not many 60+ Buckinghamshire Tories big in the Twitter game these days.
This is what caused the Nick effect in 2010 – anyone solely using social media as a way to guess the election result would have assumed Nick Clegg would do amazingly well. In the end, his party got a fairly mediocre result, less than what many pollsters put them at just before the election.
It’s tempting to use Twitter, as a constant stream of its users thoughts, as a way to predict an election result – but it’s just not. Don’t do it.