Sky Views: Polls are good for democratic debate

Adam Boulton, Editor At Large

Britain's national opinion polls during the campaigns did not provide a useful guide to the outcomes of either the 2015 General Election or the 2016 EU referendum.

Privately the polling companies acknowledge this. Why else would they have held several enquiries into what went wrong? And why else are all the pollsters individually substantially altering their methods of obtaining a result?

If the polls are useless you may be puzzled that last Sunday's papers were packed with them, projecting the outcome of the General Election this Thursday.

There are a couple of fairly shabby reasons for this.

Firstly a "shock poll" pointing to an unnecessary outcome generates a good story.

Secondly there is push-polling, hoping that the news being reported may help shift opinion. Newspapers take sides at election times. It is often argued that prominent coverage of "knife-edge" polls can be targeted to encourage worried voters to go and vote in the favoured direction.

It is increasingly difficult to use the background of people to predict how they are going to vote. Class, sex, education, race and family ties matter less and less.

Overwhelmingly though, I think that the national newspapers are sincere in their desire to inform their readers. Opinion polls are the best guide we have to the public mood and how an election might go. It's just a pity that they are not better at it.

This election Paste BN has deliberately downgraded use of national voting intention opinion polls because of their questionable value. We don't base news stories around them and we are not reporting them constantly with an on-screen ticker. The other main broadcasters have done much the same.

But we still report what the newspapers are saying - so we too should ask ourselves hard questions about what the polls are for.

My own view - and I'm not a statistician - is that it's not just the methodology but the model of national opinion polls which is broken. Taking a "representative sample" of a few thousand and using it to project how more than 40 million people might or might not vote doesn't work. To use the cliché, it is "analogue in a digital age".

While pollsters are still struggling to build "representative samples", the truly powerful - Google, Facebook and those waging the "invisible election online" - are using "big data" to research voters directly, rather than focusing on people who might be like them.

In almost every way people are less predictable than they used to be. For a start we don't know who is going to vote.

The current gap between the best and the worst polls for Labour, largely depends on how many of the 18-25s different pollsters think are going to vote.

It is increasingly difficult to use the background of people to predict how they are going to vote. Class, sex, education, race and family ties matter less and less. In Britain the EU referendum was a seismic shock which brought people who have never voted before to the polls and which split families and friends into opposite camps. Those bonds have not been restored.

We also know that a country does not work as a unit, either in the way voting patterns swing or in the way that interprets into an election result. In 2015, there was actually a slight swing overall to Labour but the Conservatives won a majority of MPs for two reasons: The Tories squeezed out the Lib Dems especially in South West England and the SNP crushed Labour in Scotland.

On top of all that, a growing proportion of the electorate is making up its mind at the last moment how to vote. If lots of minds change a "snapshot" mid campaign may not be worth very much.

Funnily enough the most controversial poll of this campaign - YouGov's "survey" in The Times, pointing to a hung parliament, is an attempt to address many of these weaknesses. It takes a much bigger sample and tries to breakdown the electorate into sections. But it is still very much a work in progress.

A similar model in last year's US Presidential Election got it right that Hillary Clinton would get more votes than Donald Trump (the popular vote) but it failed to pick up that he would be elected on a state by state basis by the electoral college.

So what are polls good for in my view?

They are information which should be made public so all people can judge for themselves. We already have problems with "the invisible election", the rich and powerful conducting their own private research and not telling us. Polls are good for democratic debate. I don't support a ban.

Taken together, polls probably give us a very broad picture of the public mood. The first 10 polls at the start of this campaign showed that the Conservatives had an aggregate lead above 20% and that Mrs May was massively popular. The last 10 polls put the Tory lead at less than 10% with big drops in the leader's ratings.

That probably tells a story. So does the fact that the Conservatives have been well ahead in every single poll this campaign.

But that's about it.

Please don't confuse regular opinion polls with the very different broadcasters' exit poll. The exit poll is based on many times the number of interviews, face-to-face not on the phone or online, carried on the same day as the vote, and in a large number of carefully chosen locations. In the past three elections it has a proven record of reliability.

So please join me on Paste BN at 10pm sharp when polls close on Thursday when I will announce the only poll that will start to unravel the mystery of Vote 2017 - the exit poll.

Sky Views is a series of comment pieces by Paste BN editors and correspondents, published every morning.

Previously on Sky Views: Stuart Ramsay - UK terror attacks should not be a surprise