What are his chances this week? Photo: Ben Birchall/PA Wire/PA Images

It is tempting, but not quite true, to say that there are lies, damned lies and opinion polls. Polls, like other statistics, are often vital if we are to understand what is going on. But when they are twisted, then danger lurks.

With this week’s elections to the European Parliament, and the wider divisions in Britain over Brexit, high quality data and rigorous analysis are especially vital. The weekend polls provided examples of both good data and terrible analysis. They generated three near-certainties, one important doubt—and two giant whopping errors, one of them unwittingly amplified by Andrew Marr on his Sunday morning show.

Let’s start with the whoppers. The Sunday Expressfront page headline proclaimed: BRITAIN SAYS “LET FARAGE LEAD BREXIT.” The poll cited by the Expresssaid no such thing. ComRes asked people which of six party leaders would best at doing each of eight things. One of these was “leading the UK out of the Brexit crisis.” Farage came top with 28 per cent. However, the other five leaders together scored 72 per cent. More than twice as many people rejected Farage as backed him. But rather than acknowledge the fairly obvious truth, that public opinion is now badly fractured, the headline misled readers.

That is not all. The paper also reported: “The ComRes survey has also shown strong backing for no deal, with 63 per cent saying that the UK should just leave with no agreement if parliament rejects Theresa May’s deal again.”

This was picked up by Marr in his interview with Rory Stewart. Marr put it to the international development secretary: “A big, substantial majority of British voters now want no deal.”

That is simply not true. ComRes presented respondents with nine statements, and asked people whether they agreed or disagreed with each. One of these statements was: “If parliament will not agree to the Withdrawal Agreement on the table, then parliament has to accept No Deal as the consequence.”

The responses were: Agree 47 per cent, disagree 27 per cent, don’t know 25 per cent. The paper arrives at 63 per cent by ignoring the large number of don’t knows. To be sure, it is normal to exclude don’t knows when reporting voting intention, for the central purpose of such a question is to explore the distribution…

4151254585df045f14aa6b2.90791808