Labour's private polls showed the Tories ahead as early as Christmas 2014

Swing voters were scared by Tory warnings of SNP involvement in government

Ed Miliband delivers his resignation speech having decided to step down immediately (AFP/Getty Images)

Private polling conducted for the Labour party in the run up to the general election showed the Conservatives in the lead as early as last year, a pollster has said.

James Morris, Ed Miliband's chief private pollster, said it was a “great mystery” why newspaper opinion polls consistently showed better results for Labour than internal polling did.

“While the public polls had Labour ahead until early spring of this year, in the party's internal polls cross-over came right after conference season in 2014,” Mr Morris wrote in an article for the New Statesman.

“A four point Labour lead in early September turned into a tie in October, followed by small Tory leads; prompting the party to put reassurance on fiscal policy and immigration at the heart of the campaign launch before Christmas.”

The pollster said Labour’s reactions to the findings was initially successful in pulling back voters in marginal seats on the back of Ed Miliband’s strong performance in debates and the row over non-doms.

A final poll in April, however, showed Conservative warnings that Labour could be reliant on SNP support had convinced the tiny number of swing voters in key marginal who would decide the election to support the Conservatives.

“The campaign strongly toughened our stance on the SNP before the final Question Time, but it was not enough. The Tories successfully used the fear of Scottish influence as a way of catalysing pre-existing doubts about Labour in a way that had not been possible earlier in the campaign,” he wrote.

Mr Morris suggested that differences in polling might be due to questionnaire design.

The British Polling Council, the professional body which governs pollsters, has launched a full inquiry into why most public polling at the general election turned out to be inaccurate.