Por qué se equivocaron tanto los encuestadores en UK. ( Ingés) /How did the pollsters get the UK general election so wrong?

La incidencia: Las encuestas se equivocan , y las tenemos como infalibles.

Fuente: NewScientist, 08 May 2015 by Jacob Aron

There were three losers as the Conservative party swept to victory in the UK's general election last night: Labour, the Liberal Democrats and pollsters. The two other main parties suffered heavy losses at the ballot box, but polling companies also have egg on their face after months of suggesting a Conservative majority win was virtually impossible – the very result we see this morning.

Only one poll got it right: the exit poll announced immediately after voting ended at 10 pm last night. Pundits were shocked to see the Conservatives with a strong lead over Labour, while John Curtice of the University of Strathclyde, Glasgow, who oversees the exit poll analysis, looked a little nervous as he unveiled it on live TV.

Curtice and his team have now been vindicated by a result that closely matches their prediction, but why was a single exit poll so much more accurate than the hundreds of opinion polls that preceded the election?

"They are completely different kinds of animals," says Roger Mortimore of polling firm Ipsos MORI. The exit poll involved interviewing people outside a polling station immediately after they have cast their vote, and using that information to forecast how each of the UK's 650 constituencies will sway

Pre-election opinion polls canvas people via phone and the internet to find out which party they support and whether they intend to vote. Pollsters then predict the share of the votes each party will get.

Ipsos MORI conducts telephone polling by calling numbers at random and then weighting the results according to the demographics of those canvassed to be representative of the UK as a whole. However, the intricacies of the UK's political system make it very difficult to translate a projected share of votes to the number of seats in Parliament a party would win with those votes.

"The relationship between seats and votes is always a bit arbitrary," says Mortimore. Matters are complicated further by the fact that people might change their mind once they have a ballot paper in hand, or may not bother going to vote at all.

The firm also did the fieldwork for the exit poll, gathering data from 140 polling stations across the UK. Mortimore says Ipsos MORI tends to visit the same polling stations at each election, so their researchers can see if anything changes. That then feeds into a computer model that effectively has a number of toggles for each constituency, such as whether it is rural or urban, and attempts to identify types of seats that will experience the same electoral swing.

"In the exit poll it's not just a matter of adding up all the figures and reporting the total. We're trying to look for different things that are happening," says Mortimore.

In other words, more sophisticated modelling plus hard data on actual votes, rather than simple intention to vote, make all the difference. Pollsters are now looking at what went wrong with their pre-election predictions, in the hope of doing better next time. "A terrible night for us pollsters. I apologise for a poor performance. We need to find out why," YouGov CEO Stephan Shakespeare tweeted this morning.

One possibility is an overestimation of the turnout. Mortimore says many firms had been expecting an unusually large number of people to go to the polls, but early information suggests the real figure is not much higher than previous elections. "That may well be part of the explanation for the gap between the results and the polls: that some people have said they are going to go out and vote Labour but haven't done so," says Mortimore.

But maybe we shouldn't be too hard on the pollsters. According to the BBC, the Conservatives are on course for a 37 per cent share of the vote versus Labour's 31 per cent. With many pre-election polls pegging both parties at 34 per cent, that is within the typical 3 per cent margin of error. Perhaps this time polling firms just got extremely unlucky.