US Election Daily Update: New study suggests Obama may be a long way ahead in polls

They told me my lead was at least this!” Mexico’s President-Elect, Enrique Pea Nieto, was thought to hold double-digit leads in July’s election.

Jack Houghton asks how useful polls will be in helping us predict the winner of this Presidential election, and explains that a new study has shown that focusing on a different type of polling question may be the way forward.

When it comes to predicting the outcome of elections, opinion polls have long been the best tool we've had available, but the reliability of polls have consistently been called into question, and, as punters in this Presidential election, we have to ask ourselves: how much should we take notice of them?

In Mexico's Presidential elections this July, Institutional Revolutionary Party candidate Enrique Pea Nieto was considered to hold a double-digit lead by all the major polling organisations covering the election, with three polls showing him to hold an average lead of over 15 points in the last week of campaigning. Nieto won the election, and is now Mexico's President-Elect, but his eventual winning margin was less than six points: hardly the landslide the pollsters were predicting.

Now, it's easy to brush this aside as irrelevant to the US Presidential Election: Mexico presumably doesn't have the same regulation of opinion polls that they have in the US; knowing their candidate was certain of victory, perhaps many of Nieto's supporters were too apathetic to vote; or maybe there are social and economic reasons in Mexico, such as the relative paucity of telephones, that means their polling is never going to be especially accurate.

Perhaps, but it's not as if Western democracies are without their examples of polling error. In the US, the most famous example is probably that of the Literary Digest poll taken before the 1936 election, which predicted a landslide victory for Alfred Landon against Franklin Roosevelt. Those designing the poll, which sampled telephone and car owners, didn't realise that Roosevelt voters were unlikely to be present in those groups in any great number. Roosevelt went on to collect over 60 per cent of the public vote.

And it's not just inexpert literary publications that have got it wrong: Gallup got the 1948 US Presidential Election drastically wrong, and, in the UK, every major polling organisation failed to predict the correct outcome of the 1992 General Election, which led to an inquiry into their techniques and a subsequent change in all of their methodologies. These examples, of course, are (in)famous because the final result was predicted incorrectly, but, increasingly, although getting the bare result right, polling companies are consistently weak at predicting the margin of that victory.

In every UK General Election since 1992, for example, pollsters have tended to overestimate Labour party popularity, and, in the US, looking at data going back to 1952, state polls have only predicted the winner of the Presidential vote in that state 69 per cent of the time, which will make those who have backed Obama at 1.282/7 to be the next President, on the back of a small lead in some key swing states, a little nervous.

Part of the issue for pollsters in the US is that, only 15 years ago, around 40 per cent of those asked were prepared to be surveyed. Nowadays, that number is closer to 10 per cent, which means there is a danger that polls tend to reflect only a vociferous minority, rather than the population at large.

There might still be hope for the pollsters, though. A recent study has shown that if, instead of asking voters which candidate they support, pollsters ask them which candidate they think will win, the predictive power of polls increases dramatically. Rather than the 69 per cent success rate outlined above, respondents now correctly predict their state vote for President 81 per cent of the time.

So how does this translate to this election? Well, when asked to predict the outcome of the race to be President in 2012, national respondents have typically shown anywhere from a 13 to 24 per cent lead for Obama, which would suggest he is going to dominate the Electoral College. That would be good news for those supporting the recent recommendations of Paul Krishnamurty, and less good news for those who have followed my Obama back-to-lay strategy.

Still, who's to say those poll respondents aren't just predicting an Obama win because that what the polls are telling them? By the time we're having our breakfast on Wednesday morning, we'll have some kind of answer.