The Betfair Prof / Leighton Vaughan Williams / 17 October 2008 /

What are the best sources for predicting presidential elections? Are polls saying the same thing as Betfair's prediction market? The Betfair Prof tells all...

There are a variety of methodologies which claim pre-eminence in the election forecasting stakes. The most traditional of these is the opinion poll, albeit it is a tradition somewhat chequered in terms of accuracy. For example, Rasmussen Reports, which performed creditably in the 2004 Presidential election, had George W. Bush up by nine points against Al Gore on polling day, 2000. In the event, Gore won the popular vote with more than a little to spare (though not the electoral vote, after the US Supreme Court voted 5 to 4 to stop the count in Florida).

In e-mail correspondence with Scott Rasmussen, he explained the 2000 result as a consequence of a flawed methodology, which used an automated telephone interview system, skewed towards likely voters and unweighted by expressed party political preference. Since then, he argues, the methodology of the Rasmussen poll has changed, in part to adjust for party weight and to reduce the tightness of the screening for likely voters. Should we believe that Rasmussen 2008 will perform more like Rasmussen 2004 than Rasmussen 2000? We shall see. In anticipation, it's perhaps better to simply take all the professional polls and take the average of the findings.

This is what RealClearPolitics (www.realclearpolitics.com) does, or claims to do. In fact, the site is selective in which polls they choose to include, which casts some doubt on the results, made more worrying by the fact that they simply take an unweighted average of the results of their selected polls, regardless of the historical accuracy of the polls or the methodology applied or even the sample size. Is this likely to provide a better window on popular opinion than simply taking the poll or polls which have performed best in the past, or the polls with the biggest sample sizes? This is an important question but not one which concerns RealClearPolitics.

The other problem with polls is that they are simply a snapshot of public opinion at a point in time. More sophisticated forecasting models assign different weights to different polls according to their record of success in pinpointing the actual results, as well as allowing for the time to elapse to the real poll and a range of demographic variables.

Arguably the best of these is available at www.fivethirtyeight.com, a fascinating and highly recommended site, run by Nate Silver, which revels in its claim of 'Electoral Projections Done Right.' Currently the site gives Barack Obama a better chance of winning the election than does Betfair. Perhaps this is because Betfair clients are attributing a higher weight to some sort of unexpected shock than does the polling analysis. Perhaps Betfair clients simply aren't aware of Silver's analysis or have other problems with it.

There is, by the way, another interesting analysis of polling information, run by Sam Wang and the Princeton Election Consortium (https://election.princeton.edu/). I advise checking both resources out. Who knows? You may even be led to the conclusion that one or more of the available Betfair prices are simply a little too generous to leave alone!

Professor Leighton Vaughan Williams is the Director of the Political Forecasting Unit and Betting Research Unit of Nottingham Business School, Nottingham Trent University, and Editor of the Journal of Prediction Markets

Betfair Politics Zone

'.$sign_up['title'].'

'; } } ?>

:

:

:

?

: