Leighton Vaughan Williams / 07 October 2008 Free Bet

A few weeks ago I wrote about the importance of what candidates look like as a measure of their likelihood of getting elected. The conclusion of that piece was that appearances matter a lot. One might hope, of course, that policies and ability to handle the issues count for more, but is it true? One way of testing this is to examine the extent to which knowledge about the relationship between voters' and candidates' positions is useful in forecasting the outcome of elections.

In a fascinating recent study, entitled 'Forecasting Elections from Voters' Perceptions of Candidates' Positions on Issues and Policies', Andreas Graefe and J. Scott Armstrong address this question. After eight years of George W. Bush their (perhaps surprising) conclusion is that American voters have a history of choosing the candidate they expected to do the best job in dealing with the issues facing the country.

The approach they adopt, called PollyIssues, is based on voters' perceptions of how well the candidates would deal with issues facing the country if they were to be elected. Graefe and Armstrong tested their approach for the nine US Presidential Elections from 1972 to 2004, analysing data from 315 historical polls. In seven out of nine cases, they correctly predicted the winner of the popular vote.

Moreover, for the last three elections, from 1996 to 2004, they provide evidence that PollyIssues was more accurate than four other well-established election forecasting models. For the 2008 election they are also using an approach known as PollyPolicies. This is based on voters' preference for policies and their perceptions of which policies the candidates are likely to pursue.

How valuable are these approaches likely to be in improving our estimate of the likely result of this year's US Presidential election? Might we not be as well to look at the polls? If we do, we see that Obama is currently over six points up in the RealClearPolitics polling average, and more if polls arbitrarily excluded by RCP are included (this is another story, and election-watchers are advised to turn to Pollster.com for an impartial and complete polling guide).

To some extent, yes we should look at polls - but it mustn't be forgotten that these are a snapshot in time rather than a crystal ball. Betting markets, on the other hand, do look into the future and the question in this case is whether the findings of the new study are properly assimilated into these markets.

Whatever our conclusions, it is clear that the Graefe and Armstrong study does offer us access to extra pieces of the jigsaw.

So, who will win the U.S. presidential election if voters decide based on which candidate they expect to do better at handling the issues? For 2008, their model predicts the Democratic candidate, Barack Obama, as the election winner, with 50.7% of the two-party vote. And what about PollyPolicies? It also predicts Obama to win the popular vote.

In four weeks' time, we shall see whether the models are right. Much more importantly, we shall also see whether Americans really do vote for the best President.