Original URL: http://www.theregister.co.uk/2009/07/27/resolver_election_predictions/

Tory landslide inevitable - maybe

It all depends which (software) tool you listen to

By Jane Fae Ozimek

Posted in Government, 27th July 2009 12:44 GMT

Get ready for a Tory landslide. That is the prediction of a new and sophisticated electoral forecasting tool, created by business modelling company, Resolver Systems.

One recent poll places Tories on 39.5 per cent, Labour on 28 per cent and Lib Dems on 21 per cent. Traditional forecasters predict results ranging from a Conservative majority of 26 (Electoral Calculus) to No Overall Control (Hill & Knowlton). The Resolver model sticks its neck way out in front, with a prediction of a 150 seat Tory majority.

The difference can be explained by the fact that the Resolver model does not apply what electoral analysts refer to as "Uniform Swing".

Robert Smithson, Chief Executive of Resolver Systems explains: "If you go back to the 1950s you could do simple two-way swing analysis because the Conservatives and Labour had 95 per cent of the vote.

"Now the two parties have 65 per cent that makes traditional forecasting pretty inaccurate."

Swing was originally defined as the average of one party’s per centage gain and the other party’s per centage loss between two elections. "Uniform swing" embodies the idea that votes move between parties uniformly across the nation: a swing of five per cent from Labour to Tory would see all Labour seats down five per cent and all Tory seats up five per cent.

In fact, this is a straw man position that has been knocked down so many times as to be scarcely worth tilting at any more.

The distinguished political analyst John Curtice published a paper (pdf) following the 2005 General Election, setting out the problems with uniform swing: these include the effect of different battlegrounds (Liberal vs. Labour seats react differently to Labour vs. Tory seats); marginality (marginal seats poll differently); regionality; and local effects due to individual candidates.

The question therefore becomes the much harder one: what factors need to be taken into account to forecast election outcomes accurately?

The two factors that have historically caused most confusion have been sampling error and sampling bias. A poll carried out on 1,000 adults – as many regularly have been – is subject to sampling error of plus or minus 3%. Bearing in mind that the UK uses a first-past-the-post system of voting, it means that in a close-run election, accurate prediction of individual seats is a nightmare – and allows for very wide differences in the real world result.

Bias creeps in when some underlying variable that affects voter behaviour has not been taken into account. One of the most famous examples of this surfaced in a Literary Digest poll in 1936, which predicted a landslide victory in the US presidential elections for a certain Alf Landon.

Unfortunately for the Literary Digest, 1936 was one of the first US elections to divide more extensively along class and affluence lines. Their methodology – a telephone poll – was hopelessly skewed towards the affluent, and therefore failed to spot the wider groundswell of support for Franklin D Roosevelt.

Whilst most polling companies now have a good handle on these factors – the "known unknowns" - there is much debate over how to handle other factors such as respondents who won’t declare an intention, and differential voting by party supporters. When a party is nationally unpopular, it seems that people are less likely to own up to supporting them (under-stating) – and at the same time, those who do declare support are less likely to come out and vote (over-stating).

What the Resolver model does is address these issues, using data from polling company ICM to allocate these categories. ICM has historically been more focused on the issue of intention allocation – and has tended, of late, to be more accurate in its forecasts.

But at base, this is an argument about the "unknown unknowns". The experts agree that it is impossible reliably to infer voting intention from what people tell the pollsters. The bottom line, therefore, is that all forecasting involves a degree of "finger in the air" guesswork about the future.

There are three lines to be crossed as the election results come in some time next year. First is the point when Labour loses its overall majority. Second is the point at which the Tories become the largest party. Third is the point at which the Tories themselves gain an overall majority.

Electoral geography creates a Labour bias within our electoral system. So much so that a single figure lead would not be enough to usher the Tories over the third of these lines. If you further consider that Labour may yet politely ask Gordon Brown to resign in the autumn, then the future remains very foggy indeed. Whether Cameron can cross that third line remains very much in doubt. ®

Bootnote

Robert Smithson is the son of Mike Smithson, who created the widely respected politacalbetting site for those interested in insightful political analysis and gambling on possible political outcomes. He is also a former Goldman Sachs banker.