Predicting election results from the latest opinion poll figures (or from other estimates from party support) is normally done using a uniform swing calculator. These work on the assumption that the change in each party’s vote is largely uniform across the country and applies the changes in the national shares of the vote to each constitutency. For example, if the Liberal Democrats were down 2 percentage points on the last election UNS assumes that their vote falls by 2 percentage points in each individual constituency. The winner of each seat after the UNS is determined and then totally to give a result.
Unfortunately, while UNS is an easy method of predicting election results, it doesn’t actually bear a huge amount of resemblance to reality. There is, as it happens, a good paper by John Curtice on the British Polling Council’s website about translating poll figures into projected results in the House of Commons. Curtice suggests a number of reasons why uniform swing calculations don’t work:
- Firstly there may be different swings in different “battlegrounds”, probably partially a result a tactical voting. This was certainly the case at the last election – the Liberal Democrats, for example, did far better in Labour held seats where they were in second place, getting an average increase of 7.8 points in their share of the vote. In contrast, in seats where Lib Dems were second to the Conservatives they only increased their vote by 0.5 percentage points. There were smaller, but still significant, differences for other parties. The Conservatives advanced or stayed the same in Labour vs Conservatives seats and Conservative vs Lib Dem seats, but where the Lib Dems were the main challengers to Labour the Tories went backwards. The Labour vote fell far more in seats where it was in first or second place than when it was in third place (primarily, one assumes, because it didn’t have so far to fall if it was already in third place!).
- Secondly there is marginality. As it happens the Conservative-Labour swing in marginals was much the same as it was in safe seats in 2005, but the changes in the parties votes was different – both the Conservatives and Labour did better in their key marginals than elsewhere, it’s just their mutual improved performances cancelled each other out! Common sense alone tells us that marginality should have some effect upon the size of the swing – if there was a 10% swing at the election, it is unlikely that the Conservatives would actually pick up an extra 10% of the vote in places like inner-city Glasgow and the Welsh valleys, while there are some home counties constituencies where Labour barely has 10% to lose.
- Thirdly there are regional differences. The most obvious case is Scotland, whose swing at elections has often contrasted strongly with the rest of the UK. The most notable case was 1992, when there was a 2.6 swing to Labour in England and a 2.5 in the opposite direction in Scotland. At the last election, there was a noticable larger swing to the Conservatives in London and the surrounding areas than elsewhere in the country.
- On top of all of these systemic differences, there are also the simple variations in swing between one seat and the next, be they a result of candidates, campaign organisation, demographic drift or whatever other reason. In theory, these should cancel out – the seats with above average swings should result in as many changes as those with below average swings. In practice though it doesn’t work that neatly, because seats are unevenly distributed in terms of marginality. To give an example, at the next election there will be about 9 Lab/Con seats where the Tories need a swing of between 3% and 4% to gain the seat, and about 18 Lab/Con seats that need a swing of 4% to 5% for the Tories to gain it. If there was a swing of 4% at the next election, then because of the way the seats are distributed, then the number of potential gains the Tories don’t get in that 3-4% range due to below average swings is likely to be outweighed by the number of seats they do get in the 4%-5% range because of above average swings.
Curtice and his team took the factors above into account when making the projected result from the exit polls at the 2005 election – and accurately predicted a Labour majority of 66. To some degree, I expect luck was a contributory factor in making the prediction quite so spot on – after all, no model could predict exceptional seats like Bethnal Green & Bow and Blaneau Gwent – but it was an impressive projection all the same. Alas, on normal opinion polls we don’t have the data from which to make such predictions. Polls are not broken down according to the type of seat in which respondents live in, and the regions on pollsters published tables tend to be rather vaguely defined. Even if the relevant splits were published, the sample sizes in any single poll would be far too small to draw any meaningful conclusions.
In short, projections using a uniform swing aren’t particularly useful and, while it does seem to be possible to make more accurate projections, you’d need better data than is currently available from newspaper opinion polls. That said, we live in the real world and I know what everyone loves to play Peter Snow and project what an opinion poll might mean in terms of seats won. Therefore, below is a very simple swing calculator based on a uniform swing and my notional figures for the new Parliamentary boundaries. Just remember that, as Peter Snow used to say, it’s just a bit of fun.