Predicting Regional Swing

First posted 22 August 2009

The simplest sort of prediction is called the Uniform National Swing which is a rather crude approximation and assumes no regional variation across the country. But using the results of a recent large YouGov poll, we can measure the extent of regional changes of opinion to get a more accruate prediction. This article describes how we do this and what the results are.

1. Regional opinion survey

At the 2005 election, such regional swing accounted for 14 mis-predicted seats.

To make a more accurate prediction we need to estimate the size of these regional swing effects. But this cannot be done from national opinion polls alone. National opinion polls have a sample size of about 1,000 people, equivalent to 100 people per region, which is not enough to estimate opinion accurately. We need an opinion poll which surveys at least 1,000 people in each region of the country. Such polls are infrequent, but fortunately YouGov recently conducted one on behalf of Channel 4 for the European elections (YouGov poll details). The poll of 32,268 people was conducted between 29 May and 4 June 2009, and the headline results were Con 37%, Lab 22%, Lib 19% and Other 22%.

We are very grateful to YouGov for making the regional breakdown of their poll available to Electoral Calculus.

The table below shows (1) the breakdown of support by region as measured at the start of June 2009 by YouGov, (2) regional support as at the general election in May 2005, and (3) the changes from 2005 to 2009 which we call the Regional Swings. (Please see the notes at the end of this article to explain some small numerical discrepancies.)

The national picture is that the Conservatives have gained about 5% support since May 2005, and Labour have lost 14%. The Lib Dems have lost 5% and Other parties have gained 14%.

Regionally, those trends are repeated in that the Conservatives have gained support in every region and Labour has lost support in every region. But although the direction of change is the same, the sizes of the changes are not. For instance, the Conservatives have gained only 2.4% support in the South West, but 12.5% in the North. Similarly Labour have lost only 8.8% in the South West, but 21.9% in the North.

2. Regional swing analysis

We remember that this is not a table of absolute swings. Absolute swings (table 3) show the Conservatives gaining everywhere and Labour losing everywhere. This is a table of swings relative to the national average. So, on average, everything balances out. For example, if the Conservatives perform more strongly than average in the North, they must do worse than average somewhere else, such as the North West. Thus the population-weighted average of each column is zero. (See notes below for why Nat and Others are different).

3. Comparison with recent trends

4. Impact on Predictions

The effect of the regional swing differentials in June 2009 is to give the Conservatives approximately another 30 seats at the expense of Labour. Running a similar calculation at August 2009 using the Regional Predictor showed the Conservatives gaining about 10 seats.

The Regional Predictor now has the added feature to make predictions incorporating the regional swings observed from this large YouGov poll.

Acknowledgements

We would like to express our gratitude to Peter Kellner and YouGov for making details of their regional poll available to us.

Appendix: Numerical Notes

YouGov themselves give some health warnings with regard to the regional support levels:

• During the weighting process of the raw data, there is no control between region and party affiliation. This means that the weighted sum of regional support may not exactly equal the headline national support figures.
• Figures have been adjusted to compensate for the difference between parties' predicted and actual shares of the votes at the European elections in June 2009.
• The "Others" are currently very high, and the LibDems slightly low. This may have changed by the time of the next election.

Additionally, Electoral Calculus gives its own cautions:

• Our definitions of regions may be slightly different from YouGov, so our figures for the general election result by region are different. Nonetheless, the difference of regions compared with the national trend should still be meaningful.
• There isn't a national implied swing for the Nationalist parties (SNP and Plaid Cymru), so their regional swing differentials are just the changes in support since 2005 and do not add up to zero for Nat and Oth.