A note on methodology
A note on methodology
The first is that we will be publishing a poll every week, releasing on Saturday evening, until polling day.
The second is that we?re adding some new crossbreaks to our regular tables, how respondents say they voted in the 2010 election and some geographical breaks. In the past we stopped publishing regional crossbreaks because they are too small to be reliable and were easy targets for being taken out of context. However, due to the changes in Scotland since the referendum and the importance of seeing how the constituent nations of the UK compare, we will start including columns for England, Scotland and Wales. While the caveats about smaller base sizes will still apply to Scotland and Wales, there are enough voters from England to make such a comparison worthwhile.
Third is a fairly minor weighting change which we have been meaning to do for a while. This is to split out the younger age group of 18-34 year olds into 18-24 year olds and 25-34 year olds. Our previous age group would over represent 25-34 year olds at the expense of their younger peers so this change is intended to remove that effect.
Fourth, and most significantly, is the introduction of political weighting using a variable called party propensity.
Party propensity is an evolution of the party-ID variable used in previous elections where respondents were asked if they identified with a particular party and the sample was then weighted according to pre-defined targets coming from sources such as the British Election Study.
With party propensity, we ask voters how they feel about all of the parties and, from their answers, put together a unique picture of each voter?s view of the whole spectrum of parties. From this we can divide the electorate into groups based on which combination of parties they would or would not consider voting for.
To decide how big or small each of these segments should be in our sample we have tried to make the process of creating weighting targets as organic as possible. This is based on a combination of how respondents voted in the European Parliament elections last year and a rolling average of the results of this segmentation during our trials over the last few months.
Taking January 2015 as our base therefore, the targets are updated week to week by a rolling average of recent polls. This makes sure that, while our samples will remain politically balanced, they are still able to take account of actual changes in the electorate?s underlying views. If party X suddenly catches fire or completely nosedives then we won?t be hamstrung in reporting this by how popular they were in December.
The effect of these changes has been to remove some of the variation and statistical noise we have seen between polls. It also has the effect of slightly upweighting the Conservatives and downweighting UKIP while leaving Labour, the Liberal Democrats and the Greens broadly unchanged.
In the past we have generally decided against applying political weighting to our polls as it makes it more difficult to account for changes in circumstances. That made sense for mid-term when things were in flux so that we could take account of things like the rise of UKIP and the Greens. But with three months until polling day, the circumstances of the election should be clear enough that we can harden our assumptions a little while our rolling average should allow us to respond to any changes on the ground.
For any more information please contact the political team at firstname.lastname@example.org where we?d be happy to answer any questions you might have.