About that UKIP number (continued)

Earlier this week I put a quick article on the Opinium website just explaining why we think our UKIP number is frequently higher than others. The link is here but the basic idea was that it's down to the fact that we don't do any sort of political weighting beyond whether people are likely to vote.

Just to explain, weighing means counting some people in a survey as slightly more or slightly less than one so that the overall demographics (i.e. the proportion of men vs. women or the percentage of the sample coming from each region) match up with those of the country as a whole.

Opinium figures are weighted to national demographics (that we use in all our commercial polling work as well) and then filtered to those who said they would be likely to vote and told us which party they would vote for.

Filtering to those likely to vote is our way of making sure that we aren't giving prominence to the views of people who aren't going to vote anyway but most other companies also apply a form of political weighting to make sure of this. This either takes the form of asking people how they voted in 2010 and making sure it matches the proportions of that election or giving more or less weight to peoples' answers depending on whether they voted before or not.

The other thing is that, like most companies I think, to see UKIP as an option you have to have said you'd vote for "some other party". In theory surely we'd get the most accurate results by showing the options as they appear on a ballot paper (like Survation do) but the argument against this is that presenting the other parties equally gives them a degree of false prominence given that an actual election will come at the end of a campaign dominated by coverage of the bigger parties. We can't predict what the 2015 campaign will look like exactly but it seems unlikely that UKIP will get as much coverage as they did for the last few weeks of 2012.

Anyway, our approach has worked well in the past and we're confident it's accurate now but we are experimenting with a number of different methods, particularly with forms of political weighting and turnout filtering, to see where we can improve it in the future.

Regional figures don't matter!

On a side note, there was some cheering in UKIP circles at the news that the purple party apparently pulled ahead of the Conservatives in the North West. I haven't checked the exact poll in question but it's worth pointing out that regional splits in national polls are best ignored, particularly for political ones. The sample size is much smaller and the margin of error therefore much higher.
The sample is also designed to match the country's demographics at a national level but may not match those of each region if that makes sense. Put more simply, we make sure that there are the right proportion of, say, 35-54 year olds in the sample but it's perfectly possibly that there will be too many of these in one region and too few in another. The only reason regions are included are because it's standard practise but the figures within them are indicative only and far more prone to being outliers.

UPDATE: I've checked and UKIP and the Conservatives both have 28 people selecting them in the North West. UKIP are on 18% and the Conservatives 17%. This is difference is due to weighting and certainly not significant in any way.