— Mike Smithson (@MSmithsonPB) December 20, 2013
Using a 2010 model might not be right for 2014
The publication last week of the latest round of the Bown-funded constituency polling set off a debate about methodology with efforts to attack what Survation had done. The main objections were that the firm wasn’t using what have become standard approaches to ensure politically balanced samples.
This was followed by the remarkable defence published on PB by the firm’s research director, Patrick Brione.
- I’m coming to a view that weighting by party ID or past vote might be giving us a false picture in the new world of 4 party politics.
Just look in the screen grab above from the latest Populus survey to watch this in operation. The firm asks what party participants identify with and then uses 2010 data to “model” the electorate. Because far fewer people then identified with UKIP the firm has to scale back responses from those saying they identity with UKIP to about a tenth of current responses.
YouGov has a similar but less harsh approach.
Past vote weighting in which people are asked how they voted at GE2010 could be having a similar effect. A by-product of the UKIP surge is that many more people are “remembering” that they voted for the purples in 2010 than the 3.1% that actually did so. These responses are then scaled back according to formulas linked to the 2010 result.
I haven’t got a solution but at the moment I think that many polls are understating UKIP.