Friday, April 27, 2007

Poll bias in the UK

I have found a useful data set which lists UK polls over the past two years. It also lists the commissioning newspapers or channels and the pollsters. Many pollsters work with the same newspapers and that’s why I concentrate on former. I estimate the normalised lead of Tories over Labour by controlling for a time trend (exceldate in the table) and a switch in the lead in December 2005. By normalisation I control for other parties, which means I divide by the sum of votes for Tories, Labour and LibDems.

I find that BPIX over-estimates the lead by 3 percentage points, ICM and YouGov over-estimate by 2 percentage points (see table). The R2 is surprisingly good with 74%, keeping in mind that we don’t control for political events. This can also be read in the way that all the other pollsters (eg Populus) under-estimate the lead.

1 comment:

Anonymous said...

lets hope it true for the f'ing scottish nationalists on thursday