Home About Chat Users Issues Party Candidates Polling Firms Media News Polls Calendar Key Races United States President Senate House Governors International

New User Account
"A historical political resource." 
Email: Password:

  ICM Research
Polls Conducted52
Average Error4.02536106
Partisan BiasAverage amount polls are biased compared to actual outcome by party.
ContributorEasily Offended Man
Last ModifiedUser 13 April 06, 2005 10:44pm
DescriptionThe other longstanding player in the market, ICM were at the forefront of most of the innovations introduced after 1992, briefly trying out such innovations as mock ballot papers before abandoning face-to-face interviewing in favour of quasi-random phone sampling. They were also the forerunners in adjusting for the spiral of silence, an adjustment which led to them having the most accurate predictions in 2001 and 1997. This adjustment originally favoured the Conservative party and throughout much of the 1992-1997 period ICM's polls were notably more favourable to the Conservatives than other pollsters.

Methodology: ICM use quasi-random selection to interview people on the telephone. Numbers are randomly selected from the telephone directory, with the last digit changed to a random number, so as to ensure that ex-directory numbers are not excluded. ICM then use quota sampling to decide which member of a household to actually interview (hence the system is not truly random). Normal sample size is 1,000.

ICM's results are weighed by normal demographics like age, sex and social class. To ensure a politically representative sample, they also weight samples by past vote. To take account of false memory syndrome, ICM weight to a point midway between the actual vote at the last election and the average recalled past vote in their last 20 polls.

ICM do not filter by likelihood to vote, but do weight by it, giving people who say they are unlikely to vote less weight than those who say they are certain. This takes into account not only how likely people say they are to vote at the next election, but how often they claim to hav voted in the past. Finally ICM were the pioneers of adjusting for the "spiral of silence" and continue to reallocate 50% of "don't knows" and refusals to the party they voted for at the previous election.

Wheras the adjustment for the "spiral of silence" used to favour the Conservatives, it now works in favour of the Labour party. Taking a relatively lax approach to turnout also tends to favour Labour and the Liberal Democrats more than the Conservative party. Looking at the tables published by ICM, weighting by past vote would appear to increase the number of Lib Dem and Conservative voters in samples. The combination of these factors results in ICM tending to report the highest Labour leads of the main pollsters (as of September 2004 they are the only one the main four never to have reported a Conservative lead in 2004).

Date Category Headline Article Contributor