ICM’s raw data had Labour nearly 20% ahead

An extraordinary insight into the challenge that pollsters have in finding representative samples is revealed in the detailed data for the final surveys which is just being made available.

We all know now that in terms of the popular vote in Great Britain Labour beat the Conservatives by 36.2% to 33.2% – or about eleven Labour voters for each ten Conservative ones. But base data for the final surveys ICM and Populus showed huge Labour leads which were simply not borne out by what happened.

If the ICM and Populus polls had not been adjusted for past vote recall they would have shown a landslide much bigger than 1997

The ICM sample produced LAB 335 CON 193 LD 180 of people saying they were voting for the respective parties. Populus managed a slightly less skewed raw response of LAB 437: CON 292: LD 218.

When they asked how respondents had voted four years earlier both pollsters found LAB:CON ratios of between 2:1 and 5:2 which underlined how unrepresentative their samples were.

It is here that we are totally reliant on the skill and expertise of the pollsters to process the information so that we have meaningful figures.

Thus the survey data was adjusted, amongst other factors, to reflect the past vote recall – but not fully. The pollsters assume here that more people are likely to say they votes last time than actually did so they give a greater weighting to Labour and a reduced weighting to Tories. If they had simply weighted on the 2001 result then they would have overstated the Tory position.

In the end the headline figures published by both pollsters had Labour leads of 6% – twice what actually happened but pretty good in view of the raw data on which it was based.

In a discussion on UK Polling Report the head of Populus, Andrew Cooper, has spoken of the particular difficulties of carrying out surveys during bank holiday weekends. Could it be that Conservatives are much less likely to be at home on public holidays than Labour supporters?

Mike Smithson