How come the firms have such different results?

Above is the rather anonymous office block in Bedford which houses the ICM calling centre where, more than likely, the field-work for today’s Populus poll in the Times was carried out.

“ICM calling centre..”, I can hear you saying,”.. surely some mistake?” No – because the process of doing the telephone interviews is something that Populus usually contracts out and ICM are a leader in this field.

There are other similarities as well. The whole architecture used by Populus is very close to ICM. Both use past vote weighting to ensure a politically balanced samples, both use responses to the “certainty to vote” questions to finalise their results and both use the so-called “spiral of silence adjuster” to deal with those who say they are voting but won’t name the party.

So why is it that in the space of five days one firm could be showing a 13 point Tory lead and the other a 6 point one? The answer is that in addition to any Glenrothes effect the pollsters use different forms of question and their mathematical calculations are not the same.

I’ve argued here before that the wording of the ICM question is more favourable to the Lib Dems and this seems to have a greater impact on Labour than the Tories. Secondly it’s down to the past vote weighting formula.

In order to ensure that samples are politically balanced both firms ask how respondents voted last time and then apply a weighting. This is how it works:-

Actual GB vote shares 2005 general election: CON 33.2: LAB 36.2: LD 22.7%

My estimate of latest ICM weightings: CON 32.8: LAB 37.9: LD 21.8%

My estimate of latest Populus weightings: CON 32.5: LAB 39: LD 20.6%

So the views of those who say they voted Labour in 2005 count for more when Populus work out their headline numbers than with ICM. The views of 2005 Lib Dem supporters count for less.

The reason that the weightings are not the actual shares from 2005 is to deal with an element of “misremembering”. In addition to past vote weighting variances the Populus “spiral of silence” adjustment is less favourable to the Lib Dems.

All this means that other things being equal one would expect that Populus polls would have a slightly bigger Labour share than ICM and a smaller Lib Dem one.

When both firms publish the detailed data from their latest surveys I hope to return to this subject again.

Mike Smithson