Shouldn’t poll numbers be based on responses alone?
It might come as a surprise to many that the headline figures from the three main past vote weighting phone pollsters, ICM/Populus/ComRes, are not based solely on how interviewees said they would vote.
For before finalising their voting intention figures each of them makes an adjustment based on what they think the quite significant numbers of “refused”/”“won’t say” respondees would have done if they had stated an intention.
This can cause the margin between the top two parties to change by up to three percent – which in the current context is huge and almost always at the moment favours Labour.
Thus with the last ICM poll the pre-adjusted shares were C 38: L 29: LD 21%. The firm reported headline figures of 37-30-20 – so a lead of nine points lead became a seven point one.
The pollsters do have a problems because of the number of the numbers involved. In the recent Populus poll this group represented more than a fifth of the total sample. They also only “adjust” those in this category in line likelihood to vote weightings and turnout filters.
Both Populus and ICM make the calculation by allocating about half of them in line with what they said they did at the last election. ComRes ask “Generally speaking, do you think of yourself as…?” and make the allocation based on those answers.
There is, I believe a better way. Both Ipsos-MORI and Angus Reid put a supplementary question to the refusers/DKs – “Which party are you most inclined to support?” and don’t make an adjustment. Isn’t it time for the ICM/ComRes/Populus to follow suit?
Surely it’s best to base headline numbers on what participants actually said rather than what the pollster thinks they would do?