h1

Why YouGov and ICM seem to produce different figures

February 22nd, 2011
Methodology comparison YouGov ICM
Sampling Sample is restricted to members of its polling panel – those who have registered with the firm and on whom the firm has a mass of data. The firm polls by phone making randomised calls to residential phone numbers throughout Britain. Theoretically anybody with a land-line can take part in a survey.
Political weighting To ensure balanced samples this is done by party ID based on the firm’s own data. This can be controversial – in almost every survey “others” -SNP/PC/UKIP/BNP/GRN – has to be scaled down very sharply so that on occasions those identifying with one of those parties can be worth barely a third of the norm. ICM was the pioneer of past vote weighting so that samples are weighted according to a formula linked to the May 2010 election result. There is a corrective calculation to deal with “false recall”.
Newspaper weighting Yes. The big challenge can be finding enough red-top readers to take part and on occasions those who do participate have their views trebled. No
Turnout weighting No. Because YouGov polls from its dedicated panel the level of “won’t votes” is much smaller than conventional polling and the firm does not routinely weight by turnout. Yes – by asking respondents to say on scale of 1-10 how certain it is that they would vote. Their “value” is then adjusted according to a formula.
Didn’t vote May 2010 No adjustment is made Their “value” in the polls is discounted by 50% because research shows non-voters from previous elections more likely to do the same again
The “will vote won’t says” No. This is the so-called “spiral of silence adjustment”. ICM allocates half of those who will vote but “don’t know” to the party they said they voted for last time. This increases the overall voter “pool” thus diluting the standard responses.

After the latest ICM poll I thought it might be useful setting out the main methodological differences with YouGov. For the reason why the two firms can produce very different snap-shots of opinion is down to the very different approaches of the firms.

They key differences are the last three in the table – ICM’s turnout weighting; the reduced value attached to those who didn’t vote at the last election and the firm’s so called “spiral of silence adjustment.

Mike Smithson