Polling Klaxon: Why you shouldn’t read too much into a small subsample, see this Scottish subsample as the perfect example
Earlier on this week there was much excitement about that tweet from UK Briefing that the SNP had shed nearly a quarter of their vote which would fit a certain narrative given the recent extraordinary contretemps between Nicola Sturgeon and Alex Salmond, and the original tweet was retweeted with wild abandon.
However astute poll watchers like Anthony Wells and Britain Elects plus myself pointed out that the figures quoted by UK Briefing was a subsample of 154 which is nowhere the near size to get a representative sample, it makes things far too volatile.
Later on that week we did get a proper poll of Scotland by IPSOS Mori with a sample size of 1,031 which showed the SNP demise was exaggerated, albeit they didn’t poll on the Westminster voting intention.
A sample size of 154 leads to an unsound margin of error of 7.9 whilst a sample size of 1,031 leads to a much more acceptable margin of error of 3.05. So anyone passing of a subsample of this size as a genuine poll is either being actively disingenuous or don’t understand what they are talking about, so either way they can be ignored.
I note with interest that shortly before the Redfield & Wilton poll there was a YouGov poll conducted which had a Scottish subsample which had the SNP on 55%, 31% ahead of the second placed Conservatives which didn’t attract the same level of interest.
TSE