Some thoughts on the methodology of the ComRes poll

As I mentioned yesterday, the ComRes poll showed the SNP doing a little less well on Holyrood voting intentions than in recent polls from Panelbase and YouGov - albeit still well enough for the seats projection to give them an overall single-party majority.  I wondered if there might be a methodological reason for the difference, and now that the datasets have been released a few possibilities leap out.  

First of all, unlike Panelbase, ComRes have weighted by 2016 Holyrood vote recall, rather than 2019 Westminster vote recall.  That's not necessarily illogical in a poll of Holyrood voting intentions, but if you're asking people to cast their minds back four-and-a-half years, when they've voted in no fewer than two general elections since then, there may be a potential for faulty recollection.

Secondly, although there's no suggestion in the explanatory note that the results are weighted by 2014 indyref vote recall, there is, for whatever reason, a sharp divergence between the unweighted and weighted numbers on that point.  After weighting, the 429 respondents who recalled voting Yes in 2014 ended up counting as only 330 people.

Thirdly, and perhaps most importantly, the phrasing of the question asking for voting intentions on the list is strikingly similar to Survation's, in that it describes the list vote as a "second" vote.  We know from Survation polls that this tends to produce worse results for the SNP on the list, and better results for the Greens - probably because some respondents wrongly gain the impression they're being asked about a second preference vote.