There is an updated version of this estimate here.
I wrote last week that things were happening in Scotland, but with only aggregated crosstabs to base my estimates on, I couldn’t quite be sure, because although aggregation creates a usable sample size, it doesn’t solve the problem of the weighting only being applied at GB-wide level, rather than to every individual subset.
What we needed was a Scottish poll with proper demographic weighting. And now, synthetically at least, we have one.
The electionforecast site is a joint project by experts at UEA, LSE and Durham, which uses a partial mean-reversion from current polling to forecast vote shares at the next election. From their current UK poll-of-polls, it looks like they are currently assuming that 41.5% of the swing since 2010 will be reversed (this amount will lessen as the election approaches).
But rather than forecasting using a UNS, the model uses the underlying YouGov microdata (augmented with demographic data and other polling) to estimate seats directly. This has all kinds of advantages, but the key thing in this instance is that it allows proper weighting (by them) of subsets of the data in course of producing their forecast, which they publish all the way down to vote share at seat level.
If we assume a pattern of turnout similar to 2010, we can derive forecast vote shares from their seat-level vote projections. And by running the mean reversion in reverse (based on my estimate above) we can imply current voting intention in Scotland, fully weighted to Scottish demographics. The implied VI, with the updated aggregation method for comparison is as follows:
This suggests that the weighting bias wasn’t that big at all. In particular, the crosstab SNP numbers look to be spot on and in fact it is Labour and the Conservatives that are a touch high, as is UKIP. The Lib Dems and Greens seem to have been marked a touch low. But it appears that no party’s vote share was significantly biased.
So not only does it appear that the SNP surge is entirely genuine, it also seems to have continued some more since last week. I’ll revist this analysis from time-to-time – one thing we can look forward to is some real Scottish polling from Lord Ashcroft. Praise the Lord.
I’ll provide updates to this analysis periodically, whether or not any dedicated Scottish polling is done. Be sure to get it first on Twitter via the button below.
Number Cruncher would like to thank Chris Hanretty of UEA for kindly providing helpful information on the electionforecast model. All aditional number crunching is by myself, and any errors are mine alone.