DURING the snap General Election of 2017, YouGov pioneered a two-pronged approach for predicting the outcome. Alongside conventional national polling, the firm published a forecast of the number of seats each party would win, based on vote shares for each individual constituency estimated by “multi-level regression and post-stratification”.

It became obvious as polling day approached, though, that YouGov were getting cold feet about the prominence given to their own forecast, which sharply departed from standard polling and the expectations of pundits by pointing stubbornly to the likelihood of a hung parliament.

An article was put out making crystal-clear that YouGov’s final call for the election was for a Conservative overall majority, based purely on national polling. That amounted to an acknowledgement that they had little faith in their experimental prediction model.

They must have bitterly regretted that last-minute disclaimer, because once again the pundits and national polls were (mostly) wrong, while the seats forecast proved to be bang on the money.

It’s because of that unexpected track record of success that there is so much excitement over the publication of updated figures from the prediction model, although ironically those numbers are now broadly in line with the picture shown by national polling.

The forecast for Scotland, for instance, is that the SNP will make a modest net gain of four seats to take them to 39, which will be no surprise at all to anyone who has followed the crude seats projections made after the publication of Scotland-wide polls in recent months.

The projected Scottish vote shares also look rather familiar, with the SNP’s lead over the Tories having increased from eight points to 12 since polling day in 2017.

Labour’s 20% share is on the low side by recent standards, though, and the headline story of the forecast is very much of ultra-marginal Labour-SNP battleground constituencies breaking decisively in the SNP’s favour. If anything, the surprise is that a massive national swing of 5% from Labour to SNP is only enough for four Labour-held seats to be estimated as having a better than even chance of falling into the SNP’s hands, with a fifth (Coatbridge, Chryston and Bellshill) being rated as an exactly 50-50 chance.

On a uniform swing, you’d expect no fewer than six of Labour’s seven seats to fall, with Ian Murray once again the last person standing. So there may be just a hint that the prediction model is picking up a successful rearguard defence by Labour in one or two specific seats such as East Lothian, perhaps with the help of Unionist tactical voting.

By the same token, it’s odd that no Tory seats are currently forecast to slip into the SNP’s grasp. Although the Scottish Tory vote is holding up better than the GB-wide Tory vote (only a trivial 1% drop in Scotland compared to 4% across Britain), the 3% boost for the SNP means there has been a healthy overall 2% swing from Tory to SNP, which on a uniform basis ought to be enough for Stirling to change hands.

So, again, the projection model may be finding that the Tories’ vote is particularly resilient where they are defending territory.

It’s worth remembering that although the projection model may be superior to traditional polling, it’s no different in one key respect – ie. it’s based on a snapshot of public opinion. It’s making no attempt to predict how voting intentions might change over the course of an election campaign. If there’s a substantial swing to Labour (as there was in 2017), the SNP could still very easily end up losing a large number of seats to Labour, rather than gaining them.