Why Our Predictions Come Unstuck in Scotland
There have been some valid criticisms on Twitter raised of a number of our seat predictions in Scotland. These point to shortcomings in our model that we're certainly looking to address. As a newly relaunched site, still in beta as it says on the masthead at the top, we expected there to be some teething problems and this is one such case.
Now off the bat, we don't claim to conduct proprietary polling nor to make any highly sophisticated calculations. We use uniform national swing, albeit with some minor adjustments, and make no gripes about that. This method has well known shortcomings and it's why, as I've said before, our predictions shouldn't be taken as gospel. Though we are also looking at ways we can build on it going forward. We eventually intend to run multiple models in parallel.
Onto the criticisms specifically: a major shortcoming of uniform national swing is with disparities over geographical areas. This is also something I've covered before, though is perhaps especially applicable in Scotland where the party system concretely diverges from England. By calculating the party change in vote share between 2019 and YouGov's most recent Scotland VI poll (which admittedly is not the most recent) you can see where our model might fall short.
By applying the specific Scottish vote swing, you can see we overestimate Labour's vote share by 3.5% and underestimate the Conservatives by 3.1%. On our most recent predictions, we also underestimate the SNP and the Liberal Democrats by 0.1%.
It is worth noting that at both a national level and by applying the most recent Scotland-specific polling, the SNP's level of support has remained remarkably consistent since 2019. With even the most recent polling in Scotland keeping them on 45%, the same vote share they received in 2019. So our model doesn't introduce inaccuracies on that front, though it may well in future.
The main issue comes from overestimating Labour. The impact of this on our predictions is that the SNP would hold four SNP-LAB seats we currently predict them to lose: Glasgow South West, Glasgow North, Glasgow East and Motherwell and Wishaw.
The SNP's other losses to Labour and eight predicted gains (seven from the Conservatives and one Lib Dem) remain unchanged. This puts the overall total of SNP seats at 48, with no change in number from 2019.
Applying predictions to Scotland generally has long been a challenge to pollsters. The more pluralist party system needs to be considered carefully, whilst the higher number of marginals introduces greater susceptibility to error. Our model is not exempt from these shortcomings.
