?????????? So, how did the polls and models perform in Scotland?

TL;DR: the polls did very well indeed, the models did not do quite as well.

Let's start with the polls. The predicted vote shares below were, on average, within 1.2pts of the actual vote shares for each party (1/13) https://t.co/WpXu8KsiSF pic.twitter.com/CCrWlo0dfx

— . (@markmcgeoghegan) July 6, 2024

Seat models were another story. The model aggregate was out for every party – overstating the SNP and understating everyone else:

Labour: -5
SNP: +9
Conservatives: -3
Liberal Democrats: -1

As a result, not only were the predicted seat values off but the prediction… (3/13) pic.twitter.com/puLrGg1uS1

— . (@markmcgeoghegan) July 6, 2024

But some models did badly – two MRPs called ~33% of seats wrong.

Here's the critical thing, for me: no model did better than a basic uniform swing model. UNS, using the predicted vote shares above, called *87.7%* of seats correct… (5/13)

— . (@markmcgeoghegan) July 6, 2024

Here is a map laying out the correctly and incorrectly called seats in Scotland, graded by how consensual the prediction was (lighter = more uncertain).

Let's look at the seats the model aggregate called wrong… (7/13) pic.twitter.com/IE9eQrKuH4

— . (@markmcgeoghegan) July 6, 2024

One incorrectly called seat was Dundee Central, which Labour nearly took – so, again, forgivable in my view. Certainly, the modelling captured a dynamic here that I know passed a lot of folks in the city itself… (9/13) pic.twitter.com/eFMY4P2qCr

— . (@markmcgeoghegan) July 6, 2024

3 incorrectly called seats were won by the Conservatives. These were all called as marginals, but something more interesting may be at work. The UNS model called the Conservative-held seats as Conservative holds, leading me to wonder what was going on with Tory swing… (11/13) pic.twitter.com/lJ5l71c6Y0

— . (@markmcgeoghegan) July 6, 2024

That's going to be it from me on models and poll accuracy. Lots more analyses to come in longer form, for @CCC_Research and @UofGPolicy, next week.

You can access the map above here: https://t.co/H2Jc361kh8

— . (@markmcgeoghegan) July 6, 2024

Alba managed a total of 11,784 votes across the 19 constituencies they contested – losing their deposit in all 19. Neale Hanvey was best at 2.8% in Cowdenbeath & Kirkcaldy, Kenny MacAskill 1.5% in Alloa & Grangemouth, George Kerevan 1.2% in Lothian East, Corri Wilson 1.2% in Ayr

— Philip Sim (@BBCPhilipSim) July 5, 2024

In Cowdenbeath & Kirkcaldy, the previous MP Neale Hanvey, who had switched from the SNP to Alba, dropped from coming first in 2019 to 7th in 2024. Are there any similar examples?

— Michael Crick (@MichaelLCrick) July 8, 2024