United Kingdom General Election: July 4, 2024

[Sorenroy]

It feels like the Economist has just completely given up when it comes to their election forecast/model, which is especially disappointing because it's one of the top links to show up and one of the most reputable news organizations with a forecast available. There seems to be a total failure to account for anything besides a Conservative opposition, which means that their numbers differ substantially from what everything else seems to be indicating at this point:

Labor — 390 (+191)
Conservative — 184 (-187)
SNP — 24 (-24)
Lib Dem — 22 (+12)
Other — 4 (NA)

It doesn't project Reform getting a single seat and gives the Conservatives a 1% chance at being the largest minority party in parliament. Even at their worst confidence interval, which should really be more of a hypothetical given that results outside of that should be occurring only 2.5% of the time, they still have the Conservatives winning 80 seats with the Liberal Democrats and Reforms best intervals maxing out at 58 and 23 seats respectively.

[Sorenroy]

This ended up being a much larger task than I had initially expected, but given we had three MRPs released today, I decided to go through and see what the commonalities were between the three models. Party colors are from Wikipedia's Colors Guide with seats where all three MRPs agreed on the leading party a third darker. That means that even if all three MRPs showed a 1% lead, if that lead was consistently for the same party, the darker color is shown.



Yet Another Political Map Simulator Link
Labour — 440 (387+53)
Conservative — 109 (32+77)
Liberal Democrats — 59 (36+23)
SNP — 12 (7+5)
Plaid Cymru — 4 (2+2)
Green — 1 (0+1)
Unclear — 6

Party — MRP Average (All Three Models+Two Models)

More in Common's MRP
Savanta's MRP
Yougov's MRP


Some interesting notes: because each model was done independently and modeling at the individual constituency-scale can often result in weird numbers, even the worst individual projection for each party will contain won seats the other MRPs do not. For example, More in Common modeled Labour winning 406 constituencies, yet only agreed with both other models on 387 of them. On the other end, Savanta modeled the Conservatives ahead in 53, but only agreed upon 32 individual seats. Perhaps the largest impact of this was for the Scottish National Party, where even though Yougov found them ahead in 20 seats (maintaining more of their seats around Scotland's cities) and More in Common Found them ahead in 18 (maintaining more of their seats in the far north), the overlaps of agreement only had 12 seats where at least two of the three models showed holds.

The six unclear seats are seats where all three models showed different results. Four of them - Basildon and Billericay, Clacton, Great Yarmouth, and Louth and Horncastle - were seats where More in Common found the Conservatives leading, Savanta showed a lead for Labour, and YouGov showed a lead for Reform (the only model to show any Reform wins). The other two are Chichester and St Neots and Mid Cambridgeshire, where More in Common found the Conservatives leading, Savanta showed a lead for Labour, and YougGov showed a lead for the Liberal Democrats.

There is a lot of other interesting stuff about the individual MRPs, but the biggest standout to me was the implied vote share shown by More in Common. Their modeling on the Conservative performance is completely different from the other two despite having the exact same vote percentage for Labour as Savanta and having at most only a two point difference on the Liberal Democrats' numbers from both other models. The difference is the Conservative/Reform split. More in Common shows the implied percentage for Conservatives at 28%, seven percent higher than current polling and at least five points higher than either of the other models. On the other hand, they show Reform at only 8%, seven percent lower than current polling (Nearly half of their support!) and at least five points lower than either of the other models. That massive difference is almost the entirety of the difference between their modeled outcome and Savantas, as Labours best and worst model today show them with the exact same implied percent of the vote.