Taken from another site (hope nobody minds?). This guy really knows his onions
--------------------
It's a useful tool, but it does produce what may seem surprisingly large numbers of Lab and LD seats.
If you use UNS rather than Electoral Calculus then this produces an implausibly low figure of just 10 or 11 Lab gains for each 1% swing. This isn't going to happen, but I thought I should mention that as a starting point.
www.electionpolling.co.uk/battleground/targets/labourHowever, UNS does not work for large swings, and Labour in fact achieved significantly better at 14 seats per 1% swing in 1997 (147 gains divided by 10.5% GB swing), when UNS would also have produced just 10. I am very confident that figure will be exceeded this time.
I've done some analysis of the six 'proper' MRPs we have had in the past 12 months, two each by YouGov, Survation, and FON*. In each case, I have compared the number of Labour gains shown, divided by the Con to Lab swing. So this is a measurement of the efficiency of the way labour are gaining and the Tories losing voters.
The 'Lab gains per 1% swing' are :
YouGov 14.7, and 14.1 (but the latter figure is reduced because of their 'unwinding' adjustment**)
Survation 15.7 and 17.6
FON 16.5 and 15.9
All of these show a higher movement in seats than GE1997, although in some cases not by much. The average across all of these is 15.75 seats per point of swing. So with, say, a 14-point swing (Lab wins by 16.3%) they would get close to 420 seats.
The figures produced by Electoral Calculus's seats calculator are just a little bit higher than this average, but well within the overall range of these MRPs. From various figures I have entered based around a 15% labour lead, their 'Lab gains per 1%' are around 16.8. So it would generate an extra 13 Labour gains in the example I have quoted above, with a 14% swing.
I have not mentioned Con or LD seat losses/gains, but to be brief - Survation's MRPs produce what I think is too low a figure for the LDs, but I think both YG and FON may be a bit generous to them. I think that YG and FON are wise to use samples and modelling specific to LD seats and targets, whereas it looks to me like Survation do not. However, it is a tricky issue as to how to model for the substantial LD to Lab movement shown by polling in seats where both may potentially be competitive. As for the Tories - their losses per 1% swing against them range from 16-20, growing larger the lower their vote share goes (down to around 20-23%, where would be few left).
What I haven't mentioned in this is the additional factor of tactical voting. While the benefit of this in the Lab/Con marginals is, in my view, exaggerated, it will add a few more seats and it is clear from both by-elections and polling such as that by Ashcroft that a significant proportion (50%?) of LD and Lab voters are willing to cast their vote tactically for each other. Because of this, and above all the continued proportionate pattern of the Conservatives' vote-losses, Electoral Calculus could well be about right in showing close to 17 Labour gains for each point of swing.
* See:
en.wikipedia.org/wiki/Opinion_polling_for_the_2024_United_Kingdom_general_election#Seat_predictions
I have ignored the 'Stonehaven' MRP with a total sample only 2,000, and would also ignore Focaldata, as their previous MRPs have contained basic errors, IMO.
* see YouGov's explanation of their 'unwinding' adjustment here:
yougov.co.uk/politics/articles/49061-yougov-mrp-labour-now-projected-to-win-over-400-seats