Surprises in the detail of the Best for Britain/Focaldata MRP
Welcome to this week’s edition which was going to be a leisurely look at Lord Ashcroft’s poll suggesting Boris Johnson would win a Parliamentary by-election in Uxbridge and South Ruislip. The headline figures were somewhat surprising and out of tune with national polling, at odds with Boris Johnson’s own standing with the public, with crosstabs that were certainly puzzling enough to suggest sampling problems, and with a question wording perhaps favourable to Johnson (making no mention of why an election might occur in his seat).
However, there is also some wider context that makes the poll a bit more plausible, reality usually favours the polls when the polls and political pundits collide (including in the very first by-election poll), and we are talking about Boris Johnson. So it’d have been an interesting deep dive for me to write and, I hope, for you to read.
But [pause to check the news headlines, again] as the former Prime Minister is not going to be fighting a by-election there, instead this time I’m going to look at another surprising set of poll results from this week: the latest Best for Britain/Focaldata MRP analysis.
As ever, if you have any feedback or questions prompted by what follows, or spotted some other recent polling you’d like to see covered, just hit reply. I personally read every response.
Been forwarded this email by someone else? Sign up to get your own copy here.
If you’d like more news about the Lib Dems specifically, sign up for my monthly Liberal Democrat Newswire.
My privacy policy and related legal information is available here. Links to purchase books online are usually affiliate links which pay a commission for each sale. Please note that if you are subscribed to other email lists of mine, unsubscribing from this list will not automatically remove you from the other lists. If you wish to be removed from all lists, simply hit reply and let me know.
Taking a look at the Best for Britain/Focaldata MRP
This week Best for Britain released a new MRP analysis run for them by Focaldata, saying:
The first major seat-level MRP poll under new constituency boundaries suggests Labour are set for a 140+ seat majority with 470 seats. The Conservatives are projected to win 129 seats.
How much weight should we put on this analysis? That in part is a question about political polling in general and also in part a question about what MRP methodology is. As I’ve written about both of those at some length in a certain book, for this piece it’s worth looking at how Best for Britain/Focaldata’s numbers stack up specifically.
MRP essentially models every person, adding up the individual results in each constituency to give a constituency result.1 You can then add up both the constituency seat winners and constituency vote shares to get national totals. There are therefore three questions you should ask of MRP results to see how plausible they seem.
First, how do the national vote shares implied by the MRP look? As you can see from the table in the next section, these look pretty reasonable, in line with the conventional national voting intention polls, if perhaps a little low on the Lib Dem share. That nuance will matter when we come to the third question, but overall it looks like a plausible picture of total national support.
Second, working backwards now from those national vote shares - what does the pattern of implied vote share changes in each constituency look like? This is more complicated as the Best for Britain/Focaldata work suggests a very proportional swing, when history suggests a unform swing is more likely.
To explain those terms first. Imagine that in a general election the Conservatives polled 45% and Labour 35% last time but now the polls are showing it’s Conservative 30% and Labour 44%. That is down 15 points for Conservatives and up 9 points for Labour. The swing2 therefore is 12 points (you average the 15 points and the 9 points).3
How though do you apply that national vote share change to individual constituency results? The uniform swing method is simply to assume that the same vote share changes occur in every seat: i.e. you take 15 points off the Conservatives and add 9 points to Labour in every seat.
This has the advantage of simplicity but also produces some weird results. For example, if the Conservatives only polled 6% in a seat, how do you take 9 points off that?
It may therefore seem intuitively more reasonable to make a proportional calculation. That is, you look at the Conservatives being down from 45% to 30% and say, ‘ah, the Conservatives have lost a third of their vote (15% out of 45%) and so let’s knock one third off their vote share in each seat’.
This deals with that 6% problem, as with a proportional swing you just knock a third off it, to get 4%. But in a seat where the Conservatives got 60% last time, then with uniform swing their vote would be down to 45% (uniformly applying the loss of 15 points). Yet with proportional swing it would be down to 40% (knocking a third off 60%). In other words, with proportional swing, the stronger you are in a seat, the bigger your vote share change.
The problem with proportional swing, however, is that historically it has been worse at predicting seat numbers than uniform swing. Although uniform swing has problems with very low or very high vote shares, and is harder to theoretically justify than proportional swing, it has the merit of working better.
Back now to the Best for Britain/Focaldata MRP and my second question, what does the pattern of implied vote share changes in each constituency look like? The answer is that it looks like very pronounced proportional swing.
Yet as Peter Kellner puts it in an excellent piece, “Taking Britain as a whole, past general elections have never produced proportional swings.”4 (Though it’s worth pointing out that we also saw much more proportional swings than usual in the last local elections.)
It’s possible that proportional swing is a feature of mid-term political views and that the public’s pattern of support reverts to more uniform swing patters as a general election nears. That, however, would still leave the Best for Britain/Focaldata model of very proportional swing looking unlikely to be what we’ll get at a general election.
Another way of thinking about this is to look at some individual constituencies, as Owen Winter did for one of last year’s outings for Focaldata’s MRP model (done before Caroline Lucas announced she was standing down):
One prominent example is Brighton Pavilion. Here, despite the Greens more than doubling their support nationwide in the FocalData model, Caroline Lucas is projected to have her majority cut by 15%, from 34% to 19%. Although individual constituencies do not always go the same direction as the country, what reason is there to think Brighton Pavilion would swing 15% against the Greens during the party’s best election ever? More so, what demographic or local variables in the MRP model could cause this to happen?
Another is Twickenham, a Liberal Democrat stronghold. Here the Liberal Democrat vote is projected to fall by 24% while the Liberal Democrat vote nationwide falls by 2.9%. This would be a bigger decrease for the Liberal Democrats in Twickenham than their catastrophic 2015 defeat. Part of the reason the FocalData model resulted in only two Liberal Democrat seats is similarly huge declines across the seats they won in 2019. Meanwhile the Liberal Democrat vote stays almost wholly intact across much of the rest of the country.
The Lib Dem vote share being pretty much intact in the Red Wall, say, but plunging in seats with Lib Dem MPs seems an unlikely pattern. It is though what the Best for Britain/Focaldata model pointed to.
Which brings us to the third question: how plausible looking are the individual constituency results which vary from the overall national pattern (whether that pattern is proportional or uniform swing)?
Again, there are some puzzles here as Peter Kellner has pointed out:
The latest Focaldata projection shows the Conservatives implausibly gaining support in Liverpool and parts of inner London, even as their voters flock to other parties everywhere else.
As with Owen’s examples, that seems unlikely. The individual constituency results also show Lib Dem support down by far the most in the seats where the party is the strongest challenger to the Conservatives. The more plausible the Lib Dems are as winners in a seat, the worse they do. That’s what happens in a vote share meltdown year, true, because everything melts down.5 But it isn’t a Lib Dem vote share meltdown that the MRP projects. What it does project is that the stronger the tactical voting message is for the Lib Dems to use, the worse the party will do. That seems odd especially at a time when the size of the tactical voting pool (Labour and Green support combined) is up, not down.6
Which makes it again a somewhat unlikely set of vote share distributions and to be fair, Best for Britain’s CEO caveated their Lib Dem results too:
Come a real election, the Lib Dems should pick up significantly more seats than this poll shows.
To summarise: their headline figures for a Labour majority at the next election are jaw dropping. There is though good reason to think that this may be exaggerated, as although the overall vote shares look plausible, the model produces a strongly proportional swing that is at odds with previous electoral history and which produces some pretty strange looking individual constituency results.
National voting intention polls
The Conservatives had a 12 point lead at the 2019 general election. All but one pollster currently shows Labour now with a larger lead than that (with the one exception still being an 11 point Labour lead).
Here are the latest figures from each currently active pollster:
For more details and updates through the week, see my daily updated table here and for all the historic figures, see PollBase.
Last week’s edition
Are Millennials politically that different from everyone else?
Know other people interested in political polling?
Labour rated ahead of Conservatives on immigration, and more polling news
Keep reading with a 7-day free trial
Subscribe to The Week in Polls to keep reading this post and get 7 days of free access to the full post archives.