CON 40% (41)
LAB 30 % (28)
LD 20 %(19)

Is Andy Cooke starting to have an influence?

The Populus poll for February is just out and provides further data which could be used to support the notion of a hung parliament.

But well done the Times in its report on the poll for raising the question of how much the standard projections can be relied upon.

Peter Riddell writes after giving the standard projection - “These estimates assume a uniform national swing, or switch, of votes, but, in practice, there are likely to be regional and local variations…Moreover, there is evidence that the Tories may be doing better in their key target seats which would mean an overall majority at this level of national support.”

For based on the Andy Cooke analysis tonight’s numbers would produce a range of 374 - 384 Tory seats with a 100% chance of a majority. That seat range would see Cameron home with a majority of about 100.

The poll also finds that voters are becoming more likely to believe that there will be a hung parliament - which is hardly surprising giving the way that almost all the media is reporting this.

Time to start buying the Tories?

Mike Smithson

This is the second part of an examination by long-time PBer and statistician, Andy Cooke, into the level of Tory lead that the party will need at the general election in order to ensure a majority. If you have not read Part 1 then read that first before going onto this continuation

What lead do the Tories need - Part 2?

How to measure the probable unwind? The simplest thing to do would be to set the three effects as linear from the a zero-swing minimum to the full whack at a swing of about 10% - similar to the “full Blair”. This is feasible for points one and two, but not point three – it varies too much. What I assumed was that the potential magnitude of the third effect would vary from the minimum to 100% of potential linearly over the 0-10% swing range. The potential maximums were 4.9% (0-2.5% swing), 4.3% (2.5-5% swing), 4.2% (5-7.5% swing) and 2.7% (7.5-10% swing).

What’s the minimum? Well, even if Labour got to a zero-swing scenario, I’d expect an effective swing towards Cameron of at least 0.2% for a tactical voting unwind barely larger than in 2005, and 0.3% for an unwind of the Blair Effect to the tune of about a tenth of it’s magnitude, merely due to the “Brown vs Cameron” battle failing to capture the anti-Tory sentiment for Labour the way Blair could against Hague/Howard. The constituency effect would be unpredictable – arguably it could be at the maximum as the battleground is the one where the peak effect is found, but pessimistically, I’d start it at zero.

The modelling chosen goes as follows:

So, what does that give us?

The effects were modelled in a probabilistic calculator, with the Lib Dems held at 20% and the Tory/Labour scores swung around the 35/35 mark. The seat ranges are given in a +/- 1 standard deviation figure (so we’d expect the seats to be in the given range 68% of the time. To get a 95% confidence figure, double the size of the range whilst keeping the centre point unchanged)

Conclusions

The results look staggering. A lead of 5 points – according to these assumptions - could result in a Tory majority! Six points and it’s practically in the bag. Another notable feature is the dog that didn’t bark. Look at the Lib Dems. Bearing in mind that this model currently ignores any potential anti-Labour tactical voting in favour of the Lib Dems or Lib Dem overperformance in targeted seats, the score of high-forties to mid-fifties is a lot higher than many people expect. To be fair, the Con vs Lib Dem battle is not modelled, so over- or under-performance in these marginals could have a major effect on the Lib Dem seat total.

It looks too good to be true for the Conservatives. But the entire distortion currently in the system has accumulated due to factors that are going into reverse And the majority of that accumulated electoral advantage is in the seats with majorities between 0-15% - the very seats that will form the key battleground.

It’s a reflection of just how damn good Labour has been at maximising its vote where it’s needed. Unfortunately, when it lets go, it’s a double or triple whammy back. Every advantage that’s been cited is one more factor that can unravel. Since 2001, when Labour held a landslide despite the swing against them, they’ve been fighting the electoral gravity – and doing so with significant success, it must be said

What if you disagree with my assumptions? The solution is simple – if you totally disagree, stick with the UNS predictions – but I’d suggest that assuming zero tactical vote unwind, zero marginal targeting effect and zero constituency effect is even more controversial than what I’ve written.

You could take the UNS predictions as one extreme and the above as another extreme and plump for somewhere between them. Or you may decide that I’m being too cautious and that the effects won’t just unwind but wind up towards the Conservatives.

A net anti-Labour tactical vote overwhelming last times net anti-Tory vote. That Cameron vs Brown in the marginals will do more than reset the situation to the Major vs Kinnock level – in which case, the effects will be larger than laid out above. Over to you …

Andy Cooke

Based on the concept of a pendulum - not a ratchet

Later today we’ll be introducing a totally new commons seat calculator that has been developed by long-standing PBer and statistician, Andy Cooke. This will be published here sometime after lunch.

It’s stems from the thinking outlined in Andy’s initial post yesterday in which he argued for the merits of “pendulum thinking” as we seek to project the coming election rather than the mind-set of “the ratchet.”

For the conventional approach to seat projection is simply to go back to what happened in May 2005 without any thought of the previous two elections when there were disproportionated swings in what for the coming contest are the key marginals.

What happened before is locked in as if in a ratchet. Andy wants us to think in pendulum terms and go back to Tony Blair’s landslide in 1997.

The percentage movements are huge. Andy wrote: “….when all three elections are taken into account, the 2010 Labour seats with up to 10% majorities hold an extra pro-Labour swing of 4.3-4.9%. This drops off to 4.2% for the marginals with majorities of 10-15% and 2.7% for those with 15-20% majorities, and drops down to negligible amounts beyond that.”

So what happens if a proportion of this historical swing starts to unravel - a critical question because so much of it occurred in the constituencies which are the 2010 key battle-grounds and where YouGov, ICM and Ipsos-MORI have recorded disproportionate moves.

Andy’s post this afternoon sets out the detailed assumptions and illustrates them with a table showing the effect at different levels of Tory lead/deficit. We get a range of possible seats for the three main parties together with a percentage probability of three separate outcomes - Conservative majority, Conservatives the largest party, Labour the largest party.

We also get what will be a very controversial suggestion of what the minimum vote threshold for a Tory majority government might be.

The reason we are publishing this in three separate segments is partly for our own convenience - I’m out all day and Andy won’t be able to deal comments until he gets home from work - and partly to prepare the ground. Those who have not yet read Andy’s post of yesterday morning might like to do so.

Only time will tell whether Andy is right or not but at the very minimum his challenge to the prevailing consensus will lead to a more informed approach to the coming battle.

Mike Smithson

The Labour backers continue to stay away

Surprisingly, given, the latest ICM poll, the PB Index has seen a four seat upwards movement of the projected Tory majority. This has been mostly driven by changes in the latest prices from Sporting Index which now have CON 342 - 347: LAB 218 - 223: LD 53 -56 seats. The Extrabet market does not seem to operate at the weekend

The Betfair line markets is virtually unchanged.

So the roller-coaster February on the spreads continues with the market numbers being out of alignment with the latest polls when applied to a seat calculation on a strict UNS basis.

These might be pointing to a hung parliament but very few punters, it would seem, are ready to risk their hard-earned cash on it.

I just wonder whether Andy Cooke’s analysis on the previous thread might have had something to do with it.

Mike Smithson

Ukraine exit polls update:

Now available at the Election Game:

www.electiongame.co.uk

Double Carpet

We keep hearing that the Conservatives will need a lead of “about 11%” to win – but is this really the case? There has been a bias towards Labour since Major won a majority on a 7.5% lead and the decrease in the hold that the “Big 2” have over the vote and seats doesn’t ease things. But the bias is not a fixed law of Nature. In the first of two guest slots statistician and PB regular Andy Cooke tries to answer the big question

The tilt on the electoral table

These are key factors that will skew results away from the UNS which I’m going to address:

1. Mean constituency swing versus overall swing. The UNS does not translate to an equal average swing in each constituency – differential constituency sizes hamper this. If big constituencies (and high-turnout ones) act the same was as small (and low-turnout) constituencies, it would. But they don’t. When Labour is on the way up, the tendency for smaller turnouts in their safe seats increases the effectiveness of their national vote – but this factor should work against them on the way down (and did in 2005)

2 Tactical voting. In 1992-2001, anti-Tory voters flocked to whomever was best place to defeat the Tory. In 1997, the 9.55% average constituency vote increase became 10.56% in Con vs Lab seats (as against only 8.08% in Con vs LD seats). A similar effect was seen for LDs. Each time you go to the polls you have to convince last times tactical voters to vote tactically again just to stay in the same place but in 2001, Labour held on to all but a fraction of those tactical votes. Even in 2005, only a small rewind (one quarter of the total) occurred.

3 Marginal targeting/over-performance. The “Blair Effect”. As New Labour tended to appeal most to the centrist voter in the battleground seats, Labour’s performance was concentrated above and beyond the two effects above – just where they needed it most. They achieved a swing of 12.1% (on average) in the Con-held marginals (up to 10%), and a staggering 13.08% average in the “semi-marginals” – with majorities of 10-20%. Without this effect, the tactically voted constituency swing would have been 11.2%

So a 10.23% UNS swing became one of 13.08% swing just where Labour most needed it. And most of the effects were retained or even improved in 2001.

The Labour bias in the electoral table is real – but is it now fixed? A ratchet or a pendulum?

I believe all three points can go into reverse.

Point one as the battleground shifts back, point two as tactical voting unwinds and as for point three I think the evidence is that Brown does not appeal to the centre ground in a fight against Cameron as much as Blair did against Hague and Howard.

What’s the total effect?

But by how much will it affect that “11% lead” requirement - let’s break down where we are.

Point 1: In 1997, the constituency effect gave Labour a 0.28% boost above UNS. In 2001, a further boost of 0.45%. The slide back in 2005 gave the Tories a 0.24% boost. Leaving Labour holding a 0.49% swing advantage above the tide. I’d contend that this will unwind in full as those battlegrounds come into play in the other direction. That’s a half-a-percent boost over UNS that Cameron’s Conservatives are likely to get. Equivalent to adding 1% onto their vote share. Lead required: 10%

Point 2: Tactical voting enhanced the Con->Lab swing by a further 0.7% towards Labour in 1997. It slipped slightly by 0.09% in 2001, and a further 0.15% in 2005. So there’s a 0.45% residual tactical swing to play for – and CCHQ are playing for it (“love bombing” and “detoxing”). If they can rewind it (let alone push it towards the Tories), the required lead is down to 9%

Point 3- The marginals. This is the big one. In 1997, the battleground seats had an enhanced swing of 1-2% beyond the other effects. In 2001 it was more complicated, particularly in the category of seats we’re looking at (next election’s battleground).

But when all three elections are taken into account, the 2010 Labour seats with up to 10% majorities hold an extra pro-Labour swing of 4.3-4.9%. This drops off to 4.2% for the marginals with majorities of 10-15% and 2.7% for those with 15-20% majorities, and drops down to negligible amounts beyond that.

We don’t know exactly how much of the Blair Effect we can assume to unwind, but a significant amount of those excess swings should fall back, giving a boost to the Conservatives in the battleground seats. And, of course, what’s sauce for the goose is sauce for the gander – a strong Cameron performance could easily go further than a full unwind. Lead required: significantly less than 9%

Excellent news for Cameron, yes? A huge boost to his swing where he needs it most. Not necessarily. All of the above assumes a big move back to the Conservatives. If there’s not a big pro-Tory swing anyway, we can assume most of the factors will be lessened. If – somehow – there’s no swing at all, the effects will be minimal. Not zero (detox, targeting and change of leader will have some effect), but far less.

Further, if he has a big swing, the seats where the effects are most pronounced will be in the bag anyway. The effects drop off as the swing extends.

How much? That’s the next article, I’m afraid …

Andy Cooke

Mike Smithson writes: When I first read this, together with the equally insightful Part 2, I got excited because amongst other things it provides a highly plausible explanation for the way the ICM and YouGov marginals polls together with the aggregated MORI data are suggesting there’s a disproportionate move to the Tories going on in the battle-ground seats. These seats saw disproportionate moves to Blair’s Labour which have the potential to move back sharply.

To simply say that it’s “all because of the Ashcroft money” is only a fraction of it. The Andy Cooke thesis is that the marginals are inherently more likely to move because this is where the bigger than average swings were in the Blair years. There was always a greater potential for this to unwind once the Tories moved forward.

Andy’s piece demonstrates so clearly why marginals are different and why they should be the subject of our prime focus. Andy’s Part 2 has some models and projections that could change the way we are looking at the coming encounter.

CON 39 % (40)
LAB 30% (29)
LD 20% (21)

And it reports the lowest Tory share since May

New is just coming in of the latest ICM poll for the Sunday Telegraph which has the lead down to just 9% and will provide encouragement for Brown Central.

It means that the last six published polls have all had Labour in the 30s with the last four of them putting the Tories under 40. This does look like a trend.

The last time ICM had the Tories below 40% was in May - though the change tonight is within the margin or error.

We have not got the fieldwork dates yet and my guess is that most of the survey would have taken place before yesterday’s news on the criminal cases being brought over the expenses scandal.

The numbers reflect a straight 6% swing from Labour to the Tories from the actual votes cast in the 2005 election. Compared with ICM’s final 2005 survey, which had C32-L38-LD22 tonight’s poll suggests a 7.5% swing.

We are on the margin for the simple UNS calculators and it’s worth bearing in mind what the individual pollsters did five years ago.

In the next couple of days I’m running an extensive examination from professional statistician and PB regular, Andy Cooke, on what lead would trigger a hung parliament.

Mike Smithson