PolitcsHome have released the full report for their updated marginals polls – it covered just short of 240 seats, with a sample of around 35000. It is a repeated of their 2008 exercise, using exactly the same seats and the same methodology so we can see how the picture has changed in the marginal seats.

Firstly a word about how they (or to be upfront, how I did it, since this was largely carried out by me) did it. The 238 seats are mostly Conservative target seats, held by either Labour or the Lib Dems, though there are also a small group of Lib/Lab marginals and some Plaid and SNP marginals. Last year the choice of seats could have been better – even with 238 seats in some areas the Conservatives got a swing large enough to take them all and we should in hindsight have been polling more distant prospects. Rather than expand the sample, this year we polled exactly the same seats to make sure it was comparable. With a lower Tory lead, we mostly avoided the earlier problems anyway.

Even with a sample of 33,000k+, the number of people in each individual seat is not high enough to give reliable voting intention figures for individual constituencies, so instead we split the 238 seats into 17 groups of seats that shared similar characteristics – so there is a group of Con/Lab marginals in the East Midlands, a group in the urban West Midlands, a group of Con/Lab seats in the London commuter belt. Most importantly we seperated out the seats contested between the Lib Dems and Conservatives (and indeed, between the Lib Dems south west heartland and their more suburban seats elsewhere). Note that the seat predictions in the poll are all based on uniform swings within each group of similar seats – last time lots of people got the wrong end of the stick and through they were based on the 150 or so people in each individual seat, they aren’t, since the sample would be too low.

Finally before we get to the results, there’s how we asked the questions. My theory is that when people are asked voting intention questions they give the name of the party they actually support as their first choice, when in reality they might vote tactically, or be influenced by the particular candidates in their seat. For the marginal seat poll therefore we asked normal voting intention, then asked if this was a tactical vote (just to prompt them to consider it) then asked them to think about the particular circumstances and candidates in their own seat, and how they would vote there. This makes a significant difference, particularly in Lib Dems seats.

So, with that out the way, what does the poll show?

Predicted Conservative majority down to 70.

Last year’s poll predicted a Tory majority of about 145, at a time when they were around 20 points ahead in the national polls. This year’s poll was taken at a time when the Conservative lead had fallen to the mid teens, so unsurprisingly we found a similar drop in Conservative support in the key marginal seats, with a smaller number of gains.

The pattern of support hasn’t vastly changed. The largest swing to the Conservatives is still in the Midlands and the London commuter belt (they are doing comparatively poorly in seats in London itself). Weaker areas are in the North and, interestingly enough, in seaside towns.

Support for minor parties

In most areas the narrowing of the Conservative lead isn’t due to any vast increase in Labour support, but a shift from the Conservatives to “others”. That is, incidentally probably the reason why the seaside towns group is so bad for them – because others are up by 11 points, almost all that to the Greens. The uniform swing in that group is enough for the Greens to take Brighton Pavilion (and given that the support for the Greens is probably actually concentrated in the Brighton seats rather than all the seaside towns polls, that’s probably a very good sign indeed for them in that seat).

Lib Dems holding their ground

A national uniform swing projection for the Lib Dems at the moment normally shows them losing a large number of seats. For example, taking a YouGov poll conducted at the start of the fieldwork for this poll, the national shares of CON 41%, LAB 27%, LDEM 17% showed the Lib Dems losing 26 seats. The PoliticsHome marginal poll last year already showed them doing better than national polls suggested, largely because of the local prompting in the voting intention question. This makes a huge difference in Con-vs-LD seats: about a third of people in Con-vs-LD marginals who say Labour when asked how they would in an election tomorrow, say Lib Dem when asked how they would vote in their seat. There was also some shift from the Conservatives, suggesting the effect of a strong personal vote for Lib Dem MPs.

This year the position is even stronger for the Lib Dems. There is still a swing to the Conservatives in their South-West heartlands, and a uniform swing in those seats suggests 8 Conservative gains. However, in Con-vs-LD seats elsewhere, mostly the more suburban seats, though also places like Westmoreland and Lonsdale and Norfork North, there is virtually no swing at all to the Tories. The overall prediction for the Lib Dems is 55 seats, so down by only 8.

The full report can be downloaded here, the data tables should be up here at some point later on today.


62 Responses to “More on the PoliticsHome marginals polls”

1 2
  1. Anthony – I suppose what’s confusing me is that you’re saying the sample size is only c150 but last time it was being marketed as nearly double that. Why the discrepancy?

    Also, one other thing – in terms of the demographic weighting, did you weight the demographics in individual seats, or to the regional groupings, or just to the demographics of the whole UK?

  2. Anthony , why with this poll do Yougov not have to follow the BPC guidelines/rules and publish all the detailed data tables . Is there a special dispensation so that PH can try to sell the data ?

  3. There isn’t, they have to publish the tables for all the data they publish (they are up on the PoliticsHome website). Data they don’t publish, they can do what they wish with (including selling).

    Question – weighted to match the demographics of the marginal seats, not GB as a whole (which would be horribly wrong!)

  4. Anthony

    I posted a message about missing regional data on the PH website on Saturday, but no one has reacted. Is anyone monitoring it please?

  5. Anthony,
    To your explanation above regarding weighting (“weighted to match the demographics of the marginal seats”), do you mean that the approx. 140 respondents per marginal seat (33,000 divided by 238 seats) were weighted to match the demographics of the whole population (70,000+ I guess) of the seat? When were the interviews completed, as this could easily skew the sample with regards to age group and/or employment status?

  6. Anthony,

    When you say “weighted to match the demographics of the marginal seats”, does that also apply to Political ID?

    I suspect the Scottish figures look so positive for the unionist and dismal for the SNP is at least in part because YouGov’s Scottish PI weightings are so different to their GB weightings: 16% vs 3% for “others” just for starters.

  7. It’s weighted to the total aggregate figures for the seats polled, not individual seats. Fieldwork was 11th Sep to 21st Sep.

    Brownwdov – yep, it was to the estimated party ID for the aggregated seats being polled, which was slightly different to the party ID figures YouGov use for the country as a whole.

    The SNP actually aren’t shown doing to badly, there is a 9%+ swing to them, which is not to be sniffed at (it would be their second highest share of the vote ever at a general election). It didn’t produce many seats because of the distribution of their target seats.

    In reality, that sort of swing probably would get them several more seats, since swings are not precisely uniform. Basically, there are lots of SNP targets that need a swing of just over 10%. This poll showed them getting a swing of 9%, so did not project them to win any of those seats. In reality, even given a uniform swing their swing will be randomly distributed around that 9% average, so they would get an 8% swing in some seats, 9% in others, 10% in others – 7% or 11% in a few. It’s very likely, therefore, that on this sort of swing they would end up picking up some of those seats.

  8. Anthony,

    OK. Thanks for the response. Using “estimated party ID for the aggregated seats” I presume you’re talking about for the 19 Scottish seats rather than the GB figures which would skew the Scottish and Welsh figures disproportionately.

    I suspect the “anyone but Labour” factor will ultimately play a very big part in the actual results, but accept that’s probably not an unreasonable snapshot of the situation just before the party conference season.

  9. Nope – it was done overall (but unlike YouGov’s normal weighting SNP/PC were weighted seperately, rather than lumped in with others, so it should produce the correct proportions for them)

  10. “Question – weighted to match the demographics of the marginal seats, not GB as a whole (which would be horribly wrong!)”

    That’s why I wanted to check!

    Just to check I’m following you correctly, each regional grouping was aggregated and then weighted according to the demographics of the whole group of seats?

    I just ask because some of the groups of marginal seats would seem intuitively to have quite different demographic make ups – looking at Outer London, it seems a bit surprising that Bromley, Dagenham, Harrow and Hounslow would all be lumped together apart from pure geography? Or would I be surprised by their similarity?

    Personally I’d thought that Dagenham might be more in tune with Essex and Bromley with Kent, than either does with more “London” seats like those in Harrow and Hounslow. But maybe I’m wrong…

  11. Nope, it’s weighted to the overall aggregate of all the marginal seats polled.

    I won’t go into great detail about how I worked out the target weights and came to that decision, but essentially it’s because I had the necessary data to calculate a robust target weighting for party ID for the seats as a whole, but not the groups.

    The groups are based only on my own opinion – nothing more. When we first designed it in 2008 we took a great big list of all the “battleground seats” in terms of the swing needed and the current political environment, and then added some other seats we thought might be interesting like Sunderland Central (interesting boundary changes) or Bromley (because the Lib Dems had done well in the by-election) – then I grouped them the best I could.

    There are some that are a bit hard to fit in, some that don’t sit very easily. Some of them (like Eastleigh and Winchester for example) I think I’d put in a different group than I did if starting again from scratch, but we decided to keep the groups identical to last year to allow people to draw comparisons from one year to the next.

    To take your example Bromley is normally a safe Tory seat, so doesn’t fit easily with the others there at all. If we were grouping it with others they would be Orpington, Beckenham, perhaps going up to places like Chingford and Woodford Green… all seats that weren’t polled. Really it should live in a group called London suburbia or something, along with bits of Surrey.

    Lots of seats will buck the trend of the groups I’ve put them in – either because they are a bit different from the others, or because of candidate effects, or demographic changes, or just the differing swing from one seat to the next.

    They probably won’t differ *that* much, because swing won’t differ that much anywhere – you can see from this that the real differences are where there are different parties – so the swing in Lib/Con seats really is very different from Con/Lab seats, Scotland is different and the Greens make the group with Brighton and Hove different. London does look a bit different (a negative Boris effect perhaps?), so does the far North… but otherwise? Not a vast amount.

    The poll is a long way from perfect – but it’s the best I can do. Certainly it gets us further than the standard GB polls, and it’s virtually the only clue we’ve got to how well the Lib Dems are doing against the Conservatives.

  12. Anthony,

    Thanks for the clarification. Obviously that’s fine for the SNP weighting, but for Lab and Con the Sep YouGov PI weightings for Scotland were 36 and 11 vs the GB ones of about 32 and 26.

1 2