Pick and Mix

With less than a fortnight to go, here are some of the highlights of the blog over the last two weeks:

The manifestos compared – can you guess which one talks about chaos more?

Why the ippr don’t get the BNP – and why the BNP targets Islam.

Why Newnight Review don’t understand political fiction

How the polls aren’t really moving as much as you think

Why you should be sceptical about all the talk of this being an internet election

Plus:

Labour’s secret weapon: the World Cup

The benefits of attack ads

Gordon sings a song

And how to lose a TV debate, even if your opponent has punched a child in the face.

Professor Philip Cowley

Polls stability again (well, sort of)

Again, the polls appear to be all over the place. Last night’s ComRes poll had the Conservatives in front by nine points, Populus by just one. You Gov had the Lib Dems ahead by three, Angus Reid had them ahead by just one. Lib Dem support had either ‘burst’ (ComRes had them back to 26%, albeit a level of support that many Lib Dems would have sold their souls for just a week ago) or was at its highest level (34%, with YouGov).

There’s no doubt that there’s some fluctuations. But, as with the pre-debate polls, there’s also some real continuity, once you take into account the margin of error you get from sampling.

The Conservative vote is the most stable of all. Since the Lib Dem surge, every poll has put the Conservatives on 33+/-2.

With one exception – yesterday’s Angus Reid poll – every poll has put Labour on 27+/-3.

And with two exceptions – yesterday’s ComRes and YouGov polls – poll survey has put the Lib Dems on 30+/-3.

And remember that each polling company conducts their surveys differently – different ways of doing fieldwork, and then (even more importantly) different ways of weighting and filtering the data – and so we should expect to see variation between companies.

If however we look at the spreads by companies, then we see even more stability. Excluding those organisations which have only done one poll since the Lib Dem surge:

ComRes have: Con: 33+/-2; Lab: 27+/-1; LD: 28+/-2.

YouGov have: Con: 32+/-1; Lab: 28+/-2; LD: 32+/-3.

Angus Reid have: Con: 32+/-0; Lab: 24+/-1; LD: 33+/-1.

ICM have: Con: 34+/-1; Lab: 20+/-1; LD: 29+/-2.

In other words, within polling companies, not a single poll has seen movement outside the margin of error, and the Conservative and Labour spreads are even more stable, the majority being just +/-1.

One other thing. The +/-3 margin of error is 95% accurate, but that means that one in every 20 polls will show variations beyond it. The phrase rogue polls is used wildly – as Mike Smithson of politicalbetting.com often says “a rogue poll is one whose results you disagree with” – but it technically refers just to those 5% of polls.

We are now seeing so many polls – 16 since the debates alone – that we should expect about one poll every five or six days to be a rogue.

UPDATE: This piece was written before the publication of today's Ipsos-MORI poll, but that merely confirms the argument above, with the poll nestling nicely near the mid-points, at 32/28/32.


Professor Philip Cowley

Life’s not fair, son

"Would the Lib Dems prop up a government that had come third in the popular vote? Now, there’s a question many of us didn’t think we’d be asking when this campaign began..."

Three things have changed as a result of the Lib Dems poll surge following Thursday’s debate. Two are (fairly) obvious, the third less so.

Starting with the most obvious: the Lib Dem vote has increased dramatically. Today’s YouGov poll for the Sun puts them on 33. The last YouGov poll before the debate had them on 22. In other words, the party has increased its support by 50% in the space of just four days.

Today’s poll puts them in the lead, as did a bpix poll on Sunday. We’ve seen third party surges before, but not this close to a general election. No one knows whether the Lib Dems will manage to sustain this level of support, or whether it will fade (or even increase yet further). Anyone who does tell you they know is fibbing, for we are in unchartered waters here.

Second, and as a result, both the other two parties have lost support. Prior to the debates, every opinion poll during the campaign put the Conservatives on 38+/-3, every poll put Labour on 30+/-3. That is no longer true. The Conservatives have been on 31 in two of the post-debate polls. Labour have dropped as low as 26. The two-party share of the vote in today’s poll – in what used to be seen as the archetypical two-party system – is just 58%.

The third consequence is the least obvious. When I was growing up, and used to complain that something wasn’t fair, my Dad would reply: life’s not fair, son. That applies in spade to the British electoral system. Projections of seats from votes should always be treated with caution – especially now – but most projections of seats based on recent polls would put Labour third in votes but first in seats. And the Lib Dems, first in votes, would come third in seats.

Prior to the debates, whilst most polls pointed to a hung parliament it was one in which the Conservatives would most likely have emerged with most seats. No longer. On most projections from the post-debate polls (and the precise details differ depending how you make that projection), we will have a hung parliament but with Labour as the largest party in terms of seats. And that is the real game changer from Thursday.

If they can push on, and increase support, then the Lib Dems might break through this barrier, and begin to be properly rewarded in seats. But they need around 37/38% to become the largest single party, and around the 40% mark to form a majority. And short of that, the Lib Dem surge hurts the Conservatives more than it hurts Labour.

This does, however, pose a problem for Nick Clegg. The Lib Dem position is that, in the event of a hung parliament, they are prepared to reach agreement with whichever party has the most obvious ‘mandate’ from the voters.

Nick Clegg, has, however always been very careful not to specify whether mandate means votes or seats, arguing that this is a hypothetical discussion, the sort of thing that nerdy academics worry about but which didn’t worry normal folk much.

In fact, it has always been a real possibility that Labour might emerge with more seats but with fewer votes. And on today’s poll is looks even less hypothetical.

Would the Lib Dems prop up a government that had come third in the popular vote? Now, there’s a question many of us didn’t think we’d be asking when this campaign began.

Professor Philip Cowley

More MOE

"Move along, there’s nothing to see. Yet."

Yesterday I pointed out how the polls had been remarkably stable during the campaign, despite talk of their ‘volatility’, once you allowed for the margin of error inherent in sampling. Up till then, all but one poll during the campaign had put the Conservatives on 37 +/-3, Labour on 30 +/-3.


Since then we’ve had three more national polls. YouGov had 41/32/18. ComRes had 35/29/21. And Harris had 36/27/23. Fairly obviously, one of these – the YouGov poll – doesn’t fit into that earlier pattern, because it has the Conservatives on 41.

But it’s still hardly evidence of any great shift. It just means that over the last week every poll has had the Conservatives on 38+/-3 (rather than 37) and every poll has still had Labour on 30+/-3. Indeed, this is true of every poll since the campaign began.

Move along, there’s nothing to see. Yet.

Professor Philip Cowley

Polls in stability shocker

"Most people involved in politics know about sampling error, but, as last night, they often forget its implications."

There were four opinion polls last night, all showing different sized-Conservative leads, ranging from 3 to 10 points. Cue confusion, and lots of talk about the polls being ‘all over the place’.

In fact, it’s dead easy to explain, and the polls are actually behaving remarkably stably.

Each opinion poll has a sampling error, an acceptance that sampling 1000 or so people is not going to produce an absolute accurate measure of the views of 40 or so million.

That sampling error is usually reported along with the poll – at least by the more responsible news outlets – and it is (usually) +/- three percentage points.

And every poll last night showed the Conservatives on 37, plus or minus 3. (Actually, every poll last night showed them on 37, plus or minus 2). And every poll last night showed Labour on 30, plus or minus 3.

In fact, with one exception right at the beginning of the campaign, every poll during the campaign has shown the same: the Conservatives on 37, plus or minus 3, and Labour on 30, plus or minus 3.

The veteran pollster Bob Worcester has one key maxim: focus on the share, not the lead, and he’s spot on.
Most people involved in politics know about sampling error (anyone still puzzled, there’s a good simple summary here), but, as last night, they often forget its implications.

Professor Philip Cowley

And so the first week ends...

This week’s blogs have included what polls aren’t telling you (and why the don’t knows may be crucial). Or when the first time Labour asked its supporters to design a poster (the answer is the distinctly pre-internet 1908). Why the newspapers aren’t powerful in the way everyone thinks (although they might be powerful in other ways). Or why corruption isn’t as electorally damaging as everyone thinks (although it’s harmful in other ways).

Following Sir Michael Caine’s appearance at the Conservatives launch of their youth volunteering scheme (‘Tories, Sir, thousands of ‘em...), we carried a piece on the chequered history of youth engagement. Plus the most rebellious Parliament of the post-war era draws to a close and why the Demos pamphlet calling for votes at 16 was sloppy and disingenuous.

We also carried perhaps the naffest poll of the election so far – snog, marry, avoid: the prime ministerial verdict – but only to draw attention to something much more serious.

This week, coming up amongst other stuff: independent MPs, the prime ministerial debates, party election broadcasts, and much much more. Who said elections are boring?

Professor Philip Cowley

What election polls do not tell us

"...in spite of the useful information that election polls offer, they are lacking in two important respects..."

We can expect a large number of election polls to be published in the next four weeks. Not all polling agencies arrive at their numbers in exactly the same way, and they need to be read with care. And in spite of the useful information that election polls offer, they are lacking in two important respects, both of which relate to the difference between ‘hard’ choices and ‘soft’ ones.

First, consider people who respond to opinion pollsters by saying that they do not (yet) know. These respondents are generally deleted from the calculation of the parties’ projected vote shares, which implies that, were they to make up their minds, they would not disproportionally go to any one party. But consider the group of people who hesitate between the Conservatives and UKIP, and who correctly state that they do not yet know for sure which party to support. When they make up their minds, they will not support Labour, or the Liberal Democrats, or the Greens. Therefore, omitting these ‘don’t know’ responses from the results will (in this case) underestimate the actual level of support for Conservatives and UKIP.

Similarly, when asked about the choice for a party that they intend to make, respondents may name a party and are counted towards that party’s projected share of the vote. Some of these answers come from people who have firmly made up their mind, but for others this hides the fact that they are not yet certain about their choice. The polls that are currently conducted in the UK do not take into account the certainty with which respondents’ give their answers. This is particularly unfortunate in the case of a close race. We are in the dark about the proportions of voters who may still change their stated preference because they like another party almost as much as the one that they said they would vote for.

In a survey conducted by the British Election Study team in February 2010 (file feb10spb.dta), and which contains more detailed information than most election polls, we find that of those who unequivocally state that they will vote Conservative at the next general election, 4.5% actually hesitates between Conservative and Labour, and 7.6% hesitates between Conservative and Liberal Democrats. Of those who claim that they will support Labour 2.1% might still change to Conservative, and 13.8% to the Liberal Democrats. Of the projected Lib Dem support 8.9% might still go Conservative and 26.9% Labour. In all these situations, it will not take much for these voters to still change their vote intention because they are actually quite strongly attracted to two different parties at the same time.

When looking at the “don’t know’s”, we find that 12.7% is actually strongly attracted to the Conservatives, 27% to Labour, and 22.4% to the Liberal Democrats. When prodding the “don’t know’s” to indicate which party they lean to, we find that those leaning to Labour have a much stronger preference for that party than leaning Conservatives have for theirs. In other words, if, as is usually the case during election campaigns, the number of yet undecided voters declines as polling day draws nearer, Labour in particular and the Liberal Democrats to a smaller extent stand to gain more from this than the Conservatives.

The methodology employed here is analogous to that reported in Martin Kroh, Wouter van der Brug and Cees van der Eijk ‘Prospects for electoral change’, in Wouter van der Brug and Cees van der Eijk, European Elections and Domestic Politics – Lessons from the Past and Scenarios for the Future (University of Notre Dame Press, 2007, pp. 209-225).

If we want to gauge the possible dynamics in party support over this election campaign, we have to take into account the differences in the certainty with which respondents state their intended choices. In the current situation, doing so reveals an even closer race than most of the polls suggest.

Professor Cees van der Eijk

Politics academics all ignorant lefties

"I’m sceptical about much election forecasting...but I confidently acquit my colleagues of the charge of being groupthinking lefties."

Politicalbetting.com had some fun this week with the results of a poll of politics academics, carried out back in 2006, which showed the vast majority both wanted and expected a Labour victory after the general election. A full 54% of politics academics wanted a Labour victory, compared to just 15% who wanted a Conservative win. Ditto their predictions for what the Commons would look like after the election: 53% expected a Labour victory, just 18% thought a Conservative victory was likely. And an overwhelming 80% thought Gordon Brown was the best chancellor of the post-war era. Given both this overwhelming Labour bias, and what currently look like very inaccurate predictions, why should we believe the academic election forecasts reported here a few days ago?

The problem is that we’re not comparing like with like. Politics, as a discipline, is extremely broad, a point we make repeatedly to students wanting to study it at university. At Nottingham, for example, our third year students can take modules from a list of about 30, covering everything from intervention in Africa to war and massacre, from the politics of science fiction to Middle Eastern politics, and from political utopianism to Russian security policy, as well as modules looking at parliamentary politics or the politics of public opinion.

The much-ridiculed survey was of members of the Political Studies Association, a group which represents that broad church. The result is that when the PSA does surveys of its members on things like this it often gets silly results, which is why I wish they wouldn’t, and have argued before – in an article that didn’t win me any friends – that they make us all look pretty stupid.

But the recent political forecasts didn’t come from all politics academics; they came from a very specialised group of election experts. Plenty of these aren’t in the PSA anyway (many of those presenting papers at the Manchester conference are not even UK academics). So, in short, those who did the forecasting are not all PSA members, and not all PSA members are election experts. I wouldn’t necessarily trust this lot’s views on political philosophy (nor they mine), but I do trust their knowledge when it comes to elections.

As for political bias, well, there’s no doubt that academics are on the whole left-of-centre. I was once told that academics used to be much more representative of the rest of the population, until the public sector/private sector voting cleavage became so prominent in the 1980s, although I’ve seen no evidence of that. For what it’s worth, I suspect – although again with no evidence – that the elections experts are not as left-of-centre as academia in general. But more importantly, where’s the evidence that any bias, if it exists, has impacted on the various voting forecasts? Anyone who studies the papers presented at the recent conference – or any similar work – will see that these are pretty hard headed models, into which the data go, and out of which the results flow. It’s really not clear where the bias is supposed to be.

As the original post pointed out, I’m sceptical about much election forecasting (as is my colleague Cees van der Eijk, who wrote an excellent short article for the British Journal of Politics and International Relations setting out his doubts), but I confidently acquit my colleagues of the charge of being groupthinking lefties.

They might well still be wrong, but if they are, we will be able to judge their models against the result, and say why. And it won’t be because of any bias.

Professor Philip Cowley