What went wrong

In the fullness of time I am sure much more will be said about why the polls overestimated the level of Lib Dem support at the election, but there was an interesting nugget from Andrew Cooper of Populus on More or Less on Radio 4 earlier. Populus’s final poll, conducted on the Tuesday and Wednesday of election week, had the Lib Dems on 27%. However, according to Andrew in the fieldwork conducted on Tuesday the Lib Dems were in the high twenties, in the fieldwork conducted on the Wednesday they were on 24%. That looks like evidence of late swing – that the polls weren’t wrong, people just changed their mind right at the end.

However, there is also some evidence that casts doubt on late swing. Because they published in the Evening Standard on Thursday and had a later deadline Ipsos MORI’s final poll of the campaign had the latest fieldwork of all the pollsters – all their fieldwork was conducted on Wednesday… yet they still had the Lib Dems at 27%.

Also illustrative is Ipsos MORI’s post-election poll. Most companies use some form of past vote weighting, so their post-election polls will be calibrated to the new results and won’t really be directly comparable to pre-election polls. Ipsos MORI however don’t use any political weighting, so their post election polls should be conducted in exactly the same way as their pre-election polls. In their post-election poll for the News of the World MORI asked how people had voted in the election on May 6th, and found figures of CON 35%, LAB 31%, LDEM 28%. No sign there of a big drop off in Lib Dem support compared to pre-election polls.

Of course – we know all about the problems of false recall, there may be people claiming to have voted Lib Dem who didn’t actually do so, so this isn’t conclusive either, but it isn’t screaming out late swing.


410 Responses to “What went wrong”

1 2 3 9
  1. FWIW – I think there was a late swing to Labour from LD’s as some anti-Tory votes found their best home and GB showed some passion (too little too late perhaps but that’s another story).
    Most polling companies picked this up but maybe not enough (remember You Gov ‘s penultinate Daily Poll )LD 24% I think reversen though the day after)

    Probably was some LD overstatement as well.
    I recall when they were scoring 30ish some posters suggesting that it was overstated as a higher %age of their ‘new’ support would not vote in the end – students young people.
    Also, maybe some respondents not intending to vote would rather give a party so as not to appear apathetic (odd as responses are to strangers but some people kid themselves like shy voters) and the LD’s were in the news more Clegg mania etc.
    Finally, is it possible Shy Labour was understated?

  2. Anthony,

    Excellent post.

  3. @ Anthony,

    This suggests to me that 3% of Conservative actual votes came from either:
    1. LibDems being so dumb they couldn’t get their X in the correct box; or
    2. Shy Tories on the rise again; or similarly,
    3. Right-leaning LibDems voting Conservative (tactically, ABB) but continuing to say they voted LD; or
    4. MORI isn’t connecting with a truly representative sample of the population – i.e. MoE

    It’s probably 4. MoE. Which is a bit of a dull conclusion. :-(

  4. @Eoin FPT

    Have just seen your analysis of the way conference voting works. Sorry Eoin, but while it may match how it works in say Labour, for example, it is not how it works in the Lib Dems. We don’t have DELEGATES, we have REPRESENTATIVES. In other words, each local party picks their quota of people to attend (dependent on how many members they have) and then those representatives do whatever they want. They do not have to seek any sort of additional mandate and cannot be instructed to do anything.
    As someone who is a voting conference rep, and was at the special conference on Sunday, I can tell you that it would be virtually impossible to stitch up a Lib Dem conference to suit ANY cabal in the party. And people have tried. And always failed.
    Oh, and on Sunday, the vote for the coalition was almost unanimous – not because we were forced into it, but because that was what we decided to do having listened to the debate.
    I have great respect for your opinion on some things, but on the internal workings of the Lib Dems, you have a long way to go before you are an expert!

  5. @MrsB,

    Interesting and thank you.

    How do they pick reps? Is the process formalised?

    What % of LD membership attended the meeting?

    one man’s delegate is another’s rep.

  6. I suspect a combination of uncommitted Lib Dems and shy labour voters might explain the differences.

    Unconnected, but I was struggling to understand how Cameron could shift so quickly and effortlessly into policy u turns and coalition arrangements, until I realised ‘Dave Cameron’ is an anagram of ‘A Move Dancer’. He should get on fine with Vince then.

    More worryingly, perhaps the Lib Dems really are going to wipe their feet on the Tories on the crime and civil liberties issues, as we see ‘Theresa May’ is an anagram of ‘A Heresy Mat’

  7. So the election day Mori poll suggests that not only people said that they voted for LibDems when they did not, but also that they did not admit that they voted for a smaller party or as somebody remarked, they did not admit that they did not vote.

  8. @ Alec – I like your thinking. Let us note too that ‘Nicholas Clegg’ is an anagram of Slogan Chic Leg and ‘Diane Abbot’ of A Babied Ton.

  9. The underlying methodoligical weakness in the LD share then one presumes continues.

  10. The issue was that the Tory vote was more or less correct in the polling, and labour’s a point too low. That doesn’t explain why in many polls the lib dems were given 5% more in the polls than they actually managed. That can’t be explained away by “shy tory” or “shy labour”. People who said “labour” or “tories” voted as the pollsters expected them to, not so with the lib dems.

    Is it possible there was a “over enthusiastic lib dem”? Lib dems were highest amongst younger voters, who tend to be worse at actually turning up, IIRC. Could it have been the LDs attracted the less politicised younger voter, who would have a below average propensity to vote, even when compared to those of their age, throwing the LD result?

  11. Of all the polling companies MORI and COmR measured the LD share more accurately than the others after the first debate…

    ComR regularly had them at 26% MORI dipped to 23% at one point.

    YG except for one occasion usually had them the highest (34% on one occasion).

    What do ComR and MORI do differently?

    I have long suspected the Labour disloyal rating to be outdated

  12. @Laszlo

    Lets look to the simple solution. The pollsters got their sampling approach wrong. We can hypothesise about late swings etc, but the reality is that they can never be tested, so they aren’t really great hypotheses.

    And this is a big difference – I don’t believe for a minute that in all the polls conducted about 6% of the sample lied (in a consistent way – lets assume that random lying is going to more or less cancel out). I especially don’t believe that in a post election sample 6% lied.

  13. @EOIN

    Is there a formal process for getting conference reps? Yes there is. They are elected (by STV) from amongst members of a local party, usually at the AGM. In most cases 1 Local Party = 1 Constituency. The number of reps per local party depends on the number of members, the minimum no of reps being 4 (and the minimum local party size 30).
    One thing I didn’t say before is that these are VOTING representatives. Any party member can go to conference, and can join in everything – they just don’t get to vote.
    On the difference between delegate and representative, there is actually quite a difference – one is only allowed to do what the people who sent them there have mandated them to do, the other can think for themself.

  14. Given that Mori have the actual result “wrong” in their analysis of how people voted last week, I think that the problem is more likely to be systemic rather than “false recall”. While some people might not remember what they did several months ago, how could they not remember what they did last week when it was such an unusual outcome ?

    One area I found interesting in the detailed tables is in the weighting adjustment from unweighted to weighted figures by social class. The unweighted figures were actualy clsoer to the result than the weighted figures. The adjustment is presumably made to adjust the profile of respondents to the profile of the electorate. But what if the original profile gave a better match to that of electors (ie those who actually voted) as opposed to electorate ?

    In other words, this mismatch as between voting recall and the actual result is a reflection of the differential turnout as between different social groups, such that when recalibrating the data to the profile of the electorate, the data is thereby skewed towards those parts of the electorate which failed to vote.

    In contrast, the “late swing” argument is at odds with the data in the table since it was the LDs which had the highest proportion of support determined in the last 24 hours. That should have been reflective of a late swing to, not from, LDs. Only if these voters decided they voted LD after the polls closed would it explain the discrepancy.

  15. @james Ludlow – at least Liam Fox should rise in stature now – anagram = ‘Loaf Mix’.

  16. There is an interesting piece by David Runciman of Cambridge University on this in the issue of the London Review of Books that has just come out. In it he quotes research by Goodin and Rice, ‘Waking up in the Poll Booth’, in Perspectives on Politics, Dec 2009.
    Essentially the view is that the campaign polls were not wrong, but when confronted with an actual choice of government on the day, voters turned away from the ‘lighter’ alternative stimulated by Clegg and the debates, to make a very serious decision between the two main contending parties about who to trust to govern for the ensuing years.
    This view, which is not unknown on academic circles, has always struck me as valid. Voting is not a consumer choice between packets of cornflakes, more like choosing professionals to represent over a long period of time.
    It is backed up by evidence not only from 40 years of British elections but also from Australia, where voting is compulsory, so it suggests it is not a question of differential turnout so much as genuine late changes of mind between what is said to pollsters during the froth of the campaign and the serious ‘priced’ decision (one which actually has consequences, which answering polls does not) in the voting booth.
    When I was a pollster, I very much believed in this distinction. for example, people always say they would welcome tax rises to pay for more and wider services. Their actual voting behaviour often suggests otherwise.

  17. Throughout the years before the election there were big swings in every direction suggesting a lot of uncertainty. In the end on the final day most of these people decided to ‘play it safe’ and voted for the Conservatives and Labour.

    In the last few days before the GE the other big news was the serious economic problems in Greece. And confidence regarding who would best take care of the economy in these uncertain times was the DECIDING FACTOR in not voting Lib Dems who have not been trusted with the economy in modern times – ‘no time for a novice’ used in Labour’s conference the year prevented Labour having a disasterous GE.

  18. @ Robert Waller

    It sounds very plausible – I would also add that it is likely that it happens well before the voting day, only it is not verbalised, not admitted to the person himself or herself.

    If it is combined with the social stratification mentioned by Paul H-J in a different contexts – you can make a decent political analysis, I suppose.

  19. @ Duncan @ 5:48

    Sorry, somehow I missed your post. Apologies.

    I meant my post ironically. Having organised voting in my youth, I know perfectly that in most cases you have to persuade people not only who should be striken off, but also who should be written in – so, yes, it was not people’s lying. Having said that, there might have been a small bias people having problems with admitting the voting for Labour and Conservatives (hence the bias in favour of LibDems).

    But I think it were sampling and sample correcting errors. The polling companies have the raw data so they should know what went wrong…

    It is also possible that the error was caused by the “fact” that people “knew” that Labour could not win, but did not “want” Conservatives – so the “desire” of hung parliament came in in this form and perhaps the questions were inadequate to capture this.

  20. I think that the Whig bubble didn’t even exist and people told the polsters that they were going to vote for them but were actually Shy Tories.
    However, the Whig share increased while the Tory share didn’t increase as much. Polls are only a snapshot and don’t get it right 100% of the time. But the Exit Poll was almost spot on. Way better than the 1992 or 1987 exit polls.

    I suppose there were also unknown unknown factors that burst the Whig bubble or that the public has a wicked sense of humor. Probaly both.

  21. “the polls weren’t wrong, people just changed their mind right at the end.”

    Brilliant excuse.

    I wish I’d thought of that.

  22. @Robert Waller

    If that were the case would we not have seen similar last minute drop-off in LibDem support in past elections? If anything, Libdem voting on election days has tended to match or exceed polling projections. This time the error sticks out like a sore thumb.

    My own take is that the media-induced Cleggmania produced a lot of weak positive intention, rather than the usual indifference among eventual non-voters. People were saying LibDem to pollsters as a euphemism for don’t know and don’t care, rather than there being any last minute switch to other parties.

  23. The only thing that died faster than Angus Reids reputation was Mike Smithson of Political Bettings golden rule of “always take Labours worst polling percentage as being the nearest to the actual election result”.

    As dead as a deceased dead thing. :-)

  24. Red Rag – surely what matters most is the trend over many elections not just the result of one election.

  25. As a watcher of Scottish election changes for over 50 years I am pleased to see progress in this election in that analysts now recognise not just that we no longer have the tribal voting that was common in 1952, but that we now have anti-party voters.

    In Scotland the anti-Con vote is very large, and the anti-Cons are spoilt for choice. Not only are there the LibDems and the SNP, but Greens, independents and fomerly Socialists have been elected mainly on the list vote.

    There is an anti-Lab and an anti-SNP vote too, but much smaller.

    The split vote encourages voters to be innovative even if some think it is a second choice.

  26. Another theory is as follows: the polls measured those sure to vote; let’s suppose that was accurate. But actually a few percent more people ended up voting who had previously said they weren’t sure to. These people were dragged to the polls by the GOTV operations of the two main parties: Labour and Conservative. The LibDems just don’t have the activists for a comprehensive operation. The result was to flatter Lab and Con votes at the expense of LD.

    Convincing?

  27. I guess it just comes down to the fact that people were just lying. Instead of saying they would stay at home they said they would vote Lib-Dem? Add shy Labour supporters into the mix and that pretty much explains it?

  28. The Liberal Democrats lost out in the resources battle. However some polls put the certainty to vote figure in the 70s which if accurate probably adversely affected the LD’S more.

    Anyway in a sense funny how a disappointing performance on the night didn’t count too much in the end though still a pity Nick Clegg didn’t have a bit more bargaining power. Nonetheless a reasonable deal has been achieved which was for so long the stuff of dreams and the coalition a better deal for the country than any of the alternatives. Of courrse a lot of battles ahead but I’m cautiously optimistic.

  29. Well, one factor would be all those who voted BNP – 6.7% in Newcastle – who wouldn’t admit to it and said they were voting liberal. In the local polls in the NE the liberals were polling incredibly high, yet we saw little sign of that, rather what appeared to be a straight swing from Labour to BNP.

    Although, the whole notion of a swing is ridiculous it only works in a two party state. We have no idea if people really swapped Lab to BNP or if they swapped Lab to Tory and Tory to Liberal and Liberal to BNP in equal measures. You just can’t apply a 2 dimensional tool to the British electorate any more, we need some more multi-dimensional equations to deal with it.

    The most common phrase I heard all election night, from all at our party and on the phone, was “will they stop talking about the ****** swing, there’s no such thing.”

  30. John,

    You are falling into the trap of believing the claptrap about high turnout. This election had the third LOWEST turnout on record. Better than 2001 and 2005, but still well short of 1997, which was itself the second lowest turnout since the war up until that time.

    Had turnout been 10% higher than it was, then it would have been “average” by historic standards.

  31. Daffid,

    Quite agree about this swing fixation.

    I doubt that that many people actually changed the way they voted. Far more likely is that people either stayed at home or decided to get out and vote in different measures.

    Certainly talking about a 7 or 8% swing when Lab vote fell 10-12% and Con vote rose 2-4% makes no sense at all. Yet that is what happened in most Lab held seats with large majorities in 2005.

  32. @JamesC,

    You have got closer than anyone to identifying the Lib Dem problem, certainly on this blog at least.

    Univeristy areas have the lowest turnout of all the districts in Britain.

    I live in a Uni District… our turnout was 30%. I myself did not, and will not ever vote.

    I think we all accept that uni folk are fairly liberal minded- lending themsleves to be well disposed to yellows.

    Thus, even if they say they will vote- the art of doing it rarely materialises.

    On a practical level- uni areas are silent battlefields between ‘locals’ and ‘students’. I have a wee theory- if polling booths are in childrens primary schools (classic local turf), to what extent would it dissuade students from a) locating the booth b) actually entering that domain/terrain to cast thier vote…

    we know of all the other transient problems (high rent occupancy) mobile lifestyle so perhaps not around on the day to cast their vote (or register).

    I checked the electoral roll out of curiousity… I am not on it. I was last on it at my parents house in 2002 I wonder how much of this damaged LD support..

    Aside from all of this there is the old saying “you can take a horse to the water but you cannot make him drink”

    In the case of the LDs can they even get the fecking horse to the water?

    I am very dubious as to whether or not they can.

    This is not an indictment. I have been invovled in canvassing and shuttling voters organising postals/proxies etc…. It is very tough work… do the LDs have people with the drive, ethic or ideological committment to carry out these functions accross 632 constit?

    Lastly, LDs by their own admission are ‘pragmatic’ to put it politely. I wonder if that carefree dare I say it whimsical approach to life lends them less likely to get up off their behinds? If they were any more laid back they would be horizontal ( and I am not even included the sexually liberal).

  33. In previous elections the LDs/Alliance/Liberals have picked up votes (if not seats) in the last week of the campaign whereas 2010 showed the reverse.
    I am surprised that an election which could produce a change of government (as it did) still showed a poor turnout. I thought that as the winner was not assured (it was in 2001 and 2005), you would expect greater motivation by the electorate.
    Finally as the swings were all over the place whether Lab/Con or Lab/Lib or Con/Lib, it may be that national swings are now dead and like the Exit poll the only true measure may be regional polling as Politicshome carried out in September 2008 and September 2009; pity they didn’t do it in April or May 2010!

  34. @Eoin Clarke
    “I myself did not, and will not ever vote.”

    I find that an amazing statement from someone interested enough in politics to make frequent posts on this blog. Could you enlighten me as to your reasoning for this position?

  35. @Eoin,

    ““I myself did not, and will not ever vote.””

    I have always had you down as a Socialist, Eoin. I’m guessing you don’t vote because you don’t think the system as properly democratic. Just a hunch. :-)

  36. @Pete B,

    First I must state that I have every respect for those who do. I also abide by democracy regardless of the result.

    I have a very long winded philisophical objection to it, on a personal level at least. Would you really care for that? And is this the place for it?

  37. Or call it intuition. The only possible reason that someone who has such an interest in politics (and so much knowledge) would not vote is because they are disillusioned with democracy and its processes. :-)

  38. @Matt,

    Yes, you are reasonably close.

    Technically less than 30,000 voted for Cameron.
    The pop of UK in c. 60 Mill – 10 mill voted blue.
    At the time Bush invaded Iraq c58 mill ovted for him roughly 17% of USA. They called Sadam a dictator. Surley he could have mustered 17% of the electorate? Demos Kratos they are funny old words when you consider that we conflate the opinions of 60mill into the votes of 324 white fat men (give or take a few token women).

    All in all it is pretty shoddy.

    But as I say I as a tax payer, view with intrigue how you all reach your decision.

  39. @Eoin,

    Thanks for your reply. :-)

  40. Eoin
    I can see your point, but would your resolve never to vote change if the system became more democratic?

    Our system is far from perfect, but it is surely better than some others such as rule by a tyrant.

    And finally, why not spoil your paper by writing all over it explaining your position? I have done that on occasion myself.

  41. I agree wholeheartedly with Robert Walker. The most extreme example of a Con/Lab squeeze on the LD came in the North East. Pre-election polls showed labour haemorraging support to the LD in tens of NE constituencies. Some even projected swings about 10%. In the real thing there was indeed a huge swing in the NE against Labour – but in favour of the Tories.

  42. @Anthony

    Do we have the regional breakdowns of the GE result?

  43. It’s all quite simple really. When Cleggmania happened after the first debate, most of the increase in LD support was new, mainly younger voters, rather than switchers from another party. People were confidently saying that there was such enthusiasm for the election that turnout would be back above 70%.
    In the end these new LD people’s enthusiasm didn’t last and they just never turned up to vote, and turnout ended up at a pretty miserable 65%.

  44. The Report, BBC Radio 4, available on iplayer.

    Some interesting reporting of those negotiations, Lib/Con and Lib/Lab, which led to the coalition agreement.
    Worth a listen, find out why Ed Balls concluded “No way, no chance, not in a million years”.

  45. @ Paul H-J

    A low turnout makes my argument stronger. A Get Out The Vote operation (GOTV, sorry for the jargon) is more effective when the overall turnout is low – it means a large bank of people not bothering to vote to aim at. My argument is that Lab and Con were much more effective at getting these people to vote by running an effective polling-day operation. (And unless you have ever been involved with one, you probably won’t realise how effective it is.)

  46. I think people intended to vote Lib-Dem but bottled it.

  47. Tony E
    Good point re LDs not dropping off in previous elections.
    I haven’t seen the Dec 2009 article quoted by Runciman but it was said to be based on British elections over last 40 years so there is likely to have been some evidence.
    I wonder if it is releavnt that this is the first election for a long time when one major party was not clearly going to get a majority. Maybe this would put 1979 through 2005 into a different category – no problem about voting LD then? Wonder what the polling evidence is re LD pattern in say 1970 or 1974 …

  48. @ Billy Bob @ 7:11

    Thank you for the suggestion.

    More details is learnt (it’s not difficult to put the storyline together if you make the report on four lines: Con, Lib, Lab and the one that seems to be the true).

    It also puts the coalition agreement (the long one) in a more defined context. It will be very difficult for the LibDems in the short-medium term. A lot will depend on the economy and the analysis by “independent” media with sufficient circulation/audiance.

    There are some wonderful political scenarios that could influence the polls: can Labour, without moving their ground put the coalition to the right; can the coalition claim that they are the centre (as they are clearly not there – too many committees and commissions). Can Labour reduce the differentiation between Con and Lib in the public’s eye (it will get help once the government vs. opposition discussions start – when will QT stop inviting both parties? – it won’t happen, I know) and what would be the outcome of that (depends on the economy), etc.

    I think any of these scenarios will create further upsets in the poll’s reporting on the voting preference for the LibDems – without signficant changes in the methodology.

  49. I nominate Ipsos MORI for the Worst Question of the Year award for “Do you think the new government will or will not …… be unable to make decisions?” (p51-53). It seems to have resulted in a large proportion of people misunderstanding, judging by the responses: Con 49 Will /46 Will Not, Lab 42/49, LD 44/52. Only the LD response is the “right” way round.

  50. I have a question for Anthony: did the exit poll try to take account of the postal vote, and if so how? If not, could its ‘accuracy’ be a bit of an accident? In other words, could ‘shy’ respondents on the day have been cancelled out by the postal voters having voted more decisively in the other direction.

    For example, let’s say that some voters were shy about admitting that they voted Tory or Labour, so that the exit poll on the day actually OVER-estimated LD support. But at the same time, postal voters may have voted MORE heavily for Tory and Labour than whose who cast ballots in person, so that the final result was close to the exit poll anyway.

1 2 3 9