Friday, January 9, 2015

Three different ways in which betting odds and betting markets have proved to be hopelessly unreliable predictors of election results

You've probably heard the claim that "democracies never go to war with each other".  Ironically, it's often used as an excuse for warmongering, because the theory is that democracies can eventually bring lasting peace to the world by forcibly "liberating" non-democratic countries.  And note that it's an absolutist claim, or a "golden rule" - there is supposed to be literally no example in history of an armed conflict between democracies.

That's utter garbage, of course - in reality there's a list as long as your arm of conflicts between democracies.  But in each and every case, the true believers in the golden rule have a get-out clause up their sleeve.  "That wasn't really a war, was it?  And that country wasn't a proper democracy."  Sometimes the logic becomes laughably circular - for example, a war causes instability, and no country can possibly call itself democratic if it isn't stable!

I'm reminded of that kind of nonsense when I hear people profess their absolute faith in the pseudoscience of 'the betting markets as an infallible predictor of election results'.  And I use the word 'infallible' advisedly, because you'll doubtless recall Betfair's breathtaking arrogance in feeling able to declare a few days before the referendum that No had already won, on the grounds that their exchange is "never wrong".  As a marketing stunt, they even paid out early on Sportsbook referendum bets.

One of the High Priests of this bizarre religion is self-styled superstar "risk assessor" Neil Edward Lovatt.  As regular readers may know, I had to mute Lovatt on Twitter a few months ago because he was wasting far too much of my time with his incessant trolling.  I eventually blocked him altogether when he started making some deeply unpleasant personal comments, and I also muted some of his more persistent risk assessor groupies.  However, they seem to multiply with water, and although I haven't had an exchange with Lovatt for months, my Twitter notifications timeline has recently become clogged up once again with people I've never heard of demanding that I worship at the altar of the betting markets, and warning me that if I don't, it will be proof that I'm a believer in homeopathic medicine (yes, really!).

To save myself the trouble of responding individually to these inane tweets for the remainder of time, here is a quick cut-out-and-keep guide to three examples that explode the myth of the infallibility of betting markets and betting odds.  In each case, there's a very different reason why they proved to be so embarrassingly unreliable.

Exhibit A : The 2006 Liberal Democrat leadership election.  There was a straightforward clash between conventional political wisdom, which identified Menzies Campbell as the obvious successor to Charles Kennedy, and the betting odds and markets, which counter-intuitively installed the little-known Chris Huhne as the favourite.  If the markets are infallible, or even if they're more reliable than other sources of information, then the money on Huhne should have been a very clear indication that something surprising was happening beneath the surface.  But there wasn't anything happening.  As it turned out, the conventional wisdom proved correct, and Campbell won at a canter.

Reason for the inaccuracy : It seems overwhelmingly likely that rich supporters of Huhne had placed a series of large bets to earn him the coveted 'favourite' tag, and to artificially generate momentum for his campaign.  This is the problem with the belief that the betting markets somehow reflect "the wisdom of crowds" - gamblers may be motivated by something other than the quest for a value bet, and in any case it only takes a few wealthy individuals (as opposed to crowds) to tip the balance.  We heard from the bookies during the closing stages of the referendum campaign that the vast majority of bets taken in Scotland were backing Yes to win, but the odds instead reflected a relatively small number of six-figure bets that had been placed on No, mostly south of the border.

Exhibit B : The 2010 UK general election.  A month before polling day, there was a clear divide between the polls, which were pointing to a hung parliament, and the betting markets, which pointed to a Conservative majority.  Again, if the markets reflect the "wisdom of crowds", they should have been able to easily outsmart the polls, which after all are snapshots and not predictions.  But for all their imperfections, you'd have been much better off believing the polls - the Tories fell twenty seats short of a majority.

Reason for the inaccuracy : Wishful thinking from wealthy Tory-supporting punters, who slipped into the groupthink of believing that the polls must be underestimating the Tories' chances, as they did in 1992.

Exhibit C : The 2007 Holyrood election.  This is my absolute favourite.  As you'll recall, because of difficulties with the counting machines, it took until 6pm on the day after polling for the final results to be declared, and for most of the intervening period, Labour were well ahead in the running seat tally.  The betting markets followed those numbers, and suggested that Labour had a 90%+ probability of winning.  However, as early as about 10 or 11am, BBC Scotland's political editor Brian Taylor had announced that it looked from intelligence on the ground that the SNP would end up marginally ahead by the end of the day.  There were a number of televised interviews throughout the afternoon with despondent Labour MSPs, who clearly had the same expectations of the final outcome.  And yet still - astonishingly - the "infallible" markets didn't budge.

Reason for the inaccuracy : The punters piling in with the most money, firmly believing they were earning themselves free cash, were based south of the border. They were getting their "information" from the London media, and were missing what was right under their nose - it didn't even occur to them to do something as obvious as check the live blog of Scotland's leading TV political journalist.  There must have been people in Scotland who made an absolute killing that day, simply by switching on their TV.

Lesson - the "wisdom of crowds" isn't of much use if the crowds are in the wrong country.  If you wanted to predict an election result in Poland, you wouldn't ask a crowd in Denmark, would you?

SNP and Plaid back up to 5% of the Britain-wide vote in today's YouGov poll

For the first time in 2015, the SNP and Plaid Cymru have reached 5% of the GB-wide vote in a YouGov daily poll...

Britain-wide voting intentions (YouGov, 7th-8th January) :

Conservatives 33% (+1)
Labour 33% (n/c)
UKIP 13% (-2)
Liberal Democrats 8% (+1)
Greens 7% (n/c)
SNP/Plaid Cymru 5% (+1)

And the Scottish subsample figures couldn't be more different from yesterday's : SNP 46%, Labour 24%, Conservatives 17%, Liberal Democrats 6%, UKIP 5%, Greens 2%.  We're in for greater day-to-day volatility now that YouGov are relying on much smaller Scottish samples, and the fluctuations this week are a good example of what to expect on a regular basis - the SNP's lead has varied from 6 points to 22.  (In fact, the volatility could well prove to be a lot worse than that.)

I must say I'm more confused than ever about the exact nature of YouGov's methodological change.  In his first clarification, Anthony Wells left the distinct impression that the only thing that had changed about the Scottish sampling was that they were sending out fewer invitations.  But yesterday he suggested that they are now using Scotland-specific target figures, just as they would for full-scale Scottish polls.  If that's true, I can't make any sense of the fact that SNP identifiers are still fairly obviously being downweighted according to Westminster-centric targets, rather than the political weighting being done by 2011 vote recall, as would happen in full-scale polls.  It may be that the Scottish sample is now demographically representative in a way that it hasn't been before, but if so, that good work is being largely undone by distorted party ID weightings, and by the smaller number of respondents.

Thursday, January 8, 2015

The broadcasters' logic for excluding the SNP from the leaders' debates lies in tatters as Ofcom rules that UKIP is "not a major party outside England and Wales"

If Ofcom's provisional ruling that UKIP only has major party status in England and Wales is upheld, this will be the position going into the general election campaign...

Conservatives : Major party status in England, Scotland and Wales but NOT in Northern Ireland.

Labour : Major party status in England, Scotland and Wales but NOT in Northern Ireland. (They don't even stand candidates in Northern Ireland on a minor party basis.)

Liberal Democrats : Major party status in England, Scotland and Wales but NOT in Northern Ireland.  (They don't even stand candidates in Northern Ireland on a minor party basis.)

SNP/Plaid Cymru : Major party status in Scotland and Wales but NOT in England and Northern Ireland.

UKIP : Major party status in England and Wales but NOT in Scotland and Northern Ireland.

As you can see, there is no such thing as a UK-wide party, so if that is the criterion for inclusion in UK-wide debates, there will quite literally be no-one there.  But the real problem here is that there are two parties or alliances that have major party status in two out of four constituent nations of the UK - and, as things stand, the broadcasters are proposing to invite one to the debates and exclude the other.

If this stands, it'll be hard not to conclude that the broadcasters see the so-called "UK debates" as in reality being Greater England debates, meaning that it's deemed perfectly natural for Scottish viewers to be bored to tears by a party that has major party status in England but not Scotland, but it's utterly unthinkable for English viewers to be expected to listen to a party that has major party status in Scotland but not England.

This quite simply isn't tenable.  It also means that if the SNP seek legal redress, there'll be a blatant double-standard for them to point out that wasn't there last time around.

*  *  *

UPDATE : There's a downright inaccuracy in the BBC website's report, which claims - "The regulator is seeking views on whether others should be included in its lists and and says it thinks UKIP will qualify across Great Britain." That is simply not true. The Ofcom consultation page clearly states that UKIP may qualify in England and Wales. There is no suggestion whatever that they may qualify "across Great Britain". It really does appear that some journalists literally cannot distinguish between England and Britain.

SNP lead by 12.5% in first Poll of Polls of 2015

As with every other Scottish subsample to have been published by YouGov since the referendum, today's has the SNP in the lead.  However, the gap is towards the lower end of the normal range we've become used to...

SNP 34%, Labour 28%, Conservatives 16%, UKIP 11%, Liberal Democrats 9%, Greens 2%.

Doubtless one or two of the minions at McDougall Central will be hoping that this is an early sign that Jackanory Jim's bizarre antics might finally be gaining some traction with the public.  I think that's unlikely, though, because Labour's vote share isn't untypically high.  Instead there are more UKIP supporters than usual in the sample, plus the Liberal Democrats are doing a little better.

We can probably expect bigger fluctuations from day to day now that YouGov are relying on much smaller Scottish samples - for the first time since heaven-knows-when, the whole Scottish sample has actually had to be upweighted in today's poll (from 139 to 149).  In spite of that, SNP identifiers have still been downweighted, albeit by less than usual.

We've now had enough polls this week to make it just about worthwhile updating the Poll of Polls for the first time in 2015.  Take the following numbers with an extremely heavy dose of salt, though, because they're based on just four subsamples - three from YouGov and one from Populus.  As always, the Opinium poll can't be used because the Scottish subsample wasn't published.

Scottish voting intentions for the May 2015 UK general election :

SNP 39.8% (-4.3)
Labour 27.3% (+0.8)
Conservatives 15.5% (-1.0)
Liberal Democrats 6.8% (+1.0)
Greens 5.5% (+3.8)
UKIP 4.5% (+0.5)

(The Poll of Polls uses the Scottish subsamples from all GB-wide polls that have been conducted entirely within the last seven days and for which datasets have been provided, and also all full-scale Scottish polls that have been conducted at least partly within the last seven days. Full-scale polls are given ten times the weighting of subsamples.)

Wednesday, January 7, 2015

SNP maintain 15% lead in second YouGov subsample of 2015

Today's second GB-wide YouGov poll of the year shows the Tories drawing level once again, while UKIP slip back still further...

Britain-wide voting intentions (YouGov, 5th-6th January) :

Conservatives 33% (+2)
Labour 33% (-1)
UKIP 13% (-1)
Greens 8% (n/c)
Liberal Democrats 7% (n/c)
SNP/Plaid Cymru 4% (n/c)

I think James may have been onto something in the previous thread with his theory about how the sudden slippage in the UKIP vote may be an illusion.  Many YouGov respondents have completed countless voting intention surveys from the firm in the past, and it could be that UKIP supporters have become used to selecting the "some other party" option, and have simply carried on doing that without noticing that UKIP have now been moved on to the main menu.  This theory is consistent with the unusually high support in yesterday's poll for "others" - a grouping which excludes even Respect and the BNP.

Today's Scottish subsample figures are : SNP 38%, Labour 23%, Conservatives 19%, Greens 9%, Liberal Democrats 7%, UKIP 3%.  If we were meant to discern a new pattern in the Scottish subsamples as a result of YouGov's methodological change, it certainly hasn't become apparent so far - the SNP are retaining an entirely familiar lead, but the Tories and Lib Dems have both bounced back from yesterday's abysmal lows.

Having read Anthony Wells' more detailed explanation of YouGov's tweak, I suspect all that's happened is that they've started to send out fewer invitations to respondents in Scotland, to prevent them having to continually downweight the Scottish sample as a matter of routine.  Ironically, that will simply make the Scottish subsample figures somewhat less reliable than before.

Even though the Scottish sample as a whole hasn't had to be downweighted at all in today's poll, respondents who identify with the SNP have, as per usual, been downweighted sharply (from 55 to 34) to bring them into line with Westminster-centric target figures.

Tuesday, January 6, 2015

Early straw in the wind from YouGov suggests that the SNP have emerged from the festive period with their big lead intact

YouGov have released their first (Britain-wide) poll of 2015, breaking a two-week drought...

Britain-wide voting intentions (YouGov, 4th-5th January) :

Labour 34% (-2)
Conservatives 31% (-1)
UKIP 14% (-2)
Greens 8% (+3)
Liberal Democrats 7% (+1)
SNP/Plaid Cymru 4% (n/c)

The biggest oddity is the drop in UKIP's vote, which has happened in spite of a major methodological change that ought to have boosted the reported vote for the party, arguably by quite a bit.  YouGov have started prompting for UKIP along with the other "main" parties, rather than sending respondents to a second menu of options if they say they plan to vote for "some other party".  However, there are two unusual factors at play in the poll - the first is the methodological change, and the second is the Christmas interruption.  So it's not totally impossible (albeit probably unlikely) that prompting for UKIP is indeed having an effect, but that a substantial drop in the UKIP vote over the last fortnight is more than offsetting the impact.

Most important for our purposes is of course the Scottish subsample, which is showing : SNP 46%, Labour 30%, Conservatives 12%, Greens 6%, Liberal Democrats 3%, UKIP 2%.  The 16-point SNP lead is entirely typical of the results we were seeing between the referendum and Christmas.   When considered in combination with yesterday's Populus subsample, it increases the likelihood that the festive break has had no detrimental effect on the SNP's fortunes.

Intriguingly, though, today's subsample is not directly comparable with the pre-Christmas subsamples, because YouGov seem to have followed Angus Reid's good example by introducing a measure of Scotland-specific weighting (or at least sampling) for the first time - the exact words are "we are controlling our sampling in London and Scotland more carefully".  It's also cryptically added that "anyone who regularly studies our crossbreaks may notice a difference within them", without any clue as to what that difference is likely to be.  My guess is that we may be looking at grim news for the Scottish Tories and the Scottish Lib Dems in particular, because both parties have an especially low share of the vote today.  That could add weight to the seemingly outlandish projections from full-scale Scottish polls that suggest the Tories might just lose David Mundell's seat to the SNP.  Whether there's going to be any impact on the national gap between the SNP and Labour is harder to say, although the vote share for both is a bit higher than the recent average.

The party system in London is not as distinct as Scotland's, but presumably the logic for London-specific sampling is the bucking of the England-wide trend that was seen in the city in the European election results, with Labour doing particularly well and UKIP doing particularly badly.

From today until the general election, we should never again have to wait more than 48 hours for a fresh poll - as far as I can remember, YouGov don't take an Easter break or anything like that.

*   *   *

UPDATE : Bizarrely, the former Labour MP Nick Palmer (who is a decent sort, not generally known for trolling) has leapt on the change in YouGov's methodology, and claimed that "the Scottish subsample today shows a narrower gap than usual".  All I can do is point out the bleedin' obvious - that's simply not true.  It's the opposite of the truth.  A 16-point SNP lead is within YouGov's normal range, and if anything is towards the higher end of it.

I can only guess that Nick is so fixated on Labour's vote that he automatically assumes that a slightly higher Labour share must mean a lower SNP lead.  It doesn't.

Monday, January 5, 2015

If Scottish Labour did seduction...

"When you move in with me, I'll buy you one more box of chocolates per year than that bastard boyfriend of yours buys you."

"But he doesn't buy me chocolates."

"Thank Christ for that, I'm a bit short on cash anyway."

*  *  *

Thanks to James on the previous thread for pointing out that Populus have published their first GB-wide poll of the year - in fact it's the first published poll from any firm to have been entirely conducted in 2015.  The Scottish subsample figures are : SNP 41%, Labour 28%, Conservatives 15%, Liberal Democrats 8%, Greens 5%, UKIP 2%.  Of the firms that poll regularly, Populus have consistently been the most favourable for Scottish Labour, so the SNP lead in this subsample is very much at the upper end of the recent 'normal range' - and that's in spite of the fact that (as usual) respondents who identify with the SNP have been sharply downweighted.

So we've now had two straws in the wind since the New Year - one from Opinium, suggesting a slightly lower SNP vote than usual, and one from Populus, showing the reverse.  The mists may clear a little overnight with the first YouGov poll of the year.

Gordon Bull

I haven't watched as much darts in years and years as I have over the last few days, but I chose the right time for it, with Gary Anderson becoming the first Scot to win either of the two world titles since Les Wallace way back in 1997.

Wayne Mardle is a strong contender for the coveted title of 'Most Irritating Sports Commentator Whose Name Is Not Clive Tyldesley', but nevertheless I'd quite like to hear him commentating on the Gordon declaration in the early hours of May 8th...

"Oh, he couldn't.

He couldn't, could he?


Of course he could!

He's Alex Salmond!

That's what he does!

All day long!"

Sunday, January 4, 2015

First Britain-wide poll of 2015 suggests that the SNP are probably still ahead in Scotland

This is always a scary time of year for followers of the polls.  There are rarely any polls published over the Christmas/New Year period, and there's no reason to automatically assume that nothing will have changed by the time that drought is broken.  The very fact that people switch off completely from the news for a few days can sometimes, ironically, shift their voting intentions quite markedly.  The parties that have the most to worry about that phenomenon this year are the SNP, who built up such an enormous lead in Scotland over the closing months of 2014, and Labour at GB-wide level, who pulled a few points clear of the Tories in most of the polls conducted just before Christmas.

The first GB-wide poll of the New Year has just been published by Opinium...

Britain-wide voting intentions (Opinium, 30th December - 2nd January) :

Labour 33% (-3)
Conservatives 32% (+3)
UKIP 17% (+1)
Liberal Democrats 8% (+2)
SNP 4% (-1)
Greens 4% (-1)
Plaid Cymru 1% (+1)
BNP 1% (n/c)

So it seems, on the face of it, that Labour's worst fears have been confirmed - their 7-point pre-Christmas lead has essentially evaporated, although only time will tell whether that is a real shift of opinion or simply an extreme example of margin of error "noise".

With the SNP it's even harder to say anything definite, because (irritatingly) Opinium are the only firm that don't publish the results of their Scottish subsamples.  The slippage from 5% to 4% certainly isn't anything to worry about, because the figure has consistently been either 4% or 5% in every Opinium poll since the referendum.  However, the unrounded figure in this poll is 3.62%, which is a touch lower than in other post-referendum Opinium polls - albeit we're obviously talking about tiny fractions here, given that this is a GB-wide poll.

If you assume that all respondents who said they would vote SNP are resident in Scotland, and if you assume that Scotland's share of the Opinium sample was equivalent to our actual share of the GB population, that would suggest the SNP are on 41.3% of the vote in Scotland, which would almost certainly mean a commanding lead over Labour.  Unfortunately it doesn't necessarily work like that, because Opinium (unlike YouGov) filter their results by likelihood to vote, so it's possible that Scotland makes up more than a proportionate share of the filtered sample.  It's also likely that at least one or two people who said they would vote SNP don't actually live in Scotland.

The most logical conclusion to draw is that the SNP are probably ahead in the Scottish subsample, but maybe not by quite as wide a margin as in other post-referendum Opinium subsamples.  But the margin of error for any individual subsample is huge, so we'll have to await the findings of other pollsters to discover whether this is just a statistical blip.