Saturday, April 30, 2016

Desperate Dugdale ducks for cover as latest polls reveal the Scottish people's message to Labour is simple and stark : "Exterminate! Exterminate! EX-TER-MI-NAAAAAATE!!!!!!"

Apologies for the three days of radio silence, during which we've had a couple of new Holyrood polls with slightly conflicting messages.  The TNS poll shows the SNP slipping down to comfortably their lowest share of the vote on both ballots since the UK general election one year ago.  It is, of course, the TNS figures that have been most commonly prayed in aid by the "tactical voting on the list" lobby in recent months, so the new poll vindicates the caution that many of us have been applying.  It looks like TNS may simply be converging with other firms that have never reported quite such an inflated SNP lead, because the new Ipsos-Mori poll still has the SNP very much within the firm's normal range (at least on the constituency ballot).

Constituency ballot (TNS) :

SNP 52% (-4)
Labour 22% (+1)
Conservatives 17% (+2)
Liberal Democrats 7% (+1)

Regional list ballot (TNS) :

SNP 45% (-2)
Labour 22% (+1)
Conservatives 18% (+3)
Greens 8% (n/c)
Liberal Democrats 5% (-1)

Constituency ballot (Ipsos-Mori) :

SNP 51% (-2)
Labour 19% (-1)
Conservatives 18% (+2)
Liberal Democrats 6% (n/c)

Regional list ballot (Ipsos-Mori) :

SNP 45% (-4)
Conservatives 19% (+4)
Labour 17% (-2)
Greens 10% (+4)
Liberal Democrats 7% (-1)

The fact that TNS still have Labour ahead of the Tories by a decent margin shouldn't necessarily lessen any sense of panic in the Dugdale bunker over the Ipsos-Mori findings.  The bulk of the TNS fieldwork was conducted before the Ipsos-Mori poll got underway, so it's possible that Ipsos-Mori are detecting a relatively recent swing to the Tories.  A more comforting thought for Labour is that Ipsos-Mori's constituency findings may be a better guide than the list findings to how the party is likely to fare on the all-important list ballot.  We know that constituency polling has tended to be more accurate than list polling over the years, and that a minority of poll respondents tend to misinterpret the question about list voting as if they're being asked for some kind of second preference vote.  To believe that Ipsos-Mori are exactly right about Labour being in second place in the constituencies but not on the list, you'd have to think that Labour voters are more likely to drift off to the Greens on the list than Tory voters are to drift off to UKIP or wherever, and I can't see any compelling evidence for that being the case.  That said, even if Ipsos-Mori's constituency numbers are the better guide, the race for second place suddenly looks like a virtual coin toss - unless of course the Tories are being flattered in this poll because of methodological issues or normal sampling variation.

Explanatory note : As a fond tribute to the mainstream media's restrained take on the GERS report, Scot Goes Pop headlines will feature 50% bonus hysteria for an indefinite period.

Wednesday, April 27, 2016

NEW FILM : How the voting system for the Scottish Parliament works, and why "tactical voting" is such a big risk

A couple of weeks ago, I jetted off (if you can jet off by train) to a top-secret mystery town somewhere between Dunblane and Bannockburn (it's got a big castle in it) to record a video about the Holyrood voting system for the brilliant Phantom Power.  The intention was very much to avoid this being a "propaganda" film - although I do explain my view that so-called "tactical voting on the list" is an enormous risk, I only do that in the final few minutes.  I've felt all along that the main reason some people find the arguments of the tactical voting lobby so seductive is that the basic principles behind the AMS voting system aren't widely understood.  So the bulk of the video simply explains in relatively neutral terms how AMS is supposed to work (with the composition of parliament determined by the list vote and not the constituency vote), but how the Scottish version of the system makes things a little less simple than they should be.  I then describe the theory of how "tactical voting on the list" is meant to deliver more pro-independence MSPs, before offering my own opinion that it's an extremely bad idea, and could easily backfire badly.

If you have any problem with the embedded video at the top of this post, the direct YouTube link is HERE.

Also, if you're looking for a video in which both sides of the argument get an equal airing, don't forget my Independence Live debate with Tommy Sheridan from a few weeks ago - that can be viewed HERE.

Tuesday, April 26, 2016

Latest EU referendum telephone poll shows slight improvement for Leave

It's been ages since I last posted anything about EU referendum polling, which is probably just as well, because I've been exasperated at the acres of drivel that have been written on the subject over the last couple of weeks.  It's true that there is some evidence (albeit not quite conclusive) that there was a swing back to Remain in mid-April, but if it did happen, it was nowhere near as dramatic as some of the commentary would have had you believe.  I saw one headline about the most recent Ipsos-Mori poll, which read "Remain now leads by TEN points", as if that's a figure that should have made our jaws drop to the floor.  But I just thought : what do you expect?  It's a telephone poll.  Yes, ten points represents a slight increase in the Remain lead, but it's not quite such a devastating blow for the Leave campaign when you recall that they were eighteen points behind with Ipsos-Mori as recently as February.

On the other extreme, we also had commentary that placed far too much emphasis on polls that didn't even ask for referendum voting intention, and yet purportedly showed that Obama's intervention on behalf of Remain had backfired.  You really can't read much into a poll showing that the majority of people think that what Obama said is irrelevant, because for all we know the substantial minority who take the opposite view could easily be the crucial swing voters who will decide this referendum.

As it turns out, the first partly-post-Obama poll out today shows a small improvement for Leave.  But the change is well within the margin of error, and unfortunately the poll is the latest in the ORB telephone series, which so far seems to have been fairly detached from the trend shown by all other polling.  The most that can be said is that today's findings somewhat decrease the chances that we'll eventually conclude that Obama's words had the intended effect.  But the jury is still very much out on that.

Last but not least, I absolutely despair at the way in which changes in John Curtice's Poll of Polls have been breathlessly reported by some journalists without the appropriate health warnings about the ever-shifting balance between telephone and online polls in his sample.  On a meaningful pound-for-pound comparison, the current 54% to 46% lead for Remain in a sample almost entirely comprised of telephone polls is actually pretty similar to the previous 50-50 split in a sample that was almost entirely comprised of online polls.  Whatever criticisms might be made of the method I use for the Poll of Polls on this blog, at least it avoids that kind of totally illusory effect.

*  *  *



Remain 44.6% (-0.6)
Leave 39.7% (-1.7)


Remain 39.6% (-2.3)
Leave 39.1% (-3.3)


Remain 49.5% (+1.0)
Leave 40.3% (n/c)

(The Poll of Polls takes account of all polls that were conducted at least partly within the last three weeks. The online average is based on seven polls - three from YouGov, two from ICM and two from TNS. The telephone average is based on six polls - two from ComRes, two from ORB, one from ICM and one from Ipsos-Mori.)

Sunday, April 24, 2016

Reader's questions : a Patrick Harvie conundrum

Niall posed an interesting question on the previous thread, which I thought might be worth responding to in a bit of detail...

"You know, these kinds of discussions usually take place between people who post on blogs about politics, and one characteristic that unites these people is that they are much more likely than your average citizen to be hyper-partisan.

So let me come to the discussion from another angle. I was speaking last week with my aunt who is a floating but generally centre-left voter. She lives in Glasgow Kelvin (note that this is the only part of Glasgow where the Greens are putting up a candidate: Patrick Harvie).

Her thoughts "I think the Greens have good policies and I like Patrick Harvie so I will vote for him for my MSP. But I will also vote SNP on the regional list because I want to see Nicola Sturgeon back as first minister"

I remarked that considering her two objectives it would make more sense, because her votes would be more efficient and effective, to vote the other way around: SNP on the constituency ballot, Green on the regional ballot. Was that wrong?"

My own view is that Niall is half-right and half-wrong. It's certainly true that Patrick Harvie doesn't have a hope in hell of actually winning the Glasgow Kelvin constituency outright (he's presumably standing as a candidate to make an investment for the future) and that his chances of returning to Holyrood as an MSP depend entirely on list votes. So, if your number one priority is getting Harvie re-elected, you should vote Green on the Glasgow regional list. But that's the nub of the matter - you really have to choose a number one priority, because the Additional Member System (unlike the Single Transferable Vote system used in local elections) doesn't really lend itself to pursuing two entirely different priorities simultaneously, except in a relatively narrow range of circumstances. If the first priority of Niall's aunt is instead to see Nicola Sturgeon as First Minister, she should probably vote SNP on both ballots. There really is no way of giving equal weight to both of her specific objectives, because depending on how the election result turns out (and that's unknowable at the moment you actually cast your vote) a Green list vote could harm the SNP. Remember the example of the North-east in 2011, when a relatively small amount of switching from SNP to Green/SSP/Solidarity could have cost the SNP the final list seat, and handed it to the Tories.

You might be looking for an example of the small number of circumstances that allow a voter to do something a little more sophisticated with AMS. Well, in a normal state of affairs, when no party is too dominant in the constituencies (ie. not this year), you can be reasonably confident that how you vote in any individual constituency is pretty unlikely to affect the overall composition of parliament. For example, if you live in an SNP/Conservative marginal constituency, the only effect of the Conservatives winning the seat would probably be to reduce the number of Tory list seats by one, and increase the number of SNP list seats by one - thus leaving the overall result totally unchanged. That might offer you a relatively 'free hit' on your constituency vote if you have a strong preference between the individual candidates. Imagine you're in Paul Kavanagh's situation, and can't bring yourself to vote for your local SNP candidate because of his views on gay marriage, or whatever. Voting against the SNP on the constituency ballot, but for the SNP on the all-important list ballot, might conceivably help 'replace' the candidate you don't like with a more palatable SNP candidate from the list. But that sort of opportunity is quite rare, because you don't have any control whatever over the ranking of the SNP's list candidates - that's predetermined by an internal ballot of SNP members. And even when an opportunity does crop up, there are still risks involved - albeit nowhere near as extreme as the risks of "tactically" giving your list vote to a fringe party like RISE, which is highly unlikely to win any list seats at all.

Just for clarity, what I've outlined above does not apply in this particular election.  Because the SNP have a chance (and it is only a chance) of total constituency dominance, voting against the SNP on the constituency ballot might well cost the party "bonus" seats that they won't be compensated for on the list.

* * *


Three new polls have been published since I last updated the Poll of Polls, with BMG entering the sample for the first time. The percentage changes are therefore a tad less glacial than normal.

Constituency ballot :

SNP 52.7% (+0.5)
Labour 19.7% (-0.7)
Conservatives 16.8% (+0.4)
Liberal Democrats 5.8% (n/c)

Regional list ballot :

SNP 46.2% (-0.4)
Labour 19.0% (-0.4)
Conservatives 16.8% (+0.6)
Greens 8.0% (+0.6)
Liberal Democrats 6.0% (-0.6)

(The Poll of Polls is based on a rolling average of the most recent poll from each of the firms that have reported Scottish Parliament voting intention numbers over the previous three months, and that adhere to British Polling Council rules. At present, there are six - Panelbase, Survation, BMG, YouGov, TNS and Ipsos-Mori. Whenever a new poll is published, it replaces the last poll from the same company in the sample.)

One week on, have the Sunday Herald learned the lessons of their wildly inaccurate reporting on so-called "tactical voting on the list"?

The short answer to that question is "only partly".  As editor Neil Mackay hinted would probably be the case, they've gone big on the subject for a second Sunday in a row, with the lead story once again penned by Peter Swindon.  It's basically another propaganda piece that is pretty unsubtle in attempting to steer readers towards abandoning the SNP on the list, but Swindon has chosen his words somewhat more carefully this time, and in particular has ditched the pretence that John Curtice actually "advised" people to vote Green or RISE. He's also replaced all traces of the totally inappropriate word "predict" with the much more sound "project".  The problem, however, is that he mostly writes as if the two words are basically synonymous, ie. "[Curtice] projected that the SNP will win all but three constituencies". That means the impression conveyed is almost as misleading as last week - but a very hurried caveat of "if an average of recent polls are correct" at least ensures that the new article can technically be claimed not to be inaccurate.

To be fair, there are some contexts in which "project" can be used as if it means exactly the same thing as "predict" - an example would be "CNN projects that Bernie Sanders has won the Michigan primary", which is a call that is never made until it is felt there is near-certainty. But that's not the sort of thing we're talking about in this case. John Curtice's report collated the results of opinion polls which were conducted well before the election, and projected what the election result would be if : a) those polls were bang-on accurate, and b) no changes whatever in public opinion occurred between the dates on which they were conducted and polling day. That's more akin to taking the pre-election polls in Michigan and using them to "project that Hillary Clinton will win comfortably", without bothering with the irritating formality of counting any votes. Indeed, as regular readers of this blog know, that sort of projection was actually made, and it was utterly meaningless - Sanders defied polls giving Clinton a double-digit lead, and won by a whisker.

Do we live in a world where that magnitude of polling fiasco only occurs once in a blue moon? Och, don't be silly. Last year's UK general election was supposed to result in a nailed-on certain hung parliament - but the Tories won an overall majority. Just a few weeks earlier in Israel, everyone thought Benjamin Netanyahu was a dead man walking, but on election night he dismayed the world with a famous victory that bore no resemblance at all to the pre-election polls. And although the Liberals certainly looked set for a return to power in Canada last autumn, the polls offered Justin Trudeau little hope of an overall majority. Come election night, he won a comfortable majority of 30.

Now, perhaps you're on the brink of offering the familiar objection that the polls may have severely misled us in many contests around the world last year, but they did pretty well here in Scotland. OK, so let's take that observation to its logical conclusion. Presumably what you'd be getting at is that the polls can always be relied upon to give an accurate indication of election results in Scotland. That would explain, for example, why we all knew well in advance that the SNP were going to win an overall majority in 2011. Oh, wait. We didn't know that at all, did we? In fact, every single projection in 2011 based upon the pre-election polls put the SNP short of a majority. It was occasionally projected that there might be a slender pro-independence majority consisting of the SNP, Greens and Margo MacDonald, but even that was thought to be reasonably unlikely.

So, no, there is no Scottish exceptionalism that means we don't have to allow for the very significant possibility that the polls and seats projections based on polls are leading us astray. Professor Curtice knows that as well as anyone - he's repeatedly pointed out on his blog that polling on the list was significantly less accurate in 2011 than constituency polling (thus explaining why nobody saw the SNP majority coming), and also that the Greens have historically been substantially overestimated on the list. And that's exactly why he went out of his way in his report to indicate that his projection based on the polls was not a prediction of the election, or functionally indistinguishable from a prediction. He specifically entertained the possibilities that the polls may be understating or overstating the SNP's support, and noted that in either scenario, attempts by SNP supporters to vote "tactically" on the list could easily backfire. And that was absolutely not, as somebody tried to make out in a comment on this blog the other day, some kind of routine, going-through-the-motions "health warning" along the lines of "we all know that the polls will probably be broadly right, but we can't totally exclude the possibility that they won't be". He was deadly serious in what he said - nobody who read the report could be in any doubt that he regarded polling error as a significant risk. That was the whole basis for his verdict that pro-independence voters face a "dilemma" on the list ballot.

Last week, the Sunday Herald totally ignored that side of the Curtice report - they literally pretended it didn't exist. This week, they've belatedly acknowledged it, but are downplaying it for all they're worth. They have a direct quote from Curtice, which in terms of tone seems to offer some comfort to their original misrepresentation of the report, but a close reading confirms that his position hasn't changed one iota. Forgive my cynicism, but I'm inclined to suspect that they conducted quite a long interview with him, and cherry-picked the quote in which he happened to place the weakest emphasis on "risk" and "dilemma". Even so, those words are still very much present.

The other experts quoted in the article are Ipsos-Mori's Mark Diffley and Dr Craig McAngus from Aberdeen University. In contrast to Curtice, I don't have any suspicions that they've been quoted selectively, because what they say seems fairly clear-cut, although obviously it's impossible to know whether the Sunday Herald kept phoning around until they found experts who said exactly what they wanted to hear (perish the thought!). I just think Diffley and McAngus are plain wrong, and some of what they say doesn't actually make any logical sense. Diffley very unwisely uses Swindon-like language and says that Curtice's projection shows that the SNP "will" only get two regional seats. I'm sorry to have to make the obvious point here, but Ipsos-Mori were as wrong as every other pollster last year - a projection based on their final poll in May 2015, and couched in Diffley's choice of words today, would have read : "Ed Miliband will be Britain's new Prime Minister". I'm sure that would make a great alternative history novel, but it wasn't much cop as a prediction.

McAngus, meanwhile, makes the extraordinary statement that "To vote for another party other than the SNP on the list as a pro-independence voter is a rational thing to do. I mean, the way that the numbers stack up, the SNP are not going to win a lot of seats on the list." I can only conclude that he actually hasn't thought this through for more than five seconds, because what all the polls are unanimous on is that the SNP are on course to win some list seats, and that both RISE and Solidarity are light-years away from winning even one list seat. So yes, let's talk about rationality - is McAngus seriously saying that it's more rational to cast a "tactical" vote that seems guaranteed to be wasted than it is to stick with the SNP and at least have a chance of contributing towards the election of a pro-independence MSP on the list? Perhaps we could be charitable and assume that "another party other than the SNP on the list" is code for the Greens, and definitely not for RISE or Solidarity - but a) that's not my impression of what he's saying, and b) even if that is what he means, it's deeply irresponsible of him not to spell it out. It would, in any case, still be a very weak argument, because while the polls offer grounds for optimism that list votes for the Greens will not be wasted in at least some regions, they fall well short of providing conclusive evidence that the Greens are set to win more list seats than the SNP.

I'm also a tad bemused by the article's suggestion that Diffley and McAngus have "backed" Curtice's projection. As the projection simply takes the raw poll numbers and directly converts them into a hypothetical number of seats, "backing" the projection amounts to no more than agreeing that Curtice is adept at using a calculator. In that sense, I'm more than happy to "back" the projection as well, but I'm not sure what that's supposed to tell us in concrete terms.

Finally (and this has nothing to do with the Sunday Herald piece), I want to tackle a line of argument that Tommy Sheridan used in the debate I had with him, and that seems to be taking root in some other quarters as well. Basically he claimed that if the SNP win all nine constituency seats in Glasgow, the d'Hondt formula will ensure that "nine out of every ten SNP list votes will be wasted, and that only one in ten will actually count". This is complete garbage, because it wrongly implies that d'Hondt literally throws away a certain number of votes, which are then not taken into account at any point during the distribution of list seats. In fact, the d'Hondt calculation is made afresh for every individual list seat, and is always based on the actual number of votes cast. The following hypothetical example in which only two parties are involved will hopefully illustrate what I mean.

Actual list vote :

SNP 99,950 votes
Solidarity 10,000 votes

Because the SNP won all nine constituency seats in the region, the d'Hondt formula divides their raw list vote by ten (nine plus one) in the count for the first list seat. Because Solidarity didn't win any constituency seats, their raw vote remains unaltered.

D'Hondt-adjusted count for first list seat :

Solidarity 10,000 votes
SNP 9,995 votes

So Solidarity take the first list seat, and on the face of it the SNP's advantage in terms of votes has been completely wiped out. If you believe Tommy Sheridan's notion about 90% of the SNP's original list votes being literally thrown away, you'd be forgiven for thinking that the two parties are now on an equal footing as far as the distribution of the remaining six list seats are concerned, and perhaps will win three apiece. But that isn't the case at all, because for each remaining seat, the d'Hondt formula refers back to each party's actual number of votes. For the second list seat, Solidarity's original vote is divided by two (one plus one), because they now have one seat under their belt. The SNP's vote remains divided by ten.

D'Hondt-adjusted count for second list seat :

SNP 9,995 votes
Solidarity 5,000 votes

For the third list seat, the SNP's original vote is divided by eleven (ten plus one), because they now have ten seats. Solidarity's vote remains divided by two.

D'Hondt-adjusted count for third list seat :

SNP 9,086 votes
Solidarity 5,000 votes

For the fourth list seat, the SNP's original vote is divided by twelve (eleven plus one), because they now have eleven seats. Solidarity's vote remains divided by two.

D'Hondt-adjusted count for fourth list seat :

SNP 8,329 votes
Solidarity 5,000 votes

For the fifth list seat, the SNP's original vote is divided by thirteen (twelve plus one), because they now have twelve seats. Solidarity's vote remains divided by two.

D'Hondt-adjusted count for fifth list seat :

SNP 7,688 votes
Solidarity 5,000 votes

For the sixth list seat, the SNP's original vote is divided by fourteen (thirteen plus one), because they now have thirteen seats. Solidarity's vote remains divided by two.

D'Hondt-adjusted count for sixth list seat :

SNP 7,139 votes
Solidarity 5,000 votes

For the seventh and final list seat, the SNP's original vote is divided by fifteen (fourteen plus one), because they now have fourteen seats. Solidarity's vote remains divided by two.

D'Hondt-adjusted count for seventh list seat :

SNP 6,663 votes
Solidarity 5,000 votes

As you can see, and entirely contrary to Sheridan's claim, the fact that the SNP has ten times as many list votes as Solidarity matters enormously. It ensures that after the initial big hit on the first count, the SNP's vote only comes down very, very gradually in subsequent counts. By contrast, Solidarity never recover from having their vote cut in half on the second count, and they lose ALL SIX remaining list seats to the SNP. Final number of list seats in this hypothetical scenario : SNP 6, Solidarity 1.  So much for nine-tenths of those SNP list votes "not counting", eh?