Friday, September 19
For literally anything you need to know about what to expect and when, Antony Green’s guide can’t be beat. Basically, we can’t expect any official results until about 3pm AEST, though presumably some manner of informal indication of how the count is going might emerge. There will not be any exit polls, but there is talk that a retrospective opinion poll might be conducted (how did you vote rather than how will you). The one final poll out was, as noted below, from Ipsos MORI, which was a phone poll of 980 respondents showing yes on 45% and no on 50%, rounding out to 53-47. I couldn’t be bothered running the poll tracker charts again because the result of the poll after bias adjustment was right on the trend, at 51.2% for no. My personal feeling is that no is likely to do it a little more easily than that, but only time will tell. The map to the right is derived from regional level polling data over the past few weeks from Survation and ICM, to give at least a broad-brush idea of where the independence cause is weakest and strongest. The actual results will emerge at the level of Scotland’s 32 local government authorities at the link above, Antony Green maps their past voting behaviour, since support for the Scottish National Party is very likely a good proxy for how the referendum vote will go.
Thursday, September 18
A few hours after polls open, one final poll from Ipsos MORI 53-47 to no. Ninety-five per cent report they will vote, which by early reports of turnout can almost be believed.
Okay, to tidy up the mess below: we’ve had 51-49 from the very authoritative Ipsos MORI, a phone poll with a sample of 1405; two online polls at 52-48, one from YouGov with a bumper sample of 3237, the other from Panelbase with 1004; and a phone poll from Survation, for which I assume the sample was about 1200 given such was the case in its only previous phone poll, at 53-47. The Survation survey in particular is very fresh, having been conducted entirely within the last 24 hours. Pumping all that into the poll aggregate is slightly better for yes than you might think, since Ipsos MORI and YouGov both get bias adjusted about 1.4% towards yes, and weighted heavily in the overall result. I’m a bit nervous about this – those bias adjustments seem excessive – but the current reading, which you may take or leave, is 51.2% no, 48.8% yes.
UPDATE 4: Survation has it at 53-47.
UPDATE 3: YouGov has it at 52-48. And the fun’s not over, because Survation have just revealed they’re about to lay on a surprise phone poll.
UPDATE 2: Mike Smithson of Political Betting tweets: “So 4 pollsters have NO on 52% and 2 on 51%. Maybe they are wrong but at least they are all wrong together”. The implication, I believe, being that we may be seeing a little bit of herding going on.
UPDATE: The Ipsos MORI poll turns out to be a nailbiter 51% no, 49% yes. This is important because it’s a phone poll rather than online, and Ipsos MORI did a particularly good job of calling the last Scottish parliamentary election. YouGov to come in a few more hours, and then I’ll give the poll aggregate another run.
A new poll from Panelbase does nothing to relieve the monotony, once again producing a result of 52-48 in favour of no. However, this is the best 52-48 so far for the no camp, as Panelbase has been the most yes-leaning of the regularly reporting pollsters. The full numbers are 49.5% no, 45.4% yes, 5.1% don’t know. Full results here. The eagerly awaited Ipsos MORI poll will be along at 3am EST, followed by the final YouGov poll at 5am.
Wednesday, September 17
Three new polls have come in overnight from Survation, Opinium and ICM and every one of them finds no with a lead of 52-48. My poll tracker (methodology explained in the entry below from Saturday) now has no leading 51.7% to 48.3%, slightly higher than the 51.4% to 48.6% recorded following Sunday’s polls. More tellingly, the trendlines provide a fairly clear indication that the momentum to yes which was evident over a period of weeks has tapered off:
Of the three pollsters to have reported new results, Survation has the no lead narrowing from its result on Sunday, which had it at 54-46; Opinium also narrows slightly from a poll on Sunday, which had it at 53-47; while ICM is much better for no this time around, its previous poll being a small-sample outlier with yes leading 54-46. All three polls were conducted online; Mike Smithson at Political Betting says that, based on past form, the one we should be hanging out on is tomorrow’s final phone poll from Ipsos MORI. I had a paywalled piece in Crikey yesterday considering the possibility that the polls might just have it all wrong.
Some further findings from Opinium: 50% do not trust that new devolution powers in the event of a no vote will be delivered as promised; 47% think Scotland will keep the pound against 37% who don’t; 44% think Alex Salmond should resign as First Minister if no wins (which I find very odd); and 45% think independence will damage the Scottish economy.
Sunday, September 13
No less than four new polls have reported overnight, of which two have “no” with reasonable solid leads of six or eight points, one is lineball, and one is the best poll yet to emerge for yes. These are reviewed in detail below, but first we take an updated look at the poll tracker. This puts the current result at 51.4% for no and 48.6% for yes, all but unchanged on the 51.2% and 48.8% recorded yesterday (based on like-for-like methodological comparison). An outline of the methodology was provided yesterday, in the bottom half of this post.
The polls in turn:
The good news for the independence camp first: ICM has produced the second poll to show the yes vote in front, following on from the first of last week’s surveys by YouGov, and by a not inconsiderable margin 49% to 42%, rounding out 54-46 after exclusion of the undecided. Unlike yesterday’s ICM poll, this one uses its usual online methodology. The caveat here is the unusually small sample of 705. Also, as noted below, ICM has been one of the more yes-friendly pollsters, such that the poll tracker adjusts it downwards by 2.0%.
An online panel poll by Survation, which has tended to come in at the middle of the range, has no at 47.0% and yes at 40.8%, for a rounded result of 54-46 to no. The poll was commissioned by the pro-union Better Together campaign. Whereas the bulk of the polling for the referendum has been online, this one was conducted by telephone, from Wednesday to Friday, with a sample of 1044. As was the case with TNS yesterday, this is a first phone poll from an outfit whose previous polling was conducted online.
Opinium is an established online pollster which has made its first entry on the referendum, this being a survey of 1055 respondents. Its result is a lot closer to Survation’s than ICM’s, with 45% for yes and 49%, rounding out to 53-47.
Panelbase in the Sunday Times has no on 50.6% and yes on 49.4%, but given its relative “yes” lean in the past, it’s comes out similarly to Survation and Opinium so far as the poll tracker is concerned.
Saturday, September 12
It’s now less than a week until Scotland’s independence referendum, which will be held on Thursday with polling stations to close at 10pm local time, or 7am Friday on the east coast of Australia. An official result won’t be expected until mid-afternoon our time. Before that time, the 32 local authorities that will be taking care of the business end of proceedings will report their results, which I guess we can expect to be done more promptly in the cities than the country.
The latest poll out this evening is an ICM poll for The Guardian which confirms the recent trend of being too close to call 42% no, 40% yes and 17% don’t know, panning out to a headline figure of 51-49 with the exclusion of the undecided. According to Anthony Wells at UK Polling Report, we can expect a result reasonably soon from Panelbase, which I have determined to be one of the more “yes”-friendly pollsters. We might then see a relative lull before heavy-hitters YouGov and Ipsos MORI hold off until their final results nearer the big day, although there will surely be other results around the place between now and then.
My own polling tracker, which is laid out below, currently has “no” in the lead with 51.6%, but there is no sign that the trend to “yes” is levelling off. As I shall discuss, it would have been more like 51.3% if I had treated the latest poll differently, as maybe I should have.
A few things that have caught my eye:
For those of you who know your way around Scotland, The Guardian offers mapped results of a year’s worth of Ipsos MORI polling in eye-watering detail.
John Curtice, a political scientist of some renown, considers the contention popular in the “yes” camp that pollsters are under-representing respondents who don’t normally vote, whom they expect will give their cause a boost. However, Curtice finds that past non-voters who have been polled are leaning quite strongly towards no.
Stephen Fisher at Elections Etc observes polling before 16 constitutional referenda in Austria, Ireland, New Zealand, Northern Ireland, Quebec, Scotland, Sweden, Britain and Wales (not us, alas), and notes there does not seem to be a precedent for a close referendum at which final polls underestimated the Yes vote (he in fact said overestimated but if you read the sentence in context, this was clearly erroneous).
Now to my own poll tracker. The methodology runs roughly as follows: 1) Calculate accuracy ratings for each pollster based on the performance of their polls in the last week of the campaigns for the 2014 European, 2011 Scottish and 2010 Westminster elections; 2) Run a local area regression on the results with each poll weighted as per the relevant pollsters’ accuracy rating multiplied by the sample size; 3) Use the trend result thus produced to derive bias measures for each pollster, by averaging the deviation of their poll results from the trend; 4) Correct the pollsters’ results accordingly and run the regression again.
The bias adjustments made to the various pollsters’ “yes” results are as follows: Ipsos Mori +1.6%; YouGov +1.6%; TNS +0.6%; Survation -0.7%; ICM -1.5%; Panelbase -2.5%; Angus Reid -4.4%. The complication I mentioned earlier is that the latest ICM poll was conducted by telephone, whereas the nine previous polls from which I have derived its bias adjustment were online polls. I have nonetheless decided to apply their existing adjustment to the latest result. Since this poll is, together with the most recent YouGov, the very latest result in the model, and the bias adjustment used is a not inconsiderable penalty to “yes”, the effect is non-trivial. If no bias adjustment is applied to this poll, the “no” result comes down to 51.2%. If the poll is removed altogether, it is 51.3%.
A couple of further points to be noted. YouGov’s stunning poll result on Monday showing yes in the lead was a real outlier from a normally no-leaning pollster, and it shows up in the charts as the only data point with yes above 50%. Another poll from YouGov a few days later had no back in front. This was inevitably reported in terms of the momentum for yes having stalled, but that’s not the picture that emerges when the polling is aggregated. Panelbase’s polling before the start of this year had an enormous lean in favour of yes which has since been corrected, so the earlier results have been excluded. Angus Reid is, or has been, a fairly major pollster in Britain, but there has been no Scottish independence polling from it since August last year. Should it re-emerge in the next few days, I will have to think twice about applying the 4.4% correction noted above.