Tuesday, August 13, 2013

Why Nate Silver is wrong

It's a brave man who uses a headline like that given the accuracy of Silver's predictions in last year's US presidential contest, but nevertheless I am going to confidently take issue with his claims reported in the Scotsman that the Yes side have "almost no chance" in the independence referendum. The main reason is that there are ample signs that his opinion is based on a fairly cursory look at the polling data - by contrast he would never put his neck on the line with a US political prediction without being totally immersed in every available statistic and variable.

1) He claims that the polling evidence is "pretty definitive" in putting the Yes side at 40% (give him his due here, he's at least not making the all-too-common schoolboy error of assuming that Don't Knows can be lumped in with Nos). But this ignores the Panelbase polls, which inconveniently and consistently diverge from the "definitive" pattern. Is Silver even aware of those polls? We don't know, but we do know that Panelbase is a credible polling company that adheres to British Polling Council rules, and we also know that no less a figure than Professor John Curtice has cautioned that for as long as one polling company is diverging from the others, it would be wrong not to at least consider the possibility that they are getting it right and that the others are getting it wrong.

2) He claims that the "No side is even more dominant with younger voters, so there's not going to be any generational thing going on". That's an utterly extraordinary claim given the number of polls - and not just Panelbase polls this time - that have suggested the complete opposite is true. For example, an Ipsos-Mori poll earlier this year showed that no fewer than 58% of 18-24 year olds were planning to vote Yes. Is Silver even aware of polls like that? Or did he just look at the MRUK youth poll that was so widely reported by our impeccably neutral media, and assume that must represent the "definitive" picture?

As Marcia has pointed out in the comments section at Wings, a further problem with Silver's remarks is one that is not specific to Scotland - namely the idea that No campaigns usually gain ground rather than lose it in referendums. (Again, to be fair, he at least doesn't make a prize idiot of himself in the way that Peter Kellner did by pretending that this represents some kind of unbreakable "iron law".) Matt Qvortrup made a similar grandiose claim recently, before backtracking somewhat on Twitter and conceding that the Yes campaign in the Montenegrin independence referendum did indeed gain ground as the campaign progressed - which they needed to do, because they looked likely to fall short of the artificial 55% threshold. But as Marcia has noted, a much better example is the monumental gains that the Yes campaign made over the course of the 1995 Quebec independence referendum campaign. Qvortrup apparently thinks that he can completely dismiss that example because Yes ultimately still lost by a whisker, but that's incredibly woolly thinking from an academic. The salient point is that huge numbers of Quebec voters who initially told pollsters they were planning to vote against independence ultimately walked into a polling station and voted in favour of it - and moreover they did so knowing that there was a severe "risk" that their votes could swing the balance. It was no protest vote.

That was something that simply shouldn't have happened if you believe Silver, Qvortrup and Kellner - so why did it? My guess is that, paradoxically, the more important a referendum is, the less likely voters are to swing to No by default. The supposed tendency that Silver talks about is largely a side-effect of electorates so often being faced with relatively trivial matters in plebiscites. Take the AV referendum, for example - the prevailing attitude among the public seemed to be "I don't give a monkey's about this, it's irritating to even have to think about it, so unless someone can give me a very good reason I'll just vote to keep what we already have". That kind of lazy thinking clearly went out of the window for Quebec voters when they were faced with the most important choice of their lives, and I'm confident it will go out of the window for Scottish voters next year. We will be dealing with an electorate that is engaged like never before - and as recent research has shown, the better informed that voters consider themselves to be, the more likely they are to vote Yes.


  1. James, The latest Panelbase poll (and if memory serves me correctly, the one before that) have men supporting Indy both in those likely to vote and in overall. No is relying on women for its majority. This is dangerous for them (and an oppertunity for the Yes campaign) as it most likely reflects woman's disengagement with formal political campaigns.

  2. Mentioning Quebec referendums is OK. After all, Yes lost twice in Quebec.

    Looking at the detail of those campaigns is just not done. As The Times remarks today in its coverage of this non-story, the sizable No vote in 1980 didn't just result in Quebec getting nothing, it actually had powers taken away. (Take a bow Michael Kelly.) The Scotsman's not going to be printing that part of the story, is it? That's even even scarier than the major swing from No to Yes in 1994-1995.

  3. Spoke to Nate Silver today. He knows he has been spun and generally prefers to keep a lower profile. His Referendum analysis is based on averaging the polls. He says the lowest No margin he has seen is 7% in favour of No. He was unaware of the WoS weekend poll. He has visited Scotland before, socially, and now gets most of his info from the Guardian. He was complimentary about blogs from those disputing his reported comments.

  4. I don't there's anything wrong with using poll averages as such. It would obviously produce inaccurate figures some of the time (for instance the 1992 election, when every pollster had its methodology wrong) but there's a case to be made for saying that a polling average is less likely to be inaccurate than any other method you can think of. The problem is that there's a bit of leap between saying that "No appear to currently have a decent lead" and saying that "the fact that No are ahead means that they have effectively already won, even though the referendum will not take place until a year from now", which is a much more extraordinary claim. So the key to this is the almost religious belief that the No side can always expect to gain ground in any referendum. For that belief to have credibility, there needs to be some kind of explanation for why Quebec 1995 (and indeed to a lesser extent Montenegro 2006) drove a coach and horses through that "rule". What was so unique about that campaign? Silver is actually quoted in the Scotsman as mentioning the Quebec example, but he bizarrely attempts to use it as proof of his own claim - ie. "Yes were in a winning position and then lost". But that happened in the last week! Given that we're a year out from the Scottish referendum, surely the much more salient point is that Yes made up huge ground in Quebec in the months leading up to polling. How does Silver explain that away? We don't seem to have an answer.

    By the way, I see Alex Massie has linked to this post with the words "Silver...appears to have annoyed some Scottish nationalists today". Honestly, Alex, if Silver had annoyed me you'd know all about it (I refer you to numerous posts on this blog about Messrs Smithson and Baker). I do think he's wrong, but that's a different matter. Alex himself is making me somewhat dizzy with his musings on this subject, because it hasn't been that long since he was giving some credence to the idea that the campaign proper hadn't really started and the polls therefore didn't matter, which seems to flatly contradict today's verdict of "We are where we are and, right now, a little over a year from the referendum the Yes side are losing." His comment about the Wings poll is also pointlessly snide - I haven't seen anyone claim that the finding about whether Scots would want to join the Union now if we were independent means that the Yes side is "winning". But I have seen plenty of people claim that it tells us something interesting about underlying public attitudes, which it self-evidently does. And underlying attitudes matter. Did the fact that Alex Salmond led Iain Gray on personal ratings mean that the SNP were "winning" the Holyrood election back in the early weeks of 2011? Nope. But it did mean there was a plausible reason for thinking that the raw voting intention figures were not telling the whole story, and that Labour's lead was potentially soft. We all know what happened next.

  5. Excellent post, James.

    Given John's comments, I'm amazed that the Scotsman front paged this story.

    I'm not much of a polling buff and therefore I'm not sure how reliable averaging the polls is.

    But surely for the Scotsman to spin this as an expert view (when all the guy did was quote a figure that even I could have quoted, and draw dubious conclusions from past referenda) was pretty poor journalism.

    Not, of course that I'm surprised to see poor journalism from the Scotsman.

    It seems they ignored the poll carried out on behalf of WoS, because it said things they didn't want to hear, and ran with this instead, because no matter how unreliable and unresearched it was, it was music to their Britnat ears.

    Of course, I have no problem with them spinning that we have no chance. It increases the chance of non-fanatic unionists, confident of an overwhelming victory, deciding that THEIR vote isn't necessary.

  6. Silver predicted major gains for the Lib Dems in the last British general election, and how did that turn out?

  7. Good points in the article and good to hear Nate Silver was wary about how his comments were spun. Those comments also betray a lack of understanding of the context and what might swing votes as I wrote here: http://www.scottishrepublic.eu/scottish/nate-silver-should-probably-stick-to-baseball/

  8. Hi Andrew, I'm pretty sure the comments about the doubling of the oil price and the viability of wave power were made by John Curtice. The example of a potential game-changer that Nate Silver put forward was the Eurozone breaking apart, which I agree is an extremely odd suggestion, because the conventional wisdom is that an economic crisis would be a boon for the No campaign.

    I think Steve Richards was bang on the money last night when he suggested that the real game-changer would be if it looked like the Tories were heading for victory at the next general election (in my opinion that already does look highly likely, but arguably the public won't be convinced until and unless the Tories take a clear poll lead).

  9. john, etc,

    It appears Nate Silver was spun.

    However he has nowt to say about it on his Twitter feed as far as I can tell.