Unpredictable prediction errors

Yes? by Cams, on Flickr.
Readers of this blog may remember that a while ago I made a prediction of the geographical distribution of a narrow Yes vote, based on the most recent council election and some reasonable assumptions about voter behaviour.

The assumption made was that the following percentage of party voters would vote Yes: SNP — 81.7%, Labour — 25.8%, Tory — 5.9%, LibDem — 26.2%, Others — 50.0%. (That is, I expected 81.7% of the people who voted SNP in the council elections to vote Yes to independence.)

A survey made by Lord Ashcroft (PDF) found that 86% of SNP voters, 37% of Labour voters, 5% of Tories and 39% of LibDem voters voted Yes to independence, but this was based on people’s recollection of their last Westminster vote, not the council elections. Also, this was based on a small sample, so these numbers may not be entirely accurate.

To test this, I wrote a computer program to work out the percentages that would have produced the best prediction of the actual result (still based on the council election results). The results are rather surprising: SNP — 64.6%, Labour — 50.3%, Tory — 9.1%, LibDem — 33.4%, Others — 36.9%. Using these percentages produces a decent prediction of the actual result (although a few council areas are wrong, such as Dundee, which performed much better than the revised prediction, and West Lothian, which performed worse).

I don’t claim that these revised percentages are accurate — you’d need a massive exit poll to make sure — but they show that many strong SNP areas performed much worse than I had expected, and many Labour areas performed much better.

To illustrate this, look at the differences between the old prediction and the actual result (the table has been sorted by the difference):

Council area Old prediction Actual result Difference
Orkney Islands 56% 33% -23%
Shetland Islands 56% 36% -20%
Moray 60% 42% -18%
Scottish Borders 47% 33% -14%
Aberdeenshire 53% 40% -13%
Angus 57% 44% -13%
Argyll and Bute 53% 41% -12%
East Lothian 49% 38% -11%
Na h-Eileanan an Iar 58% 47% -11%
Perth and Kinross 51% 40% -11%
West Lothian 55% 45% -10%
Clackmannanshire 55% 46% -9%
Midlothian 53% 44% -9%
Dumfries and Galloway 43% 34% -9%
East Ayrshire 55% 47% -8%
Aberdeen 49% 41% -8%
Falkirk 54% 47% -7%
Highland 54% 47% -7%
Stirling 47% 40% -7%
Fife 51% 45% -6%
South Lanarkshire 51% 45% -6%
Renfrewshire 52% 47% -5%
East Dunbartonshire 44% 39% -5%
North Ayrshire 53% 49% -4%
Edinburgh 43% 39% -4%
East Renfrewshire 40% 37% -3%
North Lanarkshire 53% 51% -2%
West Dunbartonshire 56% 54% -2%
South Ayrshire 43% 42% -1%
Dundee 56% 57% 1%
Inverclyde 49% 50% 1%
Glasgow 51% 53% 2%

Orkney and Shetland might be special cases, because they are so far away from Edinburgh, but what happened in places such as Moray, Aberdeenshire and Angus? Did the focus on winning over the Labour voters in Greater Glasgow make the rural SNP voters desert independence?

Going forward, we need to ensure that independence doesn’t become solely a left-wing ambition. Independence will be good for almost everybody in Scotland, and next time we need to work harder on making independence the choice of people everywhere, not just in and around Glasgow and Dundee.

4 thoughts on “Unpredictable prediction errors

  • 20/09/2014 at 20:27

    Interesting – it seems clear that the Yes campaign’s heavily left-wing redistributive vision of an independent Scotland alienated more affluent voters. Those voters also would have been much more likely to be worried by the “uncertainties” on what would happen to Scotland economically.

    • 20/09/2014 at 22:02

      Indeed. Some of these voters were probably never going to vote Yes anyway, simply because they had too much to lose, but surely many of them were potential Yes voters.

      I’d like to hear from Yes campaigners from these areas what their experience was on the ground. Did anything in particular frighten these voters? What do they think should have been done differently?


Leave a Reply

Your email address will not be published. Required fields are marked *

Optimized with PageSpeed Ninja