How should we interpret the early results in the Scottish referendum? Part II – revised method

My previous post examined what a Scotland-wide 50-50 tie would look like in each area if the regional dispersion were the same as in the 1997 devolution referendum. Thanks to everyone that posted feedback on this site and on the various fora.
The point that came up repeatedly was “Area X is a strong/weak SNP area, so the yes vote should be higher/lower there” (and sometimes other parties were mentioned too). Also, some areas (solid Lib Dem territory was one example given) might have been very pro-devolution but very “No” on the matter of independence. These are, of course, very valid concerns. So with that in mind, I’ve come up with the following alternative approach:

  • Take the 2010 Westminster ward-level vote estimates (as per Electoral Calculus)
  • Aggregate them, not to constituencies, but to local authority totals
  • Take the referendum VI by 2010 vote from opinion polling. To weight the 2010 non-voters, I had to make an assumption about turnout, so I used 85% with an evenly-spread increase over 2010, but using 75% or 95% doesn’t make much difference
  • Combine the two to get an implied referendum vote for each local authority
  • Normalise the overal result to a national tie

This solves the main problem with the original model, but it does create some new ones:

  • Except where local authorities corespond exactly to Westminster constituencies, ward level (and hence council area) figures for 2010 are estimates. The Electoral Calculus methodology is sound (it’s the same methodology you would use to calculate notional results after boundary changes) but they are still estimates
  • Using polling data rather than purely real votes means sampling error. Because only ICM seems to have published a breakdown by 2010 vote the samples are small and hence the MoEs are rather wide (though this is arguably still preferable to using the results of another referendum 17 years ago and on an entirely different question)
  • It also means we have to trust opinion polling, although a biased sample would also have to be biased geographically to matter in this case
  • Voters for each party aren’t necessarily homogenous, and this might have a regional element to it. New approach, similar problem…
  • The turnout concern still applies

So with all of those caveats noted, here is the revised version:

We can see that Aberdeenshire, Angus, Perth and Moray are all much higher than the original model had them, as are the islands. Glasgow is quite a bit lower than before.
In relation to feedback I’ve received, Dundee is also higher but not by all that much. The big 3 cities collectively are lower (though not all individually, and not by a huge amount).
Incidentally, I was concerned that the difference between referendum and Westminster turnout patterns (ie safe seats vs marginals) could affect the regional pattern by ‘misweighting’ the national average. So I tried reweighting the average using the 1997 turnouts, and it made almost no difference whatsoever.
I’ll update this on Wednesday or Thursday once the final call polls are in, hopefully some more of them will release a breakdown by 2010 vote. Ideally I’d be able to do this by 2011 Holyrood vote, but  as far as I know, the local breakdowns aren’t readily available for Holyrood Elections.
I’ll probably post it on the forums but you can get it first on Twitter:

About The Author

Related Posts