Post Election Round-up; How Accurate Was My Calculator?

As outlined at the very inception of this blog, it’s no easy task predicting local and regional election results on the basis of national polls. The traditional method, Universal Swing, simply assumes that the change in votes will be the same everywhere – e.g., if a party is down 3% nationally, it’ll be down 3% in every constituency and region. In reality, we know that won’t be the case. In trying to account for that with my model, I used regional swings based on approximating 2015 Westminster results to Holyrood regions and seeing how the result in each region varied from the national result.

Now, as established yesterday, my actual final prediction was off significantly. But so too were the polls that informed that prediction (more on that in the next post). To see how how accurate my model was in and of itself, I’ve plugged the actual national vote shares into it.

Model vs Actual

Looking just at the headline figures, the calculator is spot on for three of the five parties in parliament – Labour, Greens and the Lib Dems. It over-estimates the SNP by 4 seats and therefore under-estimates the Tories by 4. It’s not a million miles off, then, giving me a national success rate of 121 of 129 MSPs. But what about individual regional results?

Central Real

In Central, it gets all the constituencies right, but my model had Central as the Tories second worst area and gave them only two regional seats. In reality, they won three – with Labour suffering an even larger decline in Central than expected. From a Green point of view, the model was also just a shade too pessimistic about Kirsten Robb’s chances, albeit she still unfortunately missed out.

MSPs Correct; 15 of 16.

Glasgow Real

Based on most polling, and with the model viewing Glasgow as the Tories worst region, I would have expected the Greens to beat the Tories here. However, with the Tories doing so well nationally, even the model had them squeaking ahead – but Labour were seen to still be strong enough to prevent either party from getting two MSPs. Labour’s Glasgow collapse was even worse than expected however and sadly for Zara Kitson the Greens didn’t do quite so well, meaning the Tories got that second seat.

MSPs Correct; 15 of 16

Highlands Real

A rare success for Labour in the Highlands and Islands region, as they just managed to hold on to both of their list MSPs where my model was inclined to suggest they’d drop below 10% of the vote. Likewise, the model significantly overestimated the potential for Tory success in the region, with the Lib Dems bumping up a fair bit and the Greens also performing significantly better.

MSPs Correct; 14 of 15

Lothian Real

Lothian is always one of Scotland’s most interesting regions. My model would have expected the Tories to reclaim the Edinburgh Pentlands seat they held from 2003-2011 in addition to winning Edinburgh Central. In actual fact the SNP held it rather comfortably, and wins for Labour in Edinburgh Southern and the Lib Dems in Edinburgh Western relieved enough pressure on the lists for Andy Wightman to win a second Green seat.

MSPs Correct; 11 of 16

MF Real

A shock win for Willie Rennie in North East Fife as well as a solid increase in vote share gives the Lib Dems a far better result in Mid Scotland & Fife than my model would anticipate. Meanwhile, slightly worse results in terms of vote share for the Tories and Greens, who nonetheless won the number of seats the calculator expected.

MSPs Correct; 15 of 16

NE Real

The big inaccuracy in the model for North East was between the SNP and the Tories – the SNP suffered a surprisingly large dip in support in the region, whilst the Tories pulled off such a surge that they are two seats better off than modelled, including snatching one of the constituencies. A big flaw in the model here for the Greens, with it expecting Co-convenor Maggie Chapman to have been elected.

MSPs Correct; 14 of 17

South Real

A really mixed bag in South, with the model accurately identifying Tory constituency wins but obviously completely failing to predict Iain Gray would increase his majority in East Lothian. It also over-estimated Green support a little bit, with the absolutely tremendous Sarah Beattie-Smith missing out in reality, and could not foresee a surprise collapse in the Lib Dem vote leaving them without a regional MSP where once they had a stronghold in the Borders. It was the Tories that benefited most from this inaccuracy, with two regional seats, and failing to win East Lothian gave the SNP’s Paul Wheelhouse a lucky return to Holyrood.

MSPs Correct; 8 of 16

West Real

West was always the region my model was most suspect for – I kind of knew that, even if the SNP stayed static nationally, it was likely to be by losing votes in some regions and gaining in places like West and Glasgow, whilst that enormous Tory vote share was just too high. Still, consistent predictions of a Tory win in Eastwood proved quite correct, although even hailing from Dumbarton constituency originally, neither me nor the model could see Jackie Baillie clinging on. The biggest success though was for the Greens, the result on the day being far higher than in my model, successfully electing Ross Greer as the youngest MSP ever.

MSPs Correct; 14 of 17

On the whole then, the accuracy of my model seems a little bit mixed. Summed up across constituency and regional seats, my model correctly identified 106 of the 129 MSPs. I’m not counting constituency winners I’d thought would get in on the list or vice versa as correct, as even though I identified the MSP correctly I got their method of election wrong.

How does that square up with Universal Swing? I’ll fire through this with less comment, as it’s the numbers that matter.

National Universal Swing

National MSPs Correct; 121 of 129.
National Parties Correct; 1 of 5 (parliamentary).

Central US

MSPs Correct; 14 of 16

Glasgow US

MSPs Correct; 15 of 16

Highlands US

MSPs Correct; 14 of 15

Lothian US

MSPs Correct; 11 of 16

Mid US

MSPs Correct; 15 of 16


MSPs Correct; 14 of 17

South US

MSPs Correct; 9 of 16

West US

MSPs Correct; 15 of 17

Total; 107 of 129 MSPs.

I hereby concede narrow defeat to tradition when it comes to predicting individual MSPs – but claim a narrow victory in predicting national results.

And finally… What about the Scotland Votes calculator, from which we can only get the national overview for regions?

Scotland Votes.PNG

It doesn’t accurately predict any of the national seat shares, coming in at a total of 12 seats at odds with the actual results, for 117 of the 129 national MSPs. Now, that, I’m claiming as a significant victory!


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s