does cutbot cut it?

Tell us which Cutbot services you’d like to try out and we'll be back in touch to set up a free trial for you. We’ll also be happy to meet up or chat through how to get the most out of that trial account. Discounts are available for charities, small businesses, agencies, and the public sector.
Our online media monitoring scans over 1.7 million articles a week for you from around the world, from newspapers, magazines and broadcaster websites to blogs and organisations.
Our public affairs monitoring currently covers the Scottish Government and the Scottish Parliament, with Whitehall, Westminster and the other devolved bodies coming soon.







stay informed

cutbot blog

Our Holyrood forecaster: the small print

April 6th, 2016

This will, I’m afraid to say, be a long and boring post for most readers: it’s some of the background methodology to our seat forecaster – the main intro to it is here. For those with an anorak obsession with the Scottish Parliament’s electoral maths, though, it will probably be contentious (do let us know what you disagree with). Some parts of designing a seat forecaster are simple (given, say, the detail available on the last election), but others are not. In general, we have assumed that regional and constituency shares will change in a simple mathematical way. Not Uniform National Swing, but a method which assumes that parties’ votes change more when they won more votes in that area last time. So a national swing to a particular party on the constituencies will see its vote increase more, in absolute terms, in constituencies where they are already strong. Conversely, a 5% decline in a party’s regional vote will see 5% removed from their share in each region.

There are some wrinkles. To start with the Greens, they are standing in three constituencies, and it would be problematic to disregard those candidacies. But we have no figures for Green constituency votes from 2011. And we cannot ask visitors to the site to put in a national constituency vote share for the Greens when they’re only standing in those three seats. So what have we done? We’ve fudged it, using a few old data points. Here’s how it works.

The only time the the Greens stood for a Holyrood constituency in a general election was in 2007, conveniently enough in Glasgow Kelvin, which is largely unchanged from the boundaries of the seat of the same name which Green co-convenor Patrick Harvie will contest this time. And we have list votes available for each constituency for that election. So we can compare the Green list vote in Glasgow Kelvin with the Green constituency vote in the same area. For this purpose we have excluded all constituency and list votes for parties and candidates other than the five parties currently represented in Parliament – which may seem arbitrary but in this case is more conservative. Doing that, we see that Greens got 13.32% of the “big five” constituency vote in Glasgow Kelvin in 2007, and 12.43% of the list vote in the same area.

So, for the purposes of this exercise, we will assume that Greens do marginally better in constituency vote share than in list votes in the same constituency. Again, although this seems generous, remember that the Green constituency candidate in 2007 was Martin Bartos, at that point not elected to anything, although now the party’s councillor for Partick West. The three Green constituency candidates this time were all sitting MSPs at dissolution, so all three have a much higher profile going into the campaign: Patrick Harvie in Glasgow Kelvin, Alison Johnstone in Edinburgh Central, and John Wilson in Coatbridge and Chryston.

Anyway, back to the maths. What we’ve generated is a notional set of constituency results for these three seats: the numbers we estimate the “big five” parties would have got if the Greens had stood last time. And we’ve done that by simply taking the regional vote in those three constituencies and multiplying it by the ratio from Kelvin in 2007 (so estimating the constituency score would have been a little over 7% higher). That gives us a notional Green vote. We then looked at the 2007 Kelvin relationship between the other “big five” parties’ list and constituency shares, to estimate where those Green constituency votes would have been taken from. Almost half, according to this estimate, would have come from Labour, 37% from the SNP, 14% from the Lib Dems, and less than 1.5% from the Conservatives. So we take those fractions of the notional Green votes from the other parties’ actual constituency votes, and that gives us a full notional result for 2011.

It’s at least plausible, looking at those shares. 84.% of the Green votes coming from Labour and the SNP? Virtually none from the Tories? Let’s go with that.

So that’s level one handwaving, to get us Glasgow Kelvin figures. Edinburgh Central and Coatbridge and Chryston require level two handwaving. In those cases we’ve taken the regional list vote the Greens got in 2011, and we’ve multiplied it by the same ratio as the actual results in Kelvin showed in 2007. And then we’ve assumed that Green votes would have come from other parties in the same ratio as we estimate they did in Kelvin to generate notional figures there.

What this gives us (still excluding others) is a notional 2011 Green vote share for the Kelvin constituency of 18.35%, for Edinburgh Central of 17.23%, and for Coatbridge and Chryston of 2.13%. We will then use visitors’ predictions for the Green list vote share as a proxy for how well the Greens do in those three constituencies, starting from those notional baselines.

Obviously this really is wet finger in the air stuff. To pick one additional problem, it doesn’t take account of the changes in Scottish politics during and after the indyref, where the public’s party preferences now align much more closely with their independence views. However, we believe it would be more inaccurate to have nothing at all. There will be Green candidates, and the 2007 result shows they could at least potentially pull in an important number of votes, especially in Glasgow and Edinburgh, the party’s two strongest regions.

We can also calculate what national share of the vote the Greens would have to be at to win a constituency (depending on other parties’ scores). Although the notional score in Kelvin was slightly higher, Edinburgh Central was a tighter constituency last time. The SNP’s Marco Biagi won Edinburgh Central with just 32.7% of the vote, compared to the 43.3% with which Sandra White took Glasgow Kelvin. So at the point we predict a first Green constituency MSP, it would almost certainly be Alison Johnstone.

Another potentially complicated issue is that of the “Others”. Some of them were significant – most notably Margo MacDonald, who won in Lothian region, but who is sadly no longer with us. A “no other change” election would see the Lib Dems pick up her Lothian seat – they were the first to miss out last time. In both the Northern Isles constituencies, second place behind the Lib Dems was claimed by an independent candidate. For the purposes of this forecaster, though, we will simply ignore them. If either of the Northern Isles independents stands again, there’s no way to predict their scores based on national shares. In part that’s because it’s an aggregate, obviously. Even if “Others” got enough votes to win a list seat in a region, we will assume they’re fragmented enough for no individual independent to win. So the “others” scores predicted for each region and each constituency will be our estimated averages, simply shown so shares for the parties considered individually are more accurate (but not affecting actual seat wins in either section). A final thought for anyone who’s got this far: there are two notable other ways in which we would like to tweak the model. As Stuart MacLennan does, we would like to take account of regional changes, as far as that’s possible. Regional subsamples within Scottish polls are of course wildly inaccurate (the weighting and panel balance is done nationally), but by grouping enough polls together it’s possible to see some regional variations in swing. It would also be interesting to take account of the move to the “Yes Parties” (i.e. largely the SNP, to a smaller extent the Greens) in more “Yesish” areas of Scotland, without changing the overall national numbers, and conversely to boost the “No Parties” (largely Labour and the Conservatives, to a lesser extent the Lib Dems) in areas where independence was less popular in September 2014. Doing both would risk double-counting, of course. And the numbers for the indyref are provided only by local authority region (except in Glasgow, where the Holyrood constituency breakdowns were calculated – and in Edinburgh, where the Westminster constituency figures were done – slow handclap to the rest).

But those are ideas for another version. Feedback welcome!

2 Responses to “Our Holyrood forecaster: the small print”

  1. Hi Folks,
    Nice to see someone looking a little more in-depth at Scottish forecasting.
    Within the boundaries of my abilities, I’ve been looking at this too.

    I tried several models for looking at regional variations and came to the conclusion that, in the absence of any dedicated regional polling, the cross-breaks were just too small to give anything remotely accurate (in as much as polling can be). It also proved to be beyond my ability to model the effects of the indyref.

    I did a little bit of manual tweaking to try to account for others, but even here it was (as you say) very much a case of sticking a wet finger in the air.

    The google doc link above shows the results of my limited efforts so far.

    Please keep blogging on any refinement you make – I’m very interested to see your approach.

    Kind regards
    Tom Hefferin
    https://docs.google.com/spreadsheets/d/1UbHC-4IOm0RQRV12NJIs8vZQ3t4ZJncd_I2fxFTLQG8/edit#gid=1152402072

    • Hi Tom, thanks so much for this, I’m taking a look at your spreadsheet (and will have a longer look on Monday). It might not be the bit you’re most proud of, but the smoothed rolling average polls in sheets 3 and 5 are absolutely fantastic! We’ve also given our methodology to one of the pollsters for comment, so anything useful that comes back from that will be fed in. I agree it would be very useful to have proper local data for the indyref, although again it’d be moderately wet finger to work out how that would skew the projection for any given region or constituency.

Leave a Reply

Allowed tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>


4 × five =