Tuesday, 6 January 2015

Bedtime stories: extending my thoughts about Scottish constituencies

Always beware of a bright idea.  As I explained in my last post, I've been thinking about how the referendum may play out in individual constituencies.  And once I start worrying at a problem, I can't give up.

To set the scene, let me recap with my last post:

I've been trying to put together a methodology for working out how the rise of the SNP could work through to the May 2015 general election results.  We know that the SNP have risen sharply in the polls.  However, uniform national swing is always dangerous to use when a rise is as sharp as the one that the SNP seem to have got (it is inherently unlikely that a large increase will be uniform) and on this occasion we can be fairly confident that the increase is not uniform, because it seems to have been driven by the referendum on independence.  Since Yes did much better in some areas than others, we can expect the SNP to be increasing support disproportionately in the areas where Yes did best.

In my last post, I looked just at Glasgow.  I've now tried to produce a means of guessing how each individual constituency might pan out.

Feeling the aura of referendum voters

As I explained last time, my starting point is to look at how we got from the 2010 results to the referendum result.  I decided to proceed as follows with the following very rough and ready assumptions.  

The first critical point was to understand how much of the Yes vote came from the leading unionist party.

I would allocate all the Conservative votes in 2010 to No and all the SNP votes in 2010 to Yes.  Hopefully the exceptions would cancel each other out.

Next, in seats where Labour were ahead of the Lib Dems, I would allocate the 2010 Lib Dem votes 60:40 to No.  Why 60:40?  It's fairly arbitrary - I might have picked 2:1.  Or I might have picked 50:50.  But 60:40 felt right somehow.  (Where the Lib Dems are ahead of Labour, I decided that getting the Lib Dem share right was more critical, so I have assumed that the Labour share was 60:40 to No and sought to work out the Lib Dem proportions of Yes: No.)

As for all the other parties - other than Labour - I decided to treat them as dividing 50:50 between Yes and No.  This is loose - 2010 UKIP supporters were probably disproportionately likely to be No and 2010 Green supporters were probably disproportionately likely to be Yes.  But life is too short.

We next need to recognise that many referendum voters did not vote in 2010.  I decided to allocate these 60:40 to Yes where Yes got over 45%, 50:50 where Yes got between 35 and 45% and 40:60 where Yes got below 35%. The cut-offs and the percentages are arbitrary.  I just chose what seemed like reasonable assumptions that I could work with fairly easily.

That leaves the 2010 Labour voters (or in those few seats where Labour are behind the Lib Dems, the 2010 Lib Dem voters). I could then treat them as making up the balance on each side, knowing the referendum result in each council. 

Applying this "knowledge" to look at voting in individual constituencies

At this point, I now "knew" how the referendum result had been compiled.  What could I do with this "knowledge"?  Time for some more assumptions.

My starting point is that Yes voters who had voted in the 2010 election are very likely now to vote SNP.  I decided that 80% of such voters (whatever party I had deemed them to come from) would vote SNP next year.  That is me putting the zeal of the convert into figures.

Then there is the question of new voters in the referendum.  For some, this will be a habit to be repeated.  Given the superior ground game that the SNP will have (all those extra foot soldiers must be worth something), I have taken the view that previous non-voters who do vote will be disproportionately likely to vote SNP.

I decided that only half of new voters would vote in May, but of those half, they would vote as follows:

Where Yes is above 45%,

In a Labour-held seat:

Labour 30% 
SNP 60%
Others 10%

In a Lib Dem-held seat 

Lib Dem 20% 
Labour 10% 
SNP 60% 
Others 10% 

Where Yes is between 35% and 45%,

In a Labour-held seat 

Labour 40% 
SNP 50% 
Others 10%

In a Lib Dem-held seat 

Lib Dem 25% 
Second unionist party 15%
SNP 50% 
Others 10%

Where Yes is below 35%,

In a Labour-held seat 

Labour 50%
SNP 40%
Others 10%

In a Lib Dem-held seat 

Lib Dem 30%
Second unionist party 20%
SNP 40%
Others 10%

In the one Conservative seat, allocate new voters as follows:

Conservatives 20%
Labour 20% 
SNP 50% 
Others 10%

You will note that I haven't looked at the SNP seats.  With all six at 1/100 shots, I have better things to do with my time.  Sorry about that.
Then there's the question what to do about the Lib Dem collapse in the polls.  I hummed and hawed about this.  In the end, I decided that in seats which the Lib Dems did not hold, they would lose three quarters of their 2010 vote, with Labour, Others and abstentions evenly sharing those votes that were not accounted for by the SNP and the Lib Dems.
In Lib Dem held seats, I needed to something different to reflect the incumbency and the effort that the Lib Dems would put in.  So I replaced the usual assumption that the Lib Dems would lose 75% of their 2010 vote with an assumption that 20% of post-SNP defectors would transfer to Labour, while the remainder remain loyal to the Lib Dems.  In seats where the incumbent is not standing, I upped that percentage from 20% to 35%.

And that's all there is to it.  Except to do the sums.

That's easier said than done.  Some councils exactly match the boundaries of single constituencies.  Some councils have several Westminster constituencies.  Some constituencies straddle two or more councils.  Some councils helpfully gave a lot of detail on their referendum results, while others have been much terser.

How did I deal with this?  Well, when I came across a problem, I just took a view and either approximated using the council that seemed most appropriate or used something that looked close to an average between the relevant councils.  Or I did the calculation on two bases (fortunately each time I did this the result was in the same ballpark).  I went below council level where I could match up referendum results with the constituency, as in Edinburgh, but I didn't when the council reported on a different basis, as in Glasgow.  I didn't have time.  Go ahead if you feel like it.

Why have I spent so much time explaining this?
Because I don't want you treating this with any kind of reverence or thinking that I am impressed with my own efforts.  The intention is to produce an extremely rough and ready artist's impression of what Scotland's Westminster seats might look like after May 2015.  It is in no way a prediction (unless it turns out to be right, in which case I shall waste no opportunity to trumpet it).  It's a sense check.

You will no doubt disagree with many or even most of my assumptions.  Go ahead.  The idea is to set down an idea to kick around.  I have no illusions that this is some kind of work of genius.

I'm refusing to give precise figures.  I don't want anyone to get any spurious idea of precision about this.

Opening the envelope

So, here goes.  The headline figures are:

Labour: 15
SNP: 40
Conservatives: 3
Lib Dems: 1

And here are the constituencies in detail:

In all constituencies that the SNP do not take, they are second unless explicitly stated.

For me, though, what's more interesting is to see which seats are most vulnerable to the SNP on this type of model and which are more resistant to their charms.  The SNP would tear through the west coast of Scotland.  But Linlithgow & Falkirk East (currently priced at 5/6 for the SNP) would stay in the red column.

I was kind of expecting the SNP to do well on my model in the northern Lib Dem seats, but I was shocked at just how well they did.  (What's really driving this in my model is the relatively high vote for Yes in Highland (47%) and the high turnout in that council in the referendum relative to the 2010 general election.)  

My model makes no allowance for tactical voting.  Seats like Edinburgh West, Edinburgh North & Leith and Aberdeenshire West & Kincardineshire might very well turn out differently in practice once the campaigners get to work on the electorate.  Aside from the Lib Dems, it makes no allowance for incumbency either.  

Using this table for betting
Please don't.  At least, please don't take it literally.  It's intended to give you something to think about.  It's not a predictor (yes, I know I've said this once, but I want to say it again in case you missed it the first time around).

What it may help with is in spotting hidden opportunities.  Cumbernauld, Kilsyth and Kirkillintoch East was at 2/1 the other day, but since the SNP would take this with some ease on my model, I've had a bet on that.  Note what good value the Glasgow seats look on this model.

This is a model not a prediction
Yes, that's a third time.  But this time I have a slightly different point to make.  Things could change dramatically before May even if this model fitted the current direction of travel quite well.  Nothing is preordained.  Jim Murphy may yet turn things around, though that looks odds against to me.

Incidentally, you can get 5/6 on the SNP taking more than 26 seats with Ladbrokes.  If you have faith in all this.  But as I said at the outset, always beware of a bright idea.

No comments: