A Happy New Year?

Recently, there have been opinion polls which have put the SNP so far ahead of Labour in Westminster voting intention that, when the figures are fed into a site such as Electoral Calculus (http://electoralcalculus.co.uk/userpoll_scot.html), the predicted outcome is that the SNP might win more than 50 seats. At the same time, some people are suggesting that 20 to 30 seats would be a good result for the SNP; are they being too pessimistic?

There are three aspects to this question. The first is the degree to which the opinion polls are an accurate assessment of current voting intentions. Of course, there is a certain amount of random variation from one poll to another, which is to be expected because of the limited number of respondents to each poll, but a systematic error could arise if the adjustments which are generally made on the basis of how the respondents have previously voted are flawed. My impression, based on what I have read, is that perhaps the SNP vote share is being underestimated, but any such systematic error is probably rather small. The second, more important issue is how voting intentions will change between now and the election. All sorts of things might perhaps happen before May which could influence people one way or another, and I have no intention of making any guess as to how support for the SNP, or any of the other parties, will change. What I want to discuss here is the third aspect, namely how percentage shares of the votes might translate into seats.

The most recent figures from the Scot Goes Pop poll of polls on 24 December (http://scotgoespop.blogspot.co.uk/2014/12/christmas-poll-of-polls-shows-snp.html) are

SNP         44.1%    (19.9%)

Labour     26.5%    (42.0%)

Tories      16.5%    (16.7%)

LibDems   5.8%     (18.9%)

where the figures in brackets are the 2010 results. For the purpose of this exercise, I am going to assume that these figures are an accurate prediction of how the Scottish electorate will vote in May.

The simplest technique, Uniform National Swing (UNS), is to take the overall swings for each party in turn and apply these to the 2010 results for each constituency. Thus the SNP vote is increased by 24.2% of the total number of votes cast in 2010, the Labour vote is decreased by 15.5% and so on. I have constructed a spreadsheet on this simplistic basis, and it gives the following prediction –

SNP 46 (46)    Labour 9 (11)   Tories 2 (1)     LibDems 2 (1)

where the values in brackets are from the Electoral Calculus website, using the same poll of poll figures.

There is an obvious problem with this method; my spreadsheet predicts negative numbers of votes in four constituencies for Labour, and in many more for the LibDems. UNS has been found to work quite well for modest swings, but it seems to break down when there are both large swings and large differences from one constituency to another, as is the case at present in Scotland. Its weakness is the assumption that the change in the number of votes for a party in a particular constituency is independent of the previous level of support for the party in that constituency. As support for Labour drops, they stand to lose the greatest number of votes in those constituencies where there were the most Labour voters to start with. Similarly, as support for the SNP increases, it is the constituencies which were already SNP strongholds which should see the smallest numbers of new SNP voters, simply because there were fewer non-SNP voters available to be converted.

A more plausible approach than UNS is to say that the fraction of 2010 Labour voters who will vote Labour again next May is 26.5/42.0, and to multiply the votes  for Labour in each constituency by this fraction. For Glasgow North East, this gives 43.1%, down from 68.3%, while in Berwickshire, Roxburgh and Selkirk, Labour’s vote share would drop from 10.2% to 6.4%. The same method can be used to predict vote shares for the Tories and the LibDems. This method cannot predict negative or implausibly low vote shares.

If we apply this method to the SNP in Na h-Eileanan an Iar, we must multiply 45.7% by 44.1/19.9, which gives 101.3%, and so it appears not to work so well in the case of a party which is showing increased support. Instead, it is necessary to consider changes to the total vote for all parties except the SNP; in the case of Na h-Eileanan an Iar this is 54.3% which has to be multiplied by 55.9/80.1, giving 37.9% and hence an SNP vote share of 62.1%. The justification for this is that when we are looking at the change in support for a party, we have to consider where that change is happening. For a party which is losing support, it is happening amongst people who previously voted for that party; for a party gaining support, the change is happening amongst people who previously did not vote for that party.

A spreadsheet on this basis [see Note] predicts an overall result of SNP 57 and Labour 2, and one of the Labour seats would be very marginal. Even Glasgow North East, Labour’s safest seat, would fall to the SNP with only a further 1.2% swing from Labour to SNP. In other words, it suggests that it might be possible for the SNP to win every seat in Scotland. Surely this must be too good to be true? Probably it is – a bit – because there are factors which a simple spreadsheet cannot take into account.

Firstly, although probably most people vote for a party, there will be some who vote, at least in part, for a particular candidate whom they respect. As support for a party drops, there may be some Labour and LibDem MPs whose personal following will allow them to buck the trend. The other side of this coin is that the SNP will have to field a number of relatively inexperienced and perhaps not very well known candidates for winnable seats.

Secondly, the constituencies in which the SNP did best in 2010 are presumably those in which the SNP will have the most members available for activities such as canvassing, distributing leaflets and so on. This could partly offset the tendency, assumed in the spreadsheet, for the SNP to show the greatest gains in support in the constituencies where their share of the vote in 2010 was lowest.

Thirdly, the spreadsheet does not consider possible changes in who actually votes. The referendum encouraged a lot of people who were missing from the electoral roll to register. Is this change being reflected correctly in the opinion polls? As the fortunes of Scottish Labour slump, Labour voters may become demoralised and not bother voting, while as support for the SNP soars some independence supporters may think the SNP no longer needs their votes and shift their votes to the Scottish Greens or the SSP.

Finally, there will also be a certain amount of almost random variation from one constituency to another, due to demographic changes, local politics influencing views on the various parties, the effect of minor parties which do not contest every seat, and so on. In any case, there may be some constituencies which are just not fertile ground for the SNP – such as those next to the border with England, and Orkney and Shetland.

In conclusion, I would argue that if (and it is a big ‘if’) the SNP’s lead over Labour in the election is similar to that given by recent opinion polls, then most of those intimidating Labour majorities will melt away, the first past the post system will favour the SNP for a change, and the SNP should win somewhere around 50 seats. Then Jim ‘Labour won’t lose any seats to the SNP’ Murphy will have a lot of egg all over his face – metaphorically speaking, of course. And that could help make 2015 a happier year.

Note This appears to be essentially the Transition Model of Electoral Calculus, somewhat simplified because in Scotland there is only one major party which has gained support since 2010; however, Electoral Calculus currently use a modified version of this, called the Strong Transition Model (STM), because they assume that a party which is losing support will lose fewer seats than the Transition Model predicts. The STM involves a an arbitrary division of voters into ‘strong’ and ‘weak’, and it is difficult to know whether this is justified. My best guess (and it is just a guess) is that the STM will slightly underestimate the number of SNP seats, while the Transition Model will overestimate it.