A bit more on Uniform National Swing – and an alternative

Blog readers will be pleased that the next Welsh Political Barometer poll is due soon. The results of our regular poll, conducted in collaboration with ITV-Wales and YouGov, will be available in early December. What better Christmas present could anyone hope for?

Undoubtedly attracting attention once again will be the seat projections, for Westminster and the National Assembly, that are generated from the raw polling numbers. To remind you all of what I have said previously:

  • The seat numbers generated are projections based on current polling numbers. They are not predictions of what will happen in May 2015 or May 2016;
  • The numbers produced are generated via a uniform national swing (UNS) assumption. No attempt is made to account for local factors that might produce deviations from the national norm within particular constituencies.

I’ve discussed UNS in more detail before. Suffice to say that I regard it as a useful rough-and-ready tool for projecting the broad implications of national polls. It is quick and easy to apply, and can be understood very simply. But it is not flawless, and can be downright problematic for parties seeing large percentage changes in their vote share.

We can illustrate some of these problems with reference to the last Barometer poll, which put the Liberal Democrats on 6%. This means they are down just over 14 percentage points on their support level at the 2010 general election. UNS suggests that we should therefore apply a 14% downwards adjustment in the Lib-Dem 2010 vote share to ever seat. But applying such swings uniformly for those seats where the Lib-Dems got less than 14% of the vote in 2010 generates a mathematically impossible projection! A corollary of this is that UNS must be under-stating the average swing against the Lib-Dems in other seats: if they are down 14% nationally, and it is mathematically impossible for them to be down as much as 14% in some places, then in the rest of Wales they must be down by an average of more than 14%.

Are there alternatives? One recently suggested to me by a helpful Blog reader is Ratio swing. Whereas UNS models changes as the absolute percentage shift in a party’s vote share, Ratio swing models it – as the name implies – as a ratio of the change in support.

To illustrate (and using an example that keeps things as mathematically simple as possible): imagine a party that scored 20% at the last election, but is currently polling at 10%. UNS would project the party to lose 10 percentage points in every seat from the previous election. Ratio swing would project the party to get half the vote share in every seat that it gained in the previous election.

Ratio swing can also generate obviously problematic predictions. Imagine a hitherto very small, but rapidly growing party that won only 0.5 of the national vote at the last election but was now on 5% nationally. Ratio swing would suggest projecting that party to get ten times the vote share in every seat. But what if they had won 11% last time in one seat…

One way to test these two approaches is to apply them to actual election results. I’ve worked through the 2010 general election and 2011 National Assembly (constituency) results. For both I looked at the change in national vote share from the previous election, and then examined whether either uniform or ratio swing would have predicted the ‘correct’ result: i.e. predicting the party that actually won the seat. The table below shows the overall figures.

UNS Ratio Swing
2010 General Election 37/40 38/40
2011 NAW Constituencies 33/40 35/40

In short, there wasn’t much in it, but Ratio swing actually marginally out-performed UNS each time.

Those of you still awake may be thinking: yes, all very historically interesting. But does it make any difference to projections based on current polls? To answer that, I looked at the most recently reported poll in Wales – not a Barometer poll but the BBC/ICM poll done immediately after the Scottish referendum. To remind you, this poll produced the following general election vote intention figures for the main parties (unfortunately the poll didn’t ask about NAW vote intentions):

Labour: 38%
Conservatives: 23%
Liberal Democrats: 7%
Plaid Cymru: 13%
UKIP: 14%

This represents the following changes since 2010:

Labour: +1.8%
Conservatives: -3.1%
Liberal Democrats: -13.1%
Plaid Cymru: +1.7%
UKIP: +11.6%

As I mentioned at the time the poll was published, UNS produces the following projected seat changes:

Labour: 28 MPs. (Holding all 26 seats won in 2010, and also winning Cardiff North from the Conservatives and Cardiff Central from the Liberal Democrats).
Conservatives: 8 MPs. (Losing Cardiff North to Labour, but winning – very narrowly – Brecon & Radnor from the Liberal Democrats).
LibDems: 1 MP. (Losing Brecon & Radnor and Cardiff Central, but holding Ceredigion).
Plaid Cymru: 3 MPs (Holding their current 3 seats).

But what about Ratio swing? It generates the following projections.

Labour: 28 MPs. (Holding all 26 seats won in 2010, and also winning Cardiff North from the Conservatives and Cardiff Central from the Liberal Democrats).
Conservatives: 8 MPs. (Losing Cardiff North to Labour, but winning Brecon & Radnor from the Liberal Democrats).
LibDems: 0 MPs. (Losing all their three seats).
Plaid Cymru: 4 MPs (Holding their current 3 seats, and gaining Ceredigion).

In short, there is only one seat difference – Ceredigion. Despite the BBC/ICM poll having UKIP support more than four times the level won in 2010 in no seat does this get UKIP even close to winning a constituency. The main difference that ratio swing makes is that with regards to the Liberal Democrats: with their support running at only .35 of the level in 2010, ratio swing suggests that all three Lib-Dems seats would be lost (and lost quite decisively). Ratio swing also makes the Ynys Mon result closer, but doesn’t quite tip it the way of Plaid Cymru.

As can be seen, most of the time UNS and Ratio swing don’t produce very different findings. Differences are mostly notable only where there are big changes in the support level of a party. Personally – at least until I see strong evidence to change my mind – I think that the Liberal Democrats remain favourites to hold Ceredigion, and probably also Brecon and Radnor. So, in the figures that I report for ITV from future Barometer polls, we will continue to use UNS. This is the general ‘industry standard’ approach. But for readers of this site I will also produce ratio swing numbers, and we can see where and when they differ. That will be fun, won’t it?


No comments.

Leave a Reply

Your email address will not be published. Required fields are marked *