During the run-up to the recent general election, my old friend Uniform National Swing (UNS) was given a very hard time by lots of people. A number of journalists and other commentators, and many people from within the parties, suggested that we were going to see such a complex pattern of changes across different nations, regions and even individual constituencies that UNS was positively misleading; I was even told directly by someone from one of the parties that using it to project the potential seat implications of opinion polls was ‘utterly ludicrous’.
I think that some of these criticisms, at least, misunderstood how analysts generally deploy assumptions like UNS. Of course those using UNS to project the potential seat implications of an opinion poll don’t believe that any future election will really see a party’s support levels change by an exactly uniform percentage right across the country. (We might be stupid, but we’re not that stupid). As I have tried to explain on the blog previously, I don’t see UNS as any sort of infallible prediction tool (which was why I was always very careful to make clear here that I was making projections of the potential seat implications of polling numbers, and definitely not predictions). I think UNS provides us with a useful baseline, or benchmark, against which pre-election we can assess the tasks facing parties in individual seats, and post-election we can evaluate their achievements. But it is nothing much more than that.
That said, in the wake of the election I thought that it might be useful to look through the forty Welsh constituencies, to see how well UNS performed in ‘predicting’ the outcome in each one. That is, by taking the national swings in vote share between 2010-15, how accurately can we project the result in each constituency? If the critics of UNS were right, we should find significant numbers of seats producing results different from those implied by the national swings in vote share. And as I was doing this exercise for UNS, I thought it would also be a good idea to check on the outcomes for Ratio Swing (RS) as well.
(A brief reminder for those of you a little hazy on the detail here. UNS simply means projecting the change in national vote share for each party onto each constituency. So if a party has seen its vote share go up by 1.1% nationally, you just apply an increase of 1.1% to its result in 2010 in every seat to get the projected result in 2015. Ratio Swing models the change in the ratio of a party’s support. So if a hypothetical party, let’s call them the Liberal Democrats, were seeing their vote share fall by two-thirds nationally, you would then project them for 2015, in each constituency, to see their vote fall by two-thirds on their 2010 total).
The table below shows the forty Welsh constituencies. In the following three columns I then list the party projected to win the seat under UNS, under RS, and the party that actually won the seat. I highlight in bold any cases where the seat is projected to have been won by a different party from that which actually won it.
|Vale of Clwyd||Labour||Labour||Conservative|
|Alyn & Deeside||Labour||Labour||Labour|
|Brecon & Radnor||Conservative||Conservative||Conservative|
|Carmarthen West & South Pembs||Conservative||Conservative||Conservative|
|Carmarthen East & Dinefwr||Plaid||Plaid||Plaid|
|Vale of Glamorgan||Conservative||Conservative||Conservative|
|Cardiff South & Penarth||Labour||Labour||Labour|
|Merthyr Tydfil & Rhymney||Labour||Labour||Labour|
As we can see, there are not many cell entries in bold. UNS projected the correct winner – albeit not necessarily by the correct percentage – in 38 of the 40 seats in Wales. The only exceptions are the two surprise Conservative gains on the night: Vale of Clwyd and Gower. RS performs almost equally well – perhaps unsurprisingly as it projects the same result as UNS in all-but-one seat. That seat is Ceredigion, which UNS correctly projected to be a Liberal Democrat hold, but RS projected as a Plaid Cymru gain.
So, UNS projected the correct winner in 95 percent of all seats in Wales; and RS did so in 92.5 percent. I don’t think that’s at all bad – those are success rates that some of the very mathematically sophisticated election prediction models developed for 2015 certainly did not attain. I’ll therefore continue to use UNS as the main basis for projecting the potential seat implications of any polls conducted in Wales, while also reporting for readers of this blog figures from an RS projection. I hope you don’t think that is too ludicrous.