by Rob Marchant
Much has been made during 2012 of Labour’s solid poll lead, which has been of the order of 10% ever since the government’s disastrous Budget. For some it seems difficult to refrain from mentally converting this into a line on Peter Snow’s election night model of the House of Commons, showing a majority for Labour.
What this fails to account for, quite apart from the changed circumstances which may prevail in two years’ time (and which may be much more attractive for the Tories, as Peter Kellner cheekily points out here), is that incumbents tend to have a dip at mid-term anyway. Quantifying this effect would obviously help us to make more accurate forecasts, to the extent that this is possible.
Last week at Uncut, Atul Hatwal showed the numerical arguments against a win for any party which failed to establish a poll lead on the economy. In a complementary way, we can try and allow for this mid-term effect, to try and get a better idea of where things might end up, all other things being equal.
There are various ways of trying to gauge it*. A year ago, an excellent piece of analysis was carried out by Leo Barasi (and for which I am indebted to John Rentoul for pointing out) on post-war polls both eighteen months into a parliament and two years out from the next, which concluded, rightly, that Labour’s prospects were numerically better than many seemed to think.
Interestingly, this conclusion now looks a little less secure, not because Barasi’s model has changed or become less applicable, but because the polls have actually improved for Labour in the last year and therefore many commentators, who were very negative at the end of last year, have gravitated towards a much warmer view of Labour’s prospects.
To revisit the second part of that analysis – that of two years out from the next election – is now both timely and illustrative, which we do now in this graph:
It is interesting to look at the regression line but the data points, too, tell a story. It is also useful to apply a little political history to the figures.
First, the most striking thing to note is that over four-fifths of postwar oppositions have had a lead in the polls at this point, two years out from an election. Irrespective of whether they go on to win or lose. So having a lead in itself is hardly remarkable: in fact, allowing for outliers**, they almost all do. It is the size of the lead that counts. Conclusion: don’t get excited about being ahead in the polls at this point. It seems to be a necessary, not a sufficient condition. Not being ahead, now that would be something.
Second, since the war, no opposition has won without polling at least 6% two years out. In fact, this lowest-polling election-winner from opposition, two years out, was Labour in 1964 (which Barasi classifies as an outlier). The regression implies that, on average, it should be at least 12% – comparable to today’s lead of 11% – for a marginal win on vote-share, and in fact the other four cases were all 14% or more.
Moreover, historically, 1964 was clearly an “only just” election; Wilson won by a whisker, and if he had not had the anomalous effect of no small parties (outside the main parties, there were only an unprecedentedly small number of 9 Liberal MPs; no unionists, no nationalists, no independents), he would have had to form a coalition. In the end, he had to call another election two years later to get a proper working majority.
Third, we can get a rough measure of the “mid-term effect” by looking at the line of best fit, which is interesting. In general, the lead not only narrows very significantly at the election, it also tends to swing a few percentage points in favour of the incumbent.
So, in the case where the party ahead in the polls has gone on to win, we can see that final lead has been, on average, around a third of the mid-term lead, less 3.64%. Our mid-term lead of 11%, now projected out into a general election lead, is not in fact 11% – as our intuition might suggest – but close to zero.
There are some significant caveats to all this. For a start, there is obviously a significant margin for error on either side. And, as a general rule, we must be wary of the argument “no-one has ever won an election without X”. As this cartoon and article brilliantly point out (h/t: Anthony Wells), you can always apply this argument, irrespective of the relevance of X as a factor. For example, no African-American had ever become US President – until Obama.
However, if we can accept the idea that this case is different and that there is at least some fairly strong correlation between polling levels and election results – which seems eminently reasonable, given that one is supposed to be a direct, if error-prone, measure of the other – it should surely give us some pause for thought (there are also some more technical caveats***).
Conclusion: it is clear that the next election is far too close to call and, as that Wells points out at the excellent UK Polling Report, despite Labour and the Tories both playing down their chances, someone has to win it. It is highly questionable to try and make a prediction now on who will win or lose. It is also true that there is still time to upset these historical precedents and for Labour to secure a full majority, or for 2015 to be an outlier, like 1964.
However, one thing that we can conclude is that it is a highly unsafe bet that Labour will win a majority, despite what its current 11-point lead says, because of this mid-term effect. When we combine it with Atul’s results on economic polling, the picture for Labour looks still more tenuous.
So, Labour needs to establish a much stronger poll lead if it is seriously aiming for a majority in 2015 and allowing a sufficient margin for error. Currently, we are most likely to be right on the edge of winning (that is, winning the highest proportion of the vote, not a majority) – no more.
In fact, because of Labour’s higher standing in the polls and commentators’ more favourable perceptions, it is arguable that the reverse is now true of Barasi’s (informal) conclusion, that many seemed to be underestimating Labour. It now seems that some commentators are now seeing Labour’s win as somehow a strong probability, when its position is in fact rather more tenuous. On the basis of this model, one slip and we would not even have the highest vote-share, let alone a majority.
In short: hold the champagne.
*It might also be interesting to use the base data to regress the average “mid-term dip” for incumbent governments against (i.e. the % reduction in polling between an election win and the polling two years before the next one) another variable, something which I confess I have not had time to do.
**There were only three exceptions to this out of 16 data points, and two of these were the elections of 2001 and 2005, where the previous landslide victories for Labour meant the Tories were starting from an exceptionally low base of support and did not have time to recover. We are clearly not in that scenario because of the relative closeness of the 2010 election. So if we exclude these as outliers, it is only one exception out of 14 data points.
- The R2 is a respectable 0.4, but Barasi himself points out that, when at the margin, the polls two years out are potentially not great predictors. For this reason we don’t try to predict who will win.
- The number of data points is clearly less than we would like.
- A few odd general election data points have been skipped, for very good reasons to do with the inapplicability of two elections in quick succession such as 1974 – see Leo Barasi’s source for more details.
Rob Marchant is an activist and former Labour Party manager who blogs at The Centre Left