Final straight


With the 2017 “snap” general election now only days away, Frontier has been examining a potentially rich, and under-analysed, source of predictive information – constituency-level betting odds. These suggest that the Prime Minister may have been more successful in reinforcing her vote in existing Conservative seats than in pursuing her ambition to make big inroads into Labour ones, but that Labour – despite a narrowing in the opinion poll gap – has been shoring up its own core support without making headway elsewhere.

It’s been an embarrassing few years for UK pollsters, with “shock” results in both the 2015 general election and the Brexit referendum defying the predictions of the majority of opinion polls.  At the national level, betting odds appeared to do little more than reflect the scale of these upsets, suggesting that there was little chance of either a Tory majority in 2015 or a victory for the Leave campaign in 2016 until the results began rolling in.  At face value, this may seem to suggest that the “wisdom of crowds” that betting markets notionally reflect is of little more value than opinion polls when it comes to predicting election outcomes.

Look beyond the national level, however, and a more interesting picture emerges. Without constituency-level opinion polls, systematic insights into voters’ views at a local level can be hard to come by.  Betting markets can help fill this gap for general elections, since odds are available for each of the 650 UK parliamentary constituencies. Constituency betting odds offer a rich potential source of information to the parties because they allow them to track – in near-real time – how their policy proposals and messages are playing out in different local areas.  This is particularly useful in a first-past-the-post election, where small changes in the local distributions of voters’ preferences can have a substantial impact on the results in terms of numbers of MPs.

Moreover, changes in constituency-level betting odds may have some predictive power – at least in helping to track changes in opinion over time. This can be seen from the charts in Figure 1 below.

Figure 1.        Change in implied probabilities between 2010 and 2015 versus change in share of vote (Conservatives – left chart; Labour – right chart)

Figure 1 ru v10001

Note:  The charts report the results for more marginal constituencies where the party had odds of at least 10% in 2015.
Source: Frontier Economics analysis of odds data.

The two charts in Figure 1 focus on constituency-level changes in the performance of the Conservative and Labour parties between the 2010 and 2015 general elections. Each dot represents a specific constituency. For each chart, the horizontal axis measures the change in the party’s probability of winning implied by comparing its odds ahead of the 2010 general election against its odds ahead of the 2015 election five years later. The vertical axis plots the change in the party’s share of the vote between 2010 and 2015. These two variables appear to be positively correlated. Of course, this correlation is far from perfect (and stronger for Labour than for the Conservatives, who outperformed expectations in 2015 in a number of constituencies), but it does suggest that changes in odds have some predictive power at the constituency level.

The data are probably most useful in providing indicators of change during a campaign, at a level of disaggregation that makes it possible to test some favourite theories as to what has moved the dials of public opinion. So what do the constituency-level betting odds suggest about the story of the 2017 election race so far? What have been the campaign hits and misses? And what do they indicate about the possible outcome on 8th June? As Figure 2 below shows, the polls would suggest that the race has tightened considerably in recent weeks, and some organisations – such as YouGov – have adopted modelling approaches that are predicting a hung parliament. What do trends in constituency-level odds imply?

Figure 2.        Trends in “poll of polls” , 2017 to date

Figure 2_v2

Note: Each dot represents the results of a specific poll for a specific party; the lines show the moving average trends.
Source:  BBC.

Bolt for the blue

Figure 3 below plots the changes in implied probabilities for the two main parties between the build-up to the 2015 general election and mid-May 2017 – shortly before the Conservatives published their Manifesto: grey dotes denote constituencies where the probability of winning fell, while the parties’ respective colours indicate consistencies where the probability increased.  The darker the colour, the greater the increase in implied probability.

Figure 3.           Change in Conservative and Labour implied probabilities between 2015 and 19th May 2017

Figure 3_v2

Note:       Data excludes Northern Ireland.
Source: Frontier Economics analysis of odds data.

In some constituencies, both the Conservatives and Labour saw increases in their probabilities – reflecting the worsening odds of third parties such as UKIP. Nonetheless, the overall picture is stark: as of mid-May, the Conservatives’ probability of victory had increased in almost 90% of constituencies, whereas the equivalent figure for Labour was just 40%.  And while a Conservative win probability increased by 50 percentage points or more in 79 constituencies, there were no constituencies where Labour experienced such an increase. No wonder the Prime Minister felt comfortable calling a “snap” general election when she did.

The Corbyn comeback

Figure 4 below shows how the implied probabilities from betting odds have evolved since mid-May. It shows that the recent picture seen at the aggregate level through opinion polls has also occurred in individual constituencies: Labour’s chances have increased in almost 60% of constituencies over this period while the Conservatives’ odds have fallen back in almost 80%.

Figure 4.           Change in Conservative and Labour implied probabilities between 19th May and 2nd June 2017

Figure 4_v2

Notes:     Data excludes Northern Ireland; white dots denote constituencies with no change in probability.
Source: Frontier Economics analysis of odds data.

Campaign hits and misses

The picture painted by the figures above very much matches that painted by developments in the opinion polls – the Conservatives well ahead in the early stages of the campaign, but Labour then regaining ground from mid-May. No surprises there.

However, dig deeper and things get more interesting.  Figure 5 below suggests that the swing from the Tories to Labour over the past fortnight has been disproportionately concentrated in Labour-held seats and in Labour strongholds (Wales, North West, North East, London). By contrast, the odds in Conservative-held seats seem to have hardly shifted at all, although this aggregate result may conceal some individual variations.

This would suggest that – while Labour’s recent recovery may have damaged the Conservatives’ prospects of moving into “core” Labour territory – it has made little headway in Conservative-held constituencies.

Figure 5         Percentage point changes in Conservative and Labour implied probabilities between 19th May and 2nd June, by seat type and region

figure 5 again

Note:       Data excludes Northern Ireland.
Source:   Frontier Economics analysis of odds data.

Who cares about care?

One of the biggest policy stories of the campaign has been Mrs May’s rapid change in her Manifesto’s proposals for social care, which originally set no ceiling on the amount that might be deducted from people’s assets after their death (subject only to the maintenance of a residual £100,000).  This was reported by Tory candidates to be playing badly “on the doorstep”.

But how significant was the effect of this announcement?  How much does it help to explain the shift in opinion seen in the last few weeks? To test its effect, in the chart below we have plotted the change in the two main parties’ odds between 2nd June and 19th May against the proportion of people aged 65+ in each constituency.

As the Figure 6 below shows, there’s no clear relationship. In fact, if anything, it suggests that the Conservatives’ odds have weakened slightly more in constituencies with fewer pensioners. The picture is one of a broad-based fall in Conservative support across the spectrum since mid-May.

Figure 6.           Variation between change in implied probabilities 19th May vs 2nd June and proportion of constituency population aged 65+

Figure 6_v2

Source: Frontier Economics analysis of odds data overlaid with ONS constituency demographic profiles.

You get a similar picture if you plot the changes in probabilities against average income – a broad-based reduction in Conservative support and a broad-based recovery in Labour support. There’s some evidence that the Labour recovery has been stronger in areas with lower wages – which would play to the story of Labour regaining its grip on its traditional support base – but it’s pretty marginal.

Figure 7.           Variation between change in implied probabilities 19th May vs 2nd June and constituency median gross weekly pay

Figure 7_v2

Source: Frontier Economics analysis of odds data overlaid with ONS constituency wage profiles.

This either suggests that the social care proposals are equally unpopular with all demographic groups, which seems unlikely; or that it was the combination of policy and reversal that dented the Prime Minister’s image with all groups; or that it’s not, in fact, these proposals that have been the primary driver of the decline in the Tories’ odds.  The shoring-up of the Labour vote in Labour areas may suggest that, despite Labour debacles over the costing of their policies, traditional loyalties have kicked in; or that Jeremy Corbyn has proved more appealing, at least on his home ground, than many in his own party expected.

Home truths

However, all these results still suggest only a relative improvement in Labour’s standing in recent weeks, and do not indicate that Jeremy Corbyn is making inroads outside Labour territory.  Many factors continue to point firmly in the Tories’ direction.  Our analysis supports the assumption that they are well-positioned to attack Labour seats where the UKIP vote was strong in 2015 (though much less so to contest Labour seats where the Liberal Democrat vote was strong).

Most interestingly, perhaps, the odds in SNP-held seats do not seem to have shifted much as the Tory lead in the opinion polls has come down: these odds have hardly budged since mid-May. So the Tories can take some comfort that their odds of picking up seats in Scotland haven’t been dented by the campaign.

So what were the standings going into the final week of the election race? As the chart below shows, as of 2nd June, the number of constituencies in which the Conservatives were strong favourites still clearly outweighed the number of constituencies where Labour had a strong lead.  Excluding Northern Ireland, the betting odds imply the Conservatives have a probability of victory of 75% or more in 321 constituencies, while the equivalent figure for Labour is 133.

Figure 8.           Conservative and Labour implied probabilities by constituency as of 2nd June 2017

Figure 8_v2


Notes:     Data excludes Northern Ireland; white dot denotes the Speaker’s seat.
Source: Frontier Economics analysis of odds data.

All this suggests that while the Tories’ hopes of a landslide have receded, Labour isn’t much better placed to compete for seats the party doesn’t already hold than it was three weeks ago. And that leaves it still quite a long way from the hung parliament that YouGov’s recent models have been indicating … with the often-crucial last two days still to go.

Comments are closed.