How to read a constituency
(Making sense of vote intention polling in (very) uncertain times)
With talk of an early general election, polling volatile and unstable at the national level and the first past the post electoral system under increasing strain it is not a simple task to say with any confidence what the outcome of an election might be.
There are numerous tools around the internet which allow you to predict an outcome based on a national opinion poll but these are of limited use when tactical voting is likely and new divides are emerging.
There is, however, a new kid on the block, MRP (multi-level regression and post-stratification), which produces estimates of results at constituency level (though not limited to constituencies this is the most common application). These techniques use a combination of large scale polling and known characteristics about constituencies to infer to the constituency level. You may be thinking you have never heard of this, but almost all the constituency level estimates of EU referendum voting are based on the work of Professor Chris Hanretty using these techniques. In 2017, Yougov’s MRP model predicted a hung parliament and estimated Labour would win in Canterbury (to widespread derision at the time). It is likely that in an election with four or more competitive parties that these kinds of models will be the only game in town or at least the only game worth playing. The difficulty however is that these techniques require both expensive data collection and someone with the technical skills to produce the models; neither of which are available to the amateur political observer, campaigner or voter armed only with access to google.
This piece demonstrates how to do you own ‘modelling’ of a constituency using only data which can be accessed by anyone with an internet connection. This will not give a ‘nowcast’ of the House of Commons post-election and shouldn’t be thought of as a prediction but I hope will allow people to think about scenarios in their own constituency, or those they are campaigning in.
A little additional context is needed to explain the assumptions behind this approach. As I have argued elsewhere, our politics is structured by two dimensions of political values an ‘old’ dimension related to issues of economic justice (often simply referred to as ‘left-right’) and a ‘new’ dimension related to social issues, which here I call social liberalism — social conservatism (elsewhere I have used liberal-authoritarian for this dimension, there is much less agreement on the best way to characterise this set of values with some simply calling it the ‘other’ dimension).
This chart illustrates how voters for each of the parties (broken down by their EU referendum vote) are positioned in a ‘value space’ defined by these two dimensions (details on measurement of the scales can be found here). The chart shows how these two divides produce distinctive groups of voters in the electorate. On the basis of economic issues alone Labour (Remain and Leave), Green and UKIP voters are indistinguishable, with LibDem voters marginally closer to the centre of the scale. Conservative voters are more markedly to the right of the other groups, a feature which may explain why it has so far proven more difficult for the LibDems to capture this vote. However, on the social liberalism scale a large gap exists between Labour remain and Labour leave voters and between UKIP (leave) voters and Green (remain) voters. This set of values is what distinguishes leave and remain support. For Labour, the Liberal Democrats and the Green Party their ‘remain’ supporters are in the same part of this value space, they are all ‘fishing’ in the same pool of voters. Conservative remain supporters are a little more distinctive they are also more socially liberal than Conservative leave supporters (and increasingly their party leadership). Where voters are positioned closer together it is likely to be easier for them to move between parties. A voter whose priority is economics could very easily move from a Labour leave position to the Green party (for example).
Unfortunately, we do not have easily available data which tells us how many socially liberal voters there are in any constituency, we don’t know how many socially liberal voters there are in Hastings and Rye or how many economically left-wing voters there are in Swindon South. As a result it is difficult to use what we know about values to work out likely constituency results. But these value divides are rooted in social structure and we do know some key social characteristics of constituencies that can help us.
While the economic ‘left-right’ dimension is linked to income, wealth and social class; the social liberalism scale is most strongly connected to education levels. How education is connected to voting behaviour hasn’t been a key concern for many (it is rarely included in polling tables) but this is important for understanding the votes of parties who are primarily competing on the social liberalism axis. In 2010, when the Liberal Democrats won 57 seats and were ‘kingmakers’ in a hung parliament, they were the most popular party among those with degree level qualifications.
At that election they won around 1 in 3 votes among those with degrees. Few of the opinion polls give us breakdowns by education but for those that do this pattern is being replicated in recent polls. Moreover, the pattern of signatures to the petition to revoke article 50 (that gathered over 6 million signatures in early 2019) at constituency level correlates exceptionally strongly with education levels in constituencies.
This suggests that for understanding where the Liberal Democrats might be able to outperform their national polling (and their 2010 result) we should look for constituencies with relatively high proportions of the electorate with degree level qualifications.
Polling also suggests that the Brexit Party is around the level that UKIP achieved in 2015. Analysis of Brexit party support in the EU parliament elections suggests that there is considerable overlap between Brexit Party support and UKIP support at earlier elections. The assumption made here is that he Brexit Party will largely be a straight replacement for UKIP among voters and that the party will achieve a similar share of the vote.
The working assumption is that the 2019/20 election is likely to be a mash-up of the 2010 election and the 2015 election rather than a re-run of the two-party outcome of 2017. Using these basic assumptions about the shape of British politics we can make some informed guesses about what might happen in a constituency.
D.I.Y Constituency Modelling
Based on this I start any profile of a constituency by looking at the election results from 2010, 2015 and 2017 (these are available in lots of places, for individual constituencies I use Wikipedia, for analysing a set of constituencies at once it is more efficient to download a complete set of results.
From this the ‘starting point’ is that the LibDem share will be around their 2010 result and the Brexit Party share will match UKIP’s 2015 result.
In order to work out the impact this has on the other parties we need to make some best guesses about where these voters will come from, and to do this we can work out what happened between 2010 and 2015 (when the LibDem vote collapsed and UKIP rose) and between 2015 and 2017 when the UKIP vote collapsed. We can then apply those guess to get to our ‘base-line’ for the next election.
We can see how this works by looking at East Surrey. After the defection of Sam Gyimah from the Conservatives to the LibDems, the widespread view was this was misguided and that his East Surrey seat was as Conservative as you could get. A tweet suggesting that I could see a path to LD victory there was treated with a certain amount of derision. So let’s show my workings.
First the election results from 2010- 2017 in the constituency (one bonus of ‘snap’ general elections is that they give us longer time series of data without boundary changes intervening).
The LibDem share is likely to get back to 26% in this constituency on a par with 2010. Where are those votes going to come from? When the LibDems collapsed in 2015 both Labour and Conservative vote shares went up (despite the UKIP surge occuring at the same time). So my working hypothesis is that they both gained from the LD collapse and so to give the LibDems 26% will mean taking some votes from both parties. Assuming at the same time the Brexit Party matches the UKIP 2015 result means also taking votes off of each of the two main parties but here probably slightly more from the Conservatives. So my overall starting ‘guess’ for the constituency has Labour back to their 2010 position, the Conservatives weakened by losing voters on both flanks simultaneously (whereas previously they’d only had either strong LibDems or strong UKIP not both).
From this base we can work out a range of scenarios that could get us to a LibDem victory in the seat. First, we could imagine the LibDems gaining on their 2010 position, mostly at the expense of the Conservatives, as socially liberal Conservatives look elsewhere to cast their votes. On it’s own this doesn’t seem likely to give the LibDems victory (though keeping an eye on current polling gives you an indication if this becomes more likely as a campaign develops). But if we additionally give Sam Gyimah a small ‘personal vote’ it could be enough to tip the balance. Other possibilities could be the Brexit party polling higher than the UKIP 2015 share which would be likely to be at the expense of the Conservatives.
In some seats where the positions are even more complex, adding in further contextual information could be helpful. For example, the estimates for the remain share in the constituency (which can be found on the democratic dashboard pages) and demographic information from the census. This is a little more fiddly to access but can be found via http://www.constituencyexplorer.org.uk/.
For East Surrey we can add into our DIY model that the constituency has an estimated 45.8% remain vote in 2016, and has 31.1% of the population with level 4 qualifications (higher than average for the South East and nationally). The high level of qualifications in East Surrey suggests that there is a base for the LibDems to build on but there is also a path to an (unlikely) Labour victory here should Labour manage to unite the remain vote while the leave vote splits across the Conservative party and the Brexit party (though current polling suggests it is the remain vote that is more fragmented).
Looking across the demographics, the scenarios and polling, it would currently seem to be most likely to be a contest here between Conservatives and LibDems with the outcome likely to depend as the Conservatives holding on to more of their Brexit Party leaning support than they lose to the LibDems. But as you can see from the different possibilities, in a volatile election with multiple parties competitive in seats, even unlikely outcomes can’t be ruled out.
There are a multitude of uncertainties around the next election (not least when it will happen). Will the LibDems make further gains during a campaign or lose ground back to Labour? Can they grab some disaffected socially liberal voters from a Johnson led Conservative party? Will Corbyn again manage to overcome the odds of exceptionally poor personal ratings? And perhaps most uncertain of all how will the Brexit Party present itself and fare with voters in whatever the Brexit landscape looks like after the end of October.
Polling will help us to see how some of these factors are shaping up during a campaign and combining that will a detailed look at a constituency can at least provide a hand drawn map to guide us through what will be a fascinating election campaign.
If and when an election is finally called, I will write up this reasoning applied to selected constituencies as the campaign develops — do come back later for more!