Mapping Brexit votes using novel e-petition data and machine learning

21st Nov 2018
PGRG Blog #15
Nik Lomax

In our paper “Estimating the outcome of UKs referendum on EU membership using e-petition data and machine learning algorithms“, recently published in the Journal of Information Technology and Politics, we use novel e-petition data and machine learning algorithms to estimate the Brexit leave vote percentage for UK parliamentary constituencies.

On 23 June 2016, 52 per cent of the UK population voted in favour of leaving the EU (turnout 72 per cent of registered voters). Results were published for 382 ‘Counting Areas’ (generally the same as local authorities), but not for the 632 Westminster Parliamentary Constituencies, the geography at which elected members of Parliament are held to account by their constituents. In this paper, we argue that having an accurate view of the result for Constituencies is hugely important, given that they are the democratic geography of the UK.

Using a diverse range of eight machine learning algorithms, implemented via the R package carat, we used the known result for aggregations of Constituencies and Counting areas to ‘learn’ what the referendum result might be, using information from the UK government’s e-petition website. Anyone can set up an e-petition, relating to any topic they choose, and the petition can be signed online by members of the public. The government will respond to any e-petition that receives over 10,000 signatories, while those that receive over 100,000 signatories are additionally considered for debate in Parliament. We argue that the themes of petitions signed by constituents gives a great deal of insight in to the political sentiment of that area.

We then used these learned results to predict a sample of known outcomes. We found that our best performing Cubist algorithm correlated at the 97% level with the known results. Using the method to predict the referendum outcome for all Constituencies in the UK, we compared our results with two other studies which have attempted to do the same thing (using different methods) and found that our results compared favourably to both: a correlation of 97% with the estimates produced by Chris Hanretty and 96% when compared with estimates produced by Nigel Marriott.

This is not the only work we have done with these e-petitions data. In a paper published in EPJ Data Science last year, we mapped voter sentiment by creating a classification of Westminster Parliamentary Constituencies. We find there are four distinct clusters: Domestic Liberals; International Liberals; Nostalgic Brits and those with Rural Concerns. By focusing on the topics of the petitions signed within each Constituency, we provide insight in to the views which shape political debate, for example the two liberal clusters were generally more anti-Brexit while the Nostalgic Brits and Rural Concerns were far more conservative in their views.

From our work to date, we conclude that e-petition data is an informative and versatile source of information that allows us to gauge political sentiment in a given location. We argue that machine learning algorithms offer scope for researchers to gain confirmatory or alternative insight in to a range of problems. Both the data and the approaches outlined here would be of use to population geographers because (1) the e-petition data convey views and attitudes of a large number of people across space, and (2) the methods provide an effective way of interrogating large geospatial datasets.

Read the full Journal of Information Technology and Politics paper here, and you can see the slides from a presentation that Nik gave at the 2018 British Society for Population Studies annual conference in Winchester here.

Figure 1. Categorisation of Westminster Parliamentary Constituencies as those who voted to leave or remain in the EU. Hard Leave and Hard Remain are those areas where our results agree with those of both Hanretty and Marriott. Soft Leave and Soft Remain are areas where two out of the three different approaches agree the area was leave or remain. Both maps convey the same information but the cartogram (right), where each WPC is represented as a hexagon, provides a clearer picture. This is because in the conventional map (left), larger constituencies dominate the picture at the expense of smaller inner city constituencies. For a useful commentary on mapping election results, including the use of hexagon cartograms, see this interesting post by Kenneth Field.

 

Nik Lomax
University of Leeds
Communications Officer of the RGS-IBG Population Geography Research Group


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.