Can we predict #GE2017 vote share using social media?

Mesmotronic Data

Introduction

When a snap general election was called in the United Kingdom on 18 April 2017, the inquisitive minds here at Mesmotronic asked themselves: can we predict #GE2017 vote share using social media?

With the accuracy of standard polling techniques coming into question following the 2015 general election and Brexit referendum, we decided to put it to the test.

Jump to charts | Jump to results

Methodology

We'll be gathering millions of pieces of data from social media (primarily Twitter) and processing it using a number of filters, bespoke algorithms and other predictive analytics techniques to convert metrics like message volume and sentiment into a predicted national vote share for each of the major UK-wide political parties.

At the end of each day of the campaign, until the election on 8 June 2017, we'll process the data and publish a snapshot and multi-day average to see how each day's events have affected our predictions.

Charts

Daily snapshot

This chart shows where the UK's national political parties stand after each day of the election campaign based on that day's data. Expect significant swings as people react to breaking news stories, policy announcements and other events.

Multi-day average

This chart uses a rolling 4 day window to iron out temporary spikes caused by news and other events, aiming to make it easier to spot trends as we move towards election day.

Significant events

Noteworthy news stories that may have significantly affected opinion during the election campaign:

Date Events

Result: within margin of error for all parties except Lib Dems

Our final daily snapshot before the election, based on analysis of around 1.5 million social media messages on 7 June 2017, was exactly right for Labour and UKIP, and within the margin of error (2%) for the Conservatives and Greens, although it slightly underestimated the Liberal Democrats:

  Party Predicted #GE2017 Difference
Conservative Party 45% 43% 2%
Green Party 1% 2% 1%
Labour Party 40% 40%
Liberal Democrats 2% 7% 5%
UK Independence Party 2% 2%

Wow!

Conclusion

We're genuinely excited about how close our final daily snapshot was to the actual general election result, and will be running additional analysis on the data leading up to it to see if we can work out whether there was a genuine last minute swing to Labour or if there were other factors at play. For example, if an underestimation of just how negative people can be on social media lead to our algorithms overstating the Conservative position.

It's impossible to know at this stage whether the accuracy of the final daily snapshot is something that could be repeated, but we'll definitely be running this experiment again in the future to find out!

Related links

Take a look at the your MP's expenses at MPsExpenses.info.