Yesterday, George Osborne stepped up to the dispatch box in Parliament to deliver one of his biggest speeches of the year: the autumn statement. Surrounded by the jeers and shouts of his fellow MPs, this was one of his big chances to set out the Government’s vision for the economy, and defend their track record in managing it. As he did so, Twitter lit up.
Social media, particularly Twitter, has emerged as a new political battleground, where opinions are formed and public judgments made, whether politicians like it or not. Over the hour of the speech, 46,000 tweets were sent about the autumn statement – 780 a minute, and more than in both of the party conference leader speeches earlier in the year.
The Twittersphere cannot claim to be representative of the general population; but it does allow us to capture in-the-moment reaction, and give us insight into a set of uncensored attitudes. Demos, the pollsters Ipsos Mori, computer scientists at the University of Sussex and a tech start-up, CASM, yesterday launched a new project to build new, more robust ways to listen to all this digital activity. We used our technology to train algorithms to try to make sense of all these tweets, far more than we could manually read. Here is what we found:
This wasn’t just something for the Westminster bubble – those lobby hacks, politicians, parliamentary researchers and think tanks of professional politics. The autumn statement mattered to people. Tweets flew in from all over the UK, with strong coverage from all over the UK, including reaction from the ‘northern powerhouse’ and devolved nations.
One of the analytical challenges is being able to see the wood through the trees of Twitter: can we identify what the ‘general public’ think? In total, around 18,000 tweets came from members of the general public; close to ten thousand came from corporate Twitter accounts, especially tax, accountancy and management consultancy firms. A further 10,000 came from media voices – newspapers, the broadcasters, bloggers and writers. Politicians, charities, public sector organisations and activists were vocal too. These groups all mixed together to form a constant hum on Twitter: reporting on what Osborne was saying; criticising, praising, and spinning his announcements; fact checking his figures; and of course, joking.
The graph below shows the peaks and troughs in volume of reaction to the statement by these different user groups.
Twitter often now reacts to important political events. But when we trained algorithms to sort the autumn statement tweets that were booing from those that were cheering, we heard something strange - different from the Euro debate between Clegg and Farage, different from the independence referendum debates between Darling and Salmond, and the party conference speeches by Miliband and Cameron. Some people were cheering.
Twitter, as a rule, is unfavourable to politicians. It is a place where people join in the cross-partisan, cross-generational national sport of anti-politics, and in each of these occasions the boos have drowned out the cheers. In the first referendum debate, Darling was booed 30,000 times and cheered 435. For Miliband’s speech, he received 4 boos for every cheer. For Cameron’s, 10 boos for every cheer. And they, in the Party Conference, were playing to their home base. Boos represent not only disagreement with the political and economic cases put forward, but also the personalities of the key players: the clothes worn, the jokes made, and judgements of their character.
It started the same for Osborne. As he kicked off, the general public questioned Osborne’s economic record, previous promises to cut the deficit and his trustworthiness on key issues such as poverty and the NHS. But then he started making pledges, first on corporation tax, then on the postgraduate student loan – and some people dared to admit they like what they heard. The cheers started to rise, and by the time he came to cutting stamp duty, for a brief moment, the cheers peaked above the boos. Over the hour of the speech, Osborne received only 3 boos for every cheer, which is a remarkable achievement on the Twittersphere.
But if this was a good day for Osborne, it is only in the context of politicians struggling to have credibility or any degree of popularity on social media spaces. The hashtag #cameronmustgo surged through discussions on the autumn statement; as did the other two most popular phrases: “blah”, and “fault”. People also focused on what Osborne was not talking about – especially poverty, zero hours contracts, disabilities, and the environment. For many, Osborne’s pledges were not enough and not to be trusted.
Social media is fast becoming an important political space, and, in the run-up to the general election, it will be a hotly contested one. However, listening to this new and important space is not an exact science. To paint a picture of what is happening online, we need to build algorithms that deal with far more tweets than we could ever read - tens or hundreds of thousands of tweets – even millions. These algorithms don’t always get it right. For the research here – they probably correctly understood the broad meaning of a tweet around 70% of the time. The research struggles in other areas too – in understanding how social media reflects society and how it doesn’t, in knowing what to do about demographic biases, and in making this kind of research live up to some basic standards of how we go about researching society and studying attitudes. But if we knew how to do all these things, there wouldn’t be much need for our project.
Find out more:
Carl Miller is Research Director of the Centre for the Analysis of Social Media at the think tank Demos.
Steve Ginnis is Head of Digital Research at the Social Research Institute, Ipsos Mori.
This work was funded by Innovate UK, the ESRC and EPSRC