How Twitter predicted the election result

After the fact, it was generally agreed that a rather older technology – television – had been the real breakthrough influence. The first-ever televised debates between party leaders triggered dramatic swings in the pre-election opinion polls, while Facebook, Twitter and the like barely featured in the electoral discourse.

But as a measure of public sentiment, as opposed to an influence upon it, social media did prove its worth during the election.

Tweetminster is a UK-based political social media analysis website that allows visitors to measure political sentiment towards a given topic by analysing the popular microblogging site Twitter.

The site also served as an informal election poll by measuring the ‘buzz’ surrounding parliamentary candidates in the days and weeks before the nation went to the polls. It did this by counting the number of times that each of them was mentioned. This was, according to co-founder and creative director Andrew Walker, just a “simple number-counting exercise”. But it was also surprisingly accurate.

Tweetminster’s final election forecast, made two days before polling, correctly predicted that the Labour party would receive 30% of the popular vote, and its predictions for the Conservative and Liberal Democrat vote shares were within 3% of their actual results. These findings were closer to the mark than polls by respected bodies YouGov and ComRes.

Accuracy in volume

Walker believes that social media analysis has the potential to replace traditional political polls, thanks to the volume of content available through sites such as Twitter. “If you’ve got 2.1 million ‘tweets’, the accuracy you can get from that is significantly higher than that from a sample of 4,000 to 5,000 people,” he explains. “It’s significantly bigger than any opinion poll is going to be.”

The exercise was not without its difficulties, Walker recalls, as certain election candidates skewed its measurement model. “Esther Rantzen is tainted candidate, from a data point of view,” he explains, referring to the former television presenter whose bid to become MP for Luton South generated disproportionate ‘buzz’ due to her celebrity status. In the event, Rantzen received just 4.4% of the available votes in her constituency and consequently lost the deposit required to stand.

The team behind Tweetminster chose not to sue sentiment analysis technology in its election polls, but Walker does see potential in the company.

Last year, Tweetminster used an internally developed algorithm to measure – via Twitter – reactions to announcements made by the Labour party at its annual conference. According to Walker, spikes in positive sentiment were clearly detectable as announcements on the suspension of ID cards and on electoral reform were made.

Interestingly, Walker believes that the reason why Twitter was such a good predictor of the election outcome is not that it taps into public opinion per se, but rather that it is very popular among journalists, spin doctors and politicians themselves.

“These are all people that form opinion,” he explains. “When you take that on board, you realise that the Twitter population is probably a reasonably good indicator of what the influencers think, which then rolls out onto the national population.”

Peter Done

Peter Done is managing director of Peninsula Business Services, the personnel and employment law consultancy he set up having already built a successful betting shop business.

Related Topics