Tweetminster buzzing over election prediction

Another day, another declaration in support of the power of social media analysis in assessing public opinion – this time from Tweetminster, which has hailed as a success its attempt to predict the outcome of the UK general election based on Twitter mentions of candidates.

Like the Carnegie Mellon study reported yesterday, Tweetminster found that its projected share of vote for each party – 35% Conservatives, 30% Labour, 27% Liberal Democrats and 8% Others – generally mirrored that of the traditional opinion pollsters.

Like the pollsters, Tweetminster overstated Lib Dem support and understated that of the smaller parties. Its average error was 1.75 – less than ICM’s, on a par with Ipsos Mori, Populus and Harris, and more accurate than YouGov, ComRes, Opinium, Angus Reid and TNS BMRB.

(Full analysis of the performance of the traditional pollsters’ can be found here.)

“The results are too accurate to be accounted for by chance or coincidence,” writes Tweetminster. “They strongly suggest that the level of accuracy of the predictions are grounds for confirming the predictive power of Twitter is reliable.”

Up to a point. Tweetminster had sought to make accurate predictions for the way the vote would go in each constituency, but on that they achieved only a 69% accuracy. Here, sample size was key. In predicting the outcome at a national level, Tweetminster relied on more than two million tweets. Per constituency the average sample size was just 677 tweets.

“Predictions are susceptible to disproportionate media activity when considering predictions made from small samples,” says Tweetminter, giving the example of Esther Rantzen’s candidacy in Luton South. As a well-known TV personality, Rantzen generated a lot of buzz, and she was thus predicted to win by Tweetminster. In the event she came fourth.

So to answer our earlier question, can social media analysis be used to predict the outcome of elections? It would appear so.