Donald Trump’s success in the recent U.S. presidential elections caught most pollsters and political analysts by surprise. Both the leading poll aggregation website FiveThirtyEight and prediction market aggregation website PredictWise had their odds firmly on Hillary Clinton. A similar scenario had played out earlier this year in June with Britain’s vote in favor of the Brexit and with the French presidential election primaries in November.
Whether silent supporters or other factors are to blame for the prediction misses, the question arises: would monitoring political opinion from social media have provided a better prediction? Is measuring the vox populi a problem best left to data scientists?
In 2010, a now highly cited article studied the 2009 German Federal Election and found “that the mere number of messages [on Twitter] mentioning a party reflects the election result”. They observed that just counting the tweets with, say, #CDU vs. #SPD, the two biggest political parties in Germany, gave a good reflection of the eventual election outcome.
However, those early claims of success did not remain without criticism and attempts at replicating the approach in other settings remained unsuccessful. Still, in the context of the U.S. elections there was again evidence that, based on an analysis of Twitter data, Donald Trump was winning the social media war – and eventually the election.
So what can we gauge from social media concerning public opinion? Is simply counting tweets in a “one tweet, one vote” manner the future of election predictions? And how could this work in times of Twitter bots and astroturfing?
In our book “Twitter: A Digital Socioscope”, domain expert Daniel Gayo-Avello explores these topics in depth in his chapter on Political Opinion, providing an in-depth view of latest research dealing with the biases, misinformation, and state-of-the-art electoral forecasting. Comparing Twitter to an idealized Habermas’ “public sphere”, Gayo-Avello critically evaluates the computational approaches to public opinion tracking, identifying, for example, perhaps the most pressing issue of dubious truthfulness in social media: “There is an additional argument against the simplest approaches to mining opinion from Twitter data: […] ignoring that misleading information (disinformation) is not uncommon when discussing politics, and even expected during electoral campaigns.” This notion has recently been formalized by Oxford Dictionary as post-truth. Read more from Gayo-Avello on possible ways to overcome it.
Find out more about Twitter: A Digital Socioscope