Special for the Armenian Weekly
If you’ve watched the last season of “House of Cards,” you have an appreciation for the role that data and social media can play in elections. Few know, however, that a German-Armenian researcher was among the first to analyze public Twitter data to see if election results could be predicted by the 140-character messages that the masses launch into cyberspace.
Dr. Andranik Tumasjan’s academic journey has led him through Germany, China, and the U.S. Currently, he is a postdoctoral researcher at the Technical University of Munich. Born in Yerevan, he moved to Germany with his family in his early childhood.
In 2010, when the total volume of tweets per day was just a fraction of what it is now, Tumasjan and his colleagues published a paper comparing the content of the micro-messages to the 2009 German federal election results. They found that the number of mentions of the six different parties in the election did correlate strongly with the final election result.
The paper is heavily cited in the work that was sparked in this field. A lively scholarly debate has since ensued with the approach being applied to other countries and elections. In some cases, the results are replicated but, in others, a wider margin of error is observed.
As the 2015 Canadian federal election was drawing to a close, Twitter’s Canadian blog published an analysis of six million election-related tweets sent over two-and-a-half months. It did indicate that the eventual winner, Liberal Party leader Justin Trudeau, led his rivals in mentions as well as in new followers added during the election period. Twitter users do not represent a perfect random sample of the voting population, however. Tumasjan affirms that this inherent error can skew analysis results.
“As an overall summary you can say that Twitter is not better than traditional polling methods. However, many studies did find meaningful relationships, and Twitter analysis may thus be considered a useful complement to traditional polling methods. However, it definitely cannot replace traditional polling.” Tumasjan describes his approach as a “fast and frugal” way to peer into the political landscape.
It is an approach that isn’t being ignored. According to Bloomberg, at least three hedge funds were tipped off by their algorithms that it would be prudent to protect against a “Leave” decision in the June 23 Brexit referendum, saving millions of dollars in the ensuing market tumble. The Bloomberg article states that “Leave” supporters dominated “Remain”-ers on Instagram, Twitter, and Facebook. It also references Tumasjan’s pioneering work in this type of analysis. Indeed, several research projects are ongoing in the realm of combining multiple data sources to generate a more accurate forecast.
As this type of analysis gains momentum, Tumasjan cautions against the phenomenon of “astroturfing.” That term describes a small group trying to increase its clout by registering multiple accounts to simulate a “grassroots” movement. It is one of the reasons, along with user selection bias, why he doesn’t see Twitter replacing referenda any time soon. He adds, “I think that, for such purposes like conducting referenda, Twitter does not have a sufficiently large coverage of a country’s citizens to justify such a use. Moreover, every social media platform has its own special thematic focus and, importantly, also pursues commercial goals. So I think that government issues will always be handled through government-run platforms due to legal issues. However, I think that decision makers may well use social media sentiment analyses for getting a rough picture on the political deliberation for at least the social media users. Most governments in the Western world now run their own accounts on social media and certainly follow the reactions there. Overall, I think it definitely makes sense to continue to systematically analyze the political (and consumer etc.) sentiment on social media.”
So what are the latest tweets predicting for the 2016 US presidential election? The Electoral College system and wide interest in the election from those outside the U.S. introduce even more wrinkles into the required analysis. Those interested in unraveling such factors can dive into the world of Big Data themselves and follow in Dr. Andranik Tumasjan’s footsteps.