Study: Twitter is poor at predicting election results

Donald Trump, Twitter
Twitter

Twitter is “highly unlikely” to become an accurate source for measuring political support and predicting election results, according to a study released this week. 

The article published in the Social Science Computer Review found that Twitter data is noisy and noted that the user base does not represent the voting public. And more than measuring candidate or party support, Twitter data is more likely just measuring public attention. 

ADVERTISEMENT
“As of now there is no convincing argument why the volume of Twitter mentions should be connected to election results,” said Andreas Jungherr of the University Mannheim in Germany, one of the article’s authors. 

“The only evidence proponents point to are a few selected case studies in which some carefully selected Twitter-metrics could be shown to be related to some selected metric of electoral success.”

While the case study in Jungherr's article dealt with 2013 German elections, he said by email that he believes the “issue is a general one.” The conclusion points out that the facts of Germany’s case study do not completely translate, especially because of the country’s low Twitter adoption rate. 

The article found that Twitter metrics, more than anything, measured fluctuations in election attention. It found that Twitter mentions increase during big news events, debates and as the election nears. Mentions also spike during controversy. 

While an increase in support can sometimes coincide with an increase in attention, the two are not necessarily linked. 

“It is reasonable to assume that in some cases public attention toward politics might be correlated with political support,” according to the article. “Still, we anticipate this relationship to be far from stable, given the fickle nature of public attention and the stability of political support.”

The article found that there is no settled Twitter metric used in past studies. They have varied between metrics like the total number of mentions, total number of positive mentions, total number of individuals who are mentioning a candidate and so on. 

Over the years, various studies have tried to uncover the predictive power of Twitter and have caught a lot of buzz. But Tuesday’s article said researchers would do well to acknowledge the limitations of Twitter data.

“This might lead Twitter-based research to free itself from inflated early expectations to find proxies of public opinion in Twitter-data and instead focus on the potential of digital trace data in yielding insights into public attention toward political information,” according to the article. 

Twitter took issue with the relevance of the study. 

“I’m surprised that The Hill finds 3 year old German Twitter data relevant. I'd advise passing next time,” Twitter spokesman Nick Pacilio said.

Twitter and other social media companies have been pushing out an increasing amount of data during the 2016 presidential elections. 

They point out the data is not meant to be predictive, but the companies are also quick to boast when their data lines up with election results. Twitter’s press team did that after the Iowa caucuses, while Google did the same after the New Hampshire primary. 

—Updated at 2:50 p.m.