From ‘likes’ to votes – UC researcher finds simple, accurate predictor of voter intention

A University of Canterbury researcher has found a simple and accurate way of predicting voter intention – and it may come down to a single Facebook ‘like’.

Jakob Bæk Kristensen, a PhD student in the Media and Communication department of UC Arts, has co-written a paper that has been published in the academic journal PLOS ONE, Parsimonious data: How a single Facebook like predicts voting behaviour in multiparty systems.

His original research reveals new information about what it is possible to tell about voters’ intentions using only publicly available data from Facebook.

The effects of social media on politics is still a fairly unexplored area, Jakob says.

“We leave digital traces behind whenever we click on something on the web. This research shows that a person’s political ‘likes’ on Facebook can predict the party they vote for with fairly high accuracy,” he says.

“Marketing firms, such as the infamous Cambridge Analytica, which claim to have secured the victory of Donald Trump and the Brexit campaign, are attributing their abilities to the predictive powers of big and broad data.”

The recipient of the UC Doctoral Scholarship, Jakob says his research shows that only very little, publicly available data is needed in order to predict which way someone will vote.

“Our research shows that by liking politicians’ public Facebook posts, a person's voting intention can be predicted with a fairly high accuracy in a multiparty system,” he says.

“A few selective digital traces produce prediction accuracies that are on par or even greater than most current approaches based upon bigger and broader datasets. Combining the online and offline, we connect a subsample of surveyed respondents to their public Facebook activity and apply machine learning classifiers to explore the link between their political liking behaviour and actual voting intention.

“Through this work, we show that even a single selective Facebook ‘like’ can reveal as much about political voter intention as hundreds of heterogeneous ‘likes’.

Within the field of election forecasting, research has shown the potential for predicting election outcomes based on digital data from a diverse range of platforms including YouTube, Google, Twitter, Facebook, and even Wikipedia.

“The techniques we developed are not language dependent and can be employed in any country where Facebook is popular with the general population. Future work will involve drawing on the results from our paper and using them to make election result predictions,” he says.

Jakob, who is from Denmark and conducted his original research with a social sciences colleague at the University of Copenhagen, is currently working on an interactive interface where these results can be viewed.

ends

© Scoop Media