In June 2018, more than 6,000 Facebook users took part in our Facebook study. Thank you to all participants! In this study we investigated how Facebook likes can be linked to political preference.
Summary of the main findings:
- Based on Facebook likes, we are quite capable of predicting someone's political preference.
- In 60% of the cases we can correctly predict whether someone is on the left, in the middle or on the right of the political spectrum. We can identify voters with a clear left or right profile with higher reliability (66%).
- In 64% of the cases we can correctly predict whether someone has much sympathy for a political party. With much sympathy we mean a score of 7 or more out of 10 on the question of how likely they are to ever vote for that party.
- In 36% of the cases we can predict exactly for which party someone will vote (this seems little but is substantially better than random gambling leading to an accuracy of 1/7 = 14%). The party preference for newer or more extreme parties can also be predicted much better (40-50%) than that of the traditional center parties.
- Non-political interests such as music, films, books, festivals, etc. that you like on Facebook also say something about your political preference. Even without charging political Facebook pages, political preference can be accurately predicted.
- People with different political preferences usually also have different interests. The Facebook pages that are most correlated with a certain political preference can be used to describe the wider interests of voters.
- Via a new index, the Facebook pages can be ranked according to political homogeneity. The Facebook pages of cafes, culture, books and events more often have a homogeneous left or right audience. On the other hand, mainly sports and TV shows unite people with different political preferences. In Flanders, most popular pages have a mixed audience, and organizations or individuals with a pronounced right or left profile are often smaller niche players.
Praet, S., Van Aelst, P., & Martens, D. (2018). I like, therefore I am: predictive modeling to gain insights in political preference in a multi-party system (FBE Research Paper: RPS-2018-014). Full text (open access):