Predictive modeling to study lifestyle politics with Facebook likes

• Main Authors: Stiene Praet, Peter Van Aelst, Patrick van Erkel, Stephan Van der Veeken, David Martens
• Format: Article
• Language: English
• Published: SpringerOpen 2021-10-01
• Series: EPJ Data Science
• Subjects: Data science; Predictive modeling; Political preference; Facebook likes
• Online Access: https://doi.org/10.1140/epjds/s13688-021-00305-7

Abstract “Lifestyle politics” suggests that political and ideological opinions are strongly connected to our consumption choices, music and food taste, cultural preferences, and other aspects of our daily lives. With the growing political polarization this idea has become all the more relevant to a wide range of social scientists. Empirical research in this domain, however, is confronted with an impractical challenge; this type of detailed information on people’s lifestyle is very difficult to operationalize, and extremely time consuming and costly to query in a survey. A potential valuable alternative data source to capture these values and lifestyle choices is social media data. In this study, we explore the value of Facebook “like” data to complement traditional survey data to study lifestyle politics. We collect a unique dataset of Facebook likes and survey data of more than 6500 participants in Belgium, a fragmented multi-party system. Based on both types of data, we infer the political and ideological preference of our respondents. The results indicate that non-political Facebook likes are indicative of political preference and are useful to describe voters in terms of common interests, cultural preferences, and lifestyle features. This shows that social media data can be a valuable complement to traditional survey data to study lifestyle politics.