A burgeoning literature has studied the supply side of populism, namely which and how political parties use populist rhetoric in their communication. The few studies that disaggregated the party category by analyzing individual politicians almost exclusively focused on populist communication by leading candidates or heads of government. This narrow focus is destined to miss variations within parties: as candidates have different backgrounds (e.g., age, gender), are embedded in varying political contexts (e.g., position in party, media scrutiny) and face manifold structural conditions (e.g., rural/urban constituency, local unemployment), the incentives to appeal to “the people” using populist rhetoric vary considerably.
This paper analyzes the use of populist communication strategies during and after the German federal election campaign 2017. Our definition of populist communication is derived from Reinemann et al. (2017) who identify “appeals to the people”, “anti-elitism” and “exclusionism” as the constitutive elements of populism. We focus on candidates from parties with a realistic chance of passing the 5% threshold, i.e. AfD, CDU, CSU, FDP, Grüne, Linke and SPD. For our analysis, we collected all their posts on the social media platforms with highest political relevance, Facebook and Twitter. We then matched social media posts with publicly available data on candidates such as position on a party list, federal state, incumbency and individual characteristics like gender or age.
Social media data holds several advantages over the predominantly used data sources like party manifestos, press releases or TV advertisements. First, social media are the only source with sufficient data allowing for a disaggregated analysis of candidate communication (e.g., manifestos provide just one data point per party per election campaign). Second, static data sources like manifestos are blind towards the considerable shifts in how populist ideas develop over time and are communicated via the media. In contrast, social media capture the everyday dynamics of political communication and provide a constant and comparable flow of messages by political actors. Third, social media have gained particular importance in populist communication allowing politicians to bypass media gatekeepers, spread unfiltered messages and interact with supporters.
In order to identify populist communication in the more than half a million social media posts, we apply and evaluate quantitative text analysis methods. For this, we hand coded 5,000 posts according to our definition of populist communication as a training dataset for machine learning. We test and evaluate several machine learning classifiers that label each post in our data as populist (belonging to at least one of the three dimensions) or not. By incorporating all the context in each populist statement, we improve on the predominantly used dictionary-based approaches in the literature.
The output of the text analysis model is aggregated to the candidate level and then used as dependent variable in regression models including a range of individual, political and structural variables as predictors of populist communication. Preliminary results show variations in the use of populist communication strategies beyond party boundaries. This lends support to our theoretical assumptions and challenges the usual practice to treat (populist) parties as unitary actors.