Parties’ Group Appeals Across Space and Time: An Effort Towards an Automated, Large-Scale Analysis of Parties’ Election Manifestos

A growing political science scholarship substantively distinguishes parties’ group-based appeals from policy appeals, the latter referring to the policies parties claim to support / oppose and the former to the specific social groups mentioned in parties’ statements. While both are prevalent in parties’ election materials, existing large-scale datasets focus primarily on policy-based appeals, with little data available for empirical analyses of group-based appeals. Moreover, the extensive resources needed to generate these data via hand-coding and the complexity of automating such text analysis challenge new collection efforts. However, examining group appeals is important for understanding party behavior, representation, and party-voter linkages. In this paper, we propose to address this data gap by automating group-based appeals’ extraction from political texts using a newly validated dictionary as well as a newly trained BERT model. The automated approach aims at accomplishing two tasks: the first is to detect the mentions of social groups, and the second is to evaluate the stance taken in the statement (supportive or opposing). As a test case, we analyze manifestos of nearly all British and Irish parties that gained seats in national-level elections since the 1970s, showing our approach’s benefits for revealing important trends in party behavior, with implications for party competition, voter behavior, and representation.