Tracking Agency in Structuring Political Competition using Automated Multidimensional Topic Modelling

The legacy of the communist rule in Poland, including flattening of the social structure and destruction of organisational networks, resulted in the majority of salient political divisions being poorly explainable by the socio-demographic traits of the voters. Consequently, the political competition can be seen as shaped mainly from above, stemming from the conflicts between political elites, rather than being an aggregation of the differences in society, especially in comparison with the Western European countries. Thus, it becomes essential to analyse the dimensions of competition among the party elites and their change over time, as it could provide the description of how parties drive the process of ideological divides formation, both in terms of their salience and their shape. To achieve this, we conduct an automated analysis of transcripts of debates in the lower house of the Polish parliament over the period 1991–2013 using a generative topic-modelling method, which infers latent topics (which the parliamentary speeches are considered to be mixtures of) based on word co-occurrence in utterances. Specifically, the method we use is Latent Dirichlet Allocation (LDA), extended to account for spatial polarity in the issues covered, in addition to the salience of the issues themselves. This allows us to track the formation and change of ideological divisions in relation to the processes of continuous merging, splitting and ideological shifts of the parties. We provide examples of how the crucial events and processes, including macro-economic reforms in the 1990s and European Union accession influenced the issue salience, and how the electoral cycle affected the topic convergence among parties.