The transformation of political competition in Poland 1991-2015 – a quantitative analysis of legislative speeches and roll-call data

In new democracies of Central and Eastern Europe party systems and the main axes of political competition are still relatively unstable in comparison with the Western European countries. The high fluidity is reflected in major changes in party systems, including vanishing of the old parties and emerging of the new ones, as well as in changes of ideological positions made by key political players, sometimes within a very short period of time. It makes it difficult to precisely determine the positions of political parties and – consequently – to analyse the process of transformation of the party systems. To overcome this problem, we conducted a quantitative analysis of legislative speeches in the lower house of the Polish parliament to track the changes in framing of key political issues in time. Parliamentary debates are a very multifaceted source of information, in comparison with other methods of determining party positioning, like roll-call data analysis (especially when party discipline is enforced), party programs (in which the organisation is trying to present one, coherent approach, strategically imposed by a party elite, rather than a variety of ideological options existing within a party), or expert evaluations (in which experts are asked to provide an ideal point of a party position instead of a range). Therefore, they allow for the broader comparison of inter-party and intra-party conflicts and dimensions of competition and their change in time. We conduct an automated analyses of a selected corpus of transcripts of debates in the lower house of the Polish parliament over the period 1991–2015 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. The method we use is a Structural Topic Model (STM) which is an extension of Latent Dirichlet Allocation (LDA) for computer-assisted reading of large text corpora allowing for incorporation of metadata into the topic model. Finally, the results of legislative speeches analysis are compared to roll-call data for the cross-validation purpose.