Party People in the House: Measuring Cross-party Difference and Intra-party Dissidence with Data from Speeches Held in the German Bundestag, 1949-2013

Plenary speeches have received a renewed interest as a source of data on party positions recently. But though speeches can be retrieved as separate texts, they are embedded in a parliamentary debate. This debate is also typically but a segment of an extended legislative process that is itself often only one step on the road through the policy cycle tied to the related policy field. In our paper, we argue that this encapsulated nature of speeches merits an “itemizing” approach to measurement and interpretation. This approach allows dissecting the population of speeches as a whole into particular sub-populations within policy fields and also accounting for speeches themselves as elements of an enduring, long-term process. As an application of this, we propose a three-step procedure towards the inference of political positions of parliamentarians, both as a cross-party measure (difference) and as an intra-party measure (dissidence). We measure similarity of texts within and between parties using Explicit Semantic Analysis. This technique gives us a representation of all debates in the form of a vector and relates each text (i.e. speech) to a broad knowledge base so that individual speeches consist only of “concepts”. This allows us to both reduce each speech to its core contents, and render speeches comparable as regards their theme. Crucially, this technique also helps overcoming classic linguistic problems like ambiguity or irony. We employ data on all plenary speeches held in a total of 3861 plenary sessions of the first 17 terms of the post-war German Bundestag, 1949-2013, as derived from the Bundestag protocols. We address the issue of categorizing debates into policy fields with the help of 116.806 annexes to the Bundestag protocols. We employ German Wikipedia entries as a large-scale, open-source and public reference corpus for deriving concepts contained in speeches. The paper aims at adding both to empirical findings on party positions as derived from individual parliamentarians’ positions, and to methodological advances in the study of legislative politics using text as a source of data, utilizing advanced computational techniques for measuring similarity of statements made in political discourse.