Public Opinion in the German Parliament: A Semi-Automated Approach

The literature on the role of public opinion in shaping policy outcomes is vast, but often fails to directly identify if and how policymakers use public opinion. To contribute to this gap in our knowledge, we investigate references to public opinion by policymakers in parliamentary debates. This also gives us insights into how policymakers and parties use public opinion strategically and rhetorically to achieve their own goals. We combine qualitative coding and semi-supervised, deep learning algorithms to identify and categorize references to public opinion in the Bundestag (German parliament) using transcriptions of all parliamentary debates from 1991 to 2017. Our first finding is that public opinion appears quite often in Bundestag debates. Our second finding is that machine-based prediction does a decent job in categorizing references to public opinion. Using the semi-supervised classification algorithm (BERT) we were able to distinguish public opinion references from noise (sentences that include public opinion terms but are not actually public opinion statements) with an F1 score of 0.76 and an accuracy score of 0.78. We are currently working on improving the accuracy of our algorithm. Our next findings are still in progress, but we will develop a descriptive picture of the many uses of public opinion and how these change over time and across both parties and policy domains. Finally, we hope to compare actual public opinion surveys and policy changes to public opinion usages.