Policy Positions in the Media. Measuring the Policy Positions of Parties in Newspapers Via Sentiment and Item Response Models

Policy positions of political parties are key to the analysis of various core questions of political science from party competition to the analysis of representation. Previous research has dedicated great sophistication to the measurement of party positions, which is typically based on either expert surveys or manifestos. We extend existing approaches by a procedure to comparatively measure party positions on specific policies as conveyed by newspapers. We make use of an extensive newspaper dataset that covers all articles from El Mundo (2002-2019), El País (1976-2019), The Times (1900-2013) and The Guardian (1984-2019) and investigate the case of territorial politics. We select relevant statements via machine learning, optimized dictionaries and structural topic models. Based on the selected sentences we compare the validity of generalized item response models (including wordfish as one of its applications) and sentiment analysis on the basis of a hand coded random sample. We discuss benefits and drawbacks of the different approaches and compare it to party positions on territorial issues within manifestos and parliamentary speeches, thus discussing the party competition of territorial politics in three arenas of the polity: parliaments, election campaigns and media. Finally, we evaluate the potential bias of centre-right and centre-left outlets in their transmission of direct partisan statements.