Elff, Martin. 2017. "How Should We Extract Parties' Political Positions from their Manifestos? Problems of Conventional Approaches and their Solution Based on a Dynamic Idealpoint Model". Last presented at European Political Science Association 7th Annual Conference, 22-24 June 2017, Palazzo delle Stelline, Milano.
The compilation of coded manifestos by the Manifesto Project is an invaluable resource for the reconstruction of parties political positions and their changes over time. Nevertheless, the extraction of political positions from these data is confronted by some considerable challenges. First, the relation between positions and counts or percentages – as present in these data – is essentially non-linear. Second, Manifesto data may reflect both political positions and the salience of policy areas. Third, the policy space wherein parties take positions is not necessarily uni-dimensional. Fourth, positions taken by parties can be expected to be autocorrelated, since parties do not invent their positions in each election ex nihilo. Fifth, it is unlikely that even the Manifesto data is free of measurement error and and that the policy topic categories used may vary in terms of their discriminatory power for policy positions. The paper points out how these problems are addressed within the framework a dynamic idealpoint model of political texts.