A New Bayesian Scaling Approach to Party Position and Issue Salience

• In Event: Political Scaling

Abstract

Research on electoral politics heavily relied on the Manifesto Project dataset (MPD) for party positions and issue salience. However, existing scaling approaches to generate these widely-used indicators do not correspond well to well-established theories of electoral politics and hence do not fully respect the true data-generating process (DGP). They suffer from inappropriate distributional assumptions, negligence of the data's hierarchical structure and insufficient differentiation between position and salience. This paper proposes a new scaling method based on Bayesian semi-parametric modelling for party position and issue salience. We demonstrate this new alternative's theoretical and empirical superiority compared with existing relative-frequency and log-odds-ratio approaches. This theory-derived Bayesian approach avoids the flaws in existing approaches and better models the proportion-based response distribution of MPD's true DGP and the hierarchical structure of the observed data. Moreover, it simultaneously estimates position and salience and can recover unique parameters for both. We empirically compare the new measurement with existing ones using the MPD and validate it with expert surveys. Lastly, we replicate existing empirical studies using the new scale and demonstrate its utility in a broader context.

Authors

• Chendi Wang, Vrije Universiteit Amsterdam

• Haoyu Zhai, NYU Abu Dhabi