Wordkrill: Extending Wordfish into the multidimensional political space

TL;DR

Wordkrill extends the Wordfish model to estimate political actor positions across multiple dimensions, improving the analysis of complex political conflicts using text data.

Contribution

It introduces a multidimensional extension of Wordfish, maintaining interpretability while enabling efficient estimation of multiple latent political dimensions.

Findings

Applied to party manifestos and speeches, demonstrating practical utility.

Showed advantages of multidimensional analysis over unidimensional models.

Identified current limitations in the approach.

Abstract

Spatial models are central to the study of political conflict, yet their empirical application often depends on text-based methods. A prominent example is the Wordfish model, which estimates actor positions from political texts. However, a key limitation of Wordfish is its unidimensionality, despite the well-established multidimensional nature of political competition. This contribution introduces Wordkrill, a multidimensional extension of Wordfish that retains the original model's interpretability while allowing for efficient estimation of political positions along multiple latent dimensions. After presenting the mathematical framework of Wordkrill, its utility through brief applications to party manifestos and parliamentary speeches is demonstrated. These examples illustrate both the practical advantages and current limitations of the approach.